StillHammer
dbf1a3de8c
Added plan.md with complete architecture for format-agnostic content generation: - Support for Markdown, HTML, Plain Text, JSON formats - New FormatExporter module with neutral data structure - Integration strategy with existing ContentAssembly and ArticleStorage - Bonus features: SEO metadata generation, readability scoring, WordPress Gutenberg format - Implementation roadmap with 4 phases (6h total estimated) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
19018 lines
655 KiB
JavaScript
19018 lines
655 KiB
JavaScript
/*
|
|
code.js — bundle concaténé
|
|
Généré: 2025-09-04T15:08:51.662Z
|
|
Source: lib
|
|
Fichiers: 44
|
|
Ordre: topo
|
|
*/
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/ErrorReporting.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: lib/error-reporting.js - CONVERTI POUR NODE.JS
|
|
// Description: Système de validation et rapport d'erreur
|
|
// ========================================
|
|
|
|
const { google } = require('googleapis');
|
|
const nodemailer = require('nodemailer');
|
|
const fs = require('fs').promises;
|
|
const path = require('path');
|
|
const pino = require('pino');
|
|
const pretty = require('pino-pretty');
|
|
const { PassThrough } = require('stream');
|
|
const WebSocket = require('ws');
|
|
|
|
// Configuration
|
|
const SHEET_ID = process.env.GOOGLE_SHEETS_ID || '1iA2GvWeUxX-vpnAMfVm3ZMG9LhaC070SdGssEcXAh2c';
|
|
|
|
// WebSocket server for real-time logs
|
|
let wsServer;
|
|
const wsClients = new Set();
|
|
|
|
// Enhanced Pino logger configuration with real-time streaming and dated files
|
|
const now = new Date();
|
|
const timestamp = now.toISOString().slice(0, 10) + '_' +
|
|
now.toLocaleTimeString('fr-FR').replace(/:/g, '-');
|
|
const logFile = path.join(__dirname, '..', 'logs', `seo-generator-${timestamp}.log`);
|
|
|
|
const prettyStream = pretty({
|
|
colorize: true,
|
|
translateTime: 'HH:MM:ss.l',
|
|
ignore: 'pid,hostname',
|
|
});
|
|
|
|
const tee = new PassThrough();
|
|
tee.pipe(prettyStream).pipe(process.stdout);
|
|
|
|
// File destination with dated filename - FORCE DEBUG LEVEL
|
|
const fileDest = pino.destination({
|
|
dest: logFile,
|
|
mkdir: true,
|
|
sync: false,
|
|
minLength: 0 // Force immediate write even for small logs
|
|
});
|
|
tee.pipe(fileDest);
|
|
|
|
// Custom levels for Pino to include TRACE, PROMPT, and LLM
|
|
const customLevels = {
|
|
trace: 5, // Below debug (10)
|
|
debug: 10,
|
|
info: 20,
|
|
prompt: 25, // New level for prompts (between info and warn)
|
|
llm: 26, // New level for LLM interactions (between prompt and warn)
|
|
warn: 30,
|
|
error: 40,
|
|
fatal: 50
|
|
};
|
|
|
|
// Pino logger instance with enhanced configuration and custom levels
|
|
const logger = pino(
|
|
{
|
|
level: 'debug', // FORCE DEBUG LEVEL for file logging
|
|
base: undefined,
|
|
timestamp: pino.stdTimeFunctions.isoTime,
|
|
customLevels: customLevels,
|
|
useOnlyCustomLevels: true
|
|
},
|
|
tee
|
|
);
|
|
|
|
// Initialize WebSocket server
|
|
function initWebSocketServer() {
|
|
if (!wsServer) {
|
|
wsServer = new WebSocket.Server({ port: process.env.LOG_WS_PORT || 8081 });
|
|
|
|
wsServer.on('connection', (ws) => {
|
|
wsClients.add(ws);
|
|
logger.info('Client connected to log WebSocket');
|
|
|
|
ws.on('close', () => {
|
|
wsClients.delete(ws);
|
|
logger.info('Client disconnected from log WebSocket');
|
|
});
|
|
|
|
ws.on('error', (error) => {
|
|
logger.error('WebSocket error:', error.message);
|
|
wsClients.delete(ws);
|
|
});
|
|
});
|
|
|
|
logger.info(`Log WebSocket server started on port ${process.env.LOG_WS_PORT || 8081}`);
|
|
}
|
|
}
|
|
|
|
// Broadcast log to WebSocket clients
|
|
function broadcastLog(message, level) {
|
|
const logData = {
|
|
timestamp: new Date().toISOString(),
|
|
level: level.toUpperCase(),
|
|
message: message
|
|
};
|
|
|
|
wsClients.forEach(ws => {
|
|
if (ws.readyState === WebSocket.OPEN) {
|
|
try {
|
|
ws.send(JSON.stringify(logData));
|
|
} catch (error) {
|
|
logger.error('Failed to send log to WebSocket client:', error.message);
|
|
wsClients.delete(ws);
|
|
}
|
|
}
|
|
});
|
|
}
|
|
|
|
// 🔄 NODE.JS : Google Sheets API setup (remplace SpreadsheetApp)
|
|
let sheets;
|
|
let auth;
|
|
|
|
async function initGoogleSheets() {
|
|
if (!sheets) {
|
|
// Configuration auth Google Sheets API
|
|
// Pour la démo, on utilise une clé de service (à configurer)
|
|
auth = new google.auth.GoogleAuth({
|
|
keyFile: process.env.GOOGLE_CREDENTIALS_PATH, // Chemin vers fichier JSON credentials
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets']
|
|
});
|
|
|
|
sheets = google.sheets({ version: 'v4', auth });
|
|
}
|
|
return sheets;
|
|
}
|
|
|
|
async function logSh(message, level = 'INFO') {
|
|
// Initialize WebSocket server if not already done
|
|
if (!wsServer) {
|
|
initWebSocketServer();
|
|
}
|
|
|
|
// Convert level to lowercase for Pino
|
|
const pinoLevel = level.toLowerCase();
|
|
|
|
// Enhanced trace metadata for hierarchical logging
|
|
const traceData = {};
|
|
if (message.includes('▶') || message.includes('✔') || message.includes('✖') || message.includes('•')) {
|
|
traceData.trace = true;
|
|
traceData.evt = message.includes('▶') ? 'span.start' :
|
|
message.includes('✔') ? 'span.end' :
|
|
message.includes('✖') ? 'span.error' : 'span.event';
|
|
}
|
|
|
|
// Log with Pino (handles console output with pretty formatting and file logging)
|
|
switch (pinoLevel) {
|
|
case 'error':
|
|
logger.error(traceData, message);
|
|
break;
|
|
case 'warning':
|
|
case 'warn':
|
|
logger.warn(traceData, message);
|
|
break;
|
|
case 'debug':
|
|
logger.debug(traceData, message);
|
|
break;
|
|
case 'trace':
|
|
logger.trace(traceData, message);
|
|
break;
|
|
case 'prompt':
|
|
logger.prompt(traceData, message);
|
|
break;
|
|
case 'llm':
|
|
logger.llm(traceData, message);
|
|
break;
|
|
default:
|
|
logger.info(traceData, message);
|
|
}
|
|
|
|
// Broadcast to WebSocket clients for real-time viewing
|
|
broadcastLog(message, level);
|
|
|
|
// Force immediate flush to ensure real-time display and prevent log loss
|
|
logger.flush();
|
|
|
|
// Log to Google Sheets if enabled (async, non-blocking)
|
|
if (process.env.ENABLE_SHEETS_LOGGING === 'true') {
|
|
setImmediate(() => {
|
|
logToGoogleSheets(message, level).catch(err => {
|
|
// Silent fail for Google Sheets logging to avoid recursion
|
|
});
|
|
});
|
|
}
|
|
}
|
|
|
|
// Fonction pour déterminer si on doit logger en console
|
|
function shouldLogToConsole(messageLevel, configLevel) {
|
|
const levels = { DEBUG: 0, INFO: 1, WARNING: 2, ERROR: 3 };
|
|
return levels[messageLevel] >= levels[configLevel];
|
|
}
|
|
|
|
// Log to file is now handled by Pino transport
|
|
// This function is kept for compatibility but does nothing
|
|
async function logToFile(message, level) {
|
|
// Pino handles file logging via transport configuration
|
|
// This function is deprecated and kept for compatibility only
|
|
}
|
|
|
|
// 🔄 NODE.JS : Log vers Google Sheets (version async)
|
|
async function logToGoogleSheets(message, level) {
|
|
try {
|
|
const sheetsApi = await initGoogleSheets();
|
|
|
|
const values = [[
|
|
new Date().toISOString(),
|
|
level,
|
|
message,
|
|
'Node.js workflow'
|
|
]];
|
|
|
|
await sheetsApi.spreadsheets.values.append({
|
|
spreadsheetId: SHEET_ID,
|
|
range: 'Logs!A:D',
|
|
valueInputOption: 'RAW',
|
|
insertDataOption: 'INSERT_ROWS',
|
|
resource: { values }
|
|
});
|
|
|
|
} catch (error) {
|
|
logSh('Échec log Google Sheets: ' + error.message, 'WARNING'); // Using logSh instead of console.warn
|
|
}
|
|
}
|
|
|
|
// 🔄 NODE.JS : Version simplifiée cleanLogSheet
|
|
async function cleanLogSheet() {
|
|
try {
|
|
logSh('🧹 Nettoyage logs...', 'INFO'); // Using logSh instead of console.log
|
|
|
|
// 1. Nettoyer fichiers logs locaux (garder 7 derniers jours)
|
|
await cleanLocalLogs();
|
|
|
|
// 2. Nettoyer Google Sheets si activé
|
|
if (process.env.ENABLE_SHEETS_LOGGING === 'true') {
|
|
await cleanGoogleSheetsLogs();
|
|
}
|
|
|
|
logSh('✅ Logs nettoyés', 'INFO'); // Using logSh instead of console.log
|
|
|
|
} catch (error) {
|
|
logSh('Erreur nettoyage logs: ' + error.message, 'ERROR'); // Using logSh instead of console.error
|
|
}
|
|
}
|
|
|
|
async function cleanLocalLogs() {
|
|
try {
|
|
// Note: With Pino, log files are managed differently
|
|
// This function is kept for compatibility with Google Sheets logs cleanup
|
|
// Pino log rotation should be handled by external tools like logrotate
|
|
|
|
// For now, we keep the basic cleanup for any remaining old log files
|
|
const logsDir = path.join(__dirname, '../logs');
|
|
|
|
try {
|
|
const files = await fs.readdir(logsDir);
|
|
const cutoffDate = new Date();
|
|
cutoffDate.setDate(cutoffDate.getDate() - 7); // Garder 7 jours
|
|
|
|
for (const file of files) {
|
|
if (file.endsWith('.log')) {
|
|
const filePath = path.join(logsDir, file);
|
|
const stats = await fs.stat(filePath);
|
|
|
|
if (stats.mtime < cutoffDate) {
|
|
await fs.unlink(filePath);
|
|
logSh(`🗑️ Supprimé log ancien: ${file}`, 'INFO');
|
|
}
|
|
}
|
|
}
|
|
} catch (error) {
|
|
// Directory might not exist, that's fine
|
|
}
|
|
} catch (error) {
|
|
// Silent fail
|
|
}
|
|
}
|
|
|
|
async function cleanGoogleSheetsLogs() {
|
|
try {
|
|
const sheetsApi = await initGoogleSheets();
|
|
|
|
// Clear + remettre headers
|
|
await sheetsApi.spreadsheets.values.clear({
|
|
spreadsheetId: SHEET_ID,
|
|
range: 'Logs!A:D'
|
|
});
|
|
|
|
await sheetsApi.spreadsheets.values.update({
|
|
spreadsheetId: SHEET_ID,
|
|
range: 'Logs!A1:D1',
|
|
valueInputOption: 'RAW',
|
|
resource: {
|
|
values: [['Timestamp', 'Level', 'Message', 'Source']]
|
|
}
|
|
});
|
|
|
|
} catch (error) {
|
|
logSh('Échec nettoyage Google Sheets: ' + error.message, 'WARNING'); // Using logSh instead of console.warn
|
|
}
|
|
}
|
|
|
|
// ============= VALIDATION PRINCIPALE - IDENTIQUE =============
|
|
|
|
function validateWorkflowIntegrity(elements, generatedContent, finalXML, csvData) {
|
|
logSh('🔍 >>> VALIDATION INTÉGRITÉ WORKFLOW <<<', 'INFO'); // Using logSh instead of console.log
|
|
|
|
const errors = [];
|
|
const warnings = [];
|
|
const stats = {
|
|
elementsExtracted: elements.length,
|
|
contentGenerated: Object.keys(generatedContent).length,
|
|
tagsReplaced: 0,
|
|
tagsRemaining: 0
|
|
};
|
|
|
|
// TEST 1: Détection tags dupliqués
|
|
const duplicateCheck = detectDuplicateTags(elements);
|
|
if (duplicateCheck.hasDuplicates) {
|
|
errors.push({
|
|
type: 'DUPLICATE_TAGS',
|
|
severity: 'HIGH',
|
|
message: `Tags dupliqués détectés: ${duplicateCheck.duplicates.join(', ')}`,
|
|
impact: 'Certains contenus ne seront pas remplacés dans le XML final',
|
|
suggestion: 'Vérifier le template XML pour corriger la structure'
|
|
});
|
|
}
|
|
|
|
// TEST 2: Cohérence éléments extraits vs générés
|
|
const missingGeneration = elements.filter(el => !generatedContent[el.originalTag]);
|
|
if (missingGeneration.length > 0) {
|
|
errors.push({
|
|
type: 'MISSING_GENERATION',
|
|
severity: 'HIGH',
|
|
message: `${missingGeneration.length} éléments extraits mais non générés`,
|
|
details: missingGeneration.map(el => el.originalTag),
|
|
impact: 'Contenu incomplet dans le XML final'
|
|
});
|
|
}
|
|
|
|
// TEST 3: Tags non remplacés dans XML final
|
|
const remainingTags = (finalXML.match(/\|[^|]*\|/g) || []);
|
|
stats.tagsRemaining = remainingTags.length;
|
|
|
|
if (remainingTags.length > 0) {
|
|
errors.push({
|
|
type: 'UNREPLACED_TAGS',
|
|
severity: 'HIGH',
|
|
message: `${remainingTags.length} tags non remplacés dans le XML final`,
|
|
details: remainingTags.slice(0, 5),
|
|
impact: 'XML final contient des placeholders non remplacés'
|
|
});
|
|
}
|
|
|
|
// TEST 4: Variables CSV manquantes
|
|
const missingVars = detectMissingCSVVariables(csvData);
|
|
if (missingVars.length > 0) {
|
|
warnings.push({
|
|
type: 'MISSING_CSV_VARIABLES',
|
|
severity: 'MEDIUM',
|
|
message: `Variables CSV manquantes: ${missingVars.join(', ')}`,
|
|
impact: 'Système de génération de mots-clés automatique activé'
|
|
});
|
|
}
|
|
|
|
// TEST 5: Qualité génération IA
|
|
const generationQuality = assessGenerationQuality(generatedContent);
|
|
if (generationQuality.errorRate > 0.1) {
|
|
warnings.push({
|
|
type: 'GENERATION_QUALITY',
|
|
severity: 'MEDIUM',
|
|
message: `${(generationQuality.errorRate * 100).toFixed(1)}% d'erreurs de génération IA`,
|
|
impact: 'Qualité du contenu potentiellement dégradée'
|
|
});
|
|
}
|
|
|
|
// CALCUL STATS FINALES
|
|
stats.tagsReplaced = elements.length - remainingTags.length;
|
|
stats.successRate = stats.elementsExtracted > 0 ?
|
|
((stats.tagsReplaced / elements.length) * 100).toFixed(1) : '100';
|
|
|
|
const report = {
|
|
timestamp: new Date().toISOString(),
|
|
csvData: { mc0: csvData.mc0, t0: csvData.t0 },
|
|
stats: stats,
|
|
errors: errors,
|
|
warnings: warnings,
|
|
status: errors.length === 0 ? 'SUCCESS' : 'ERROR'
|
|
};
|
|
|
|
const logLevel = report.status === 'SUCCESS' ? 'INFO' : 'ERROR';
|
|
logSh(`✅ Validation terminée: ${report.status} (${errors.length} erreurs, ${warnings.length} warnings)`, 'INFO'); // Using logSh instead of console.log
|
|
|
|
// ENVOYER RAPPORT SI ERREURS (async en arrière-plan)
|
|
if (errors.length > 0 || warnings.length > 2) {
|
|
sendErrorReport(report).catch(err => {
|
|
logSh('Erreur envoi rapport: ' + err.message, 'ERROR'); // Using logSh instead of console.error
|
|
});
|
|
}
|
|
|
|
return report;
|
|
}
|
|
|
|
// ============= HELPERS - IDENTIQUES =============
|
|
|
|
function detectDuplicateTags(elements) {
|
|
const tagCounts = {};
|
|
const duplicates = [];
|
|
|
|
elements.forEach(element => {
|
|
const tag = element.originalTag;
|
|
tagCounts[tag] = (tagCounts[tag] || 0) + 1;
|
|
|
|
if (tagCounts[tag] === 2) {
|
|
duplicates.push(tag);
|
|
logSh(`❌ DUPLICATE détecté: ${tag}`, 'ERROR'); // Using logSh instead of console.error
|
|
}
|
|
});
|
|
|
|
return {
|
|
hasDuplicates: duplicates.length > 0,
|
|
duplicates: duplicates,
|
|
counts: tagCounts
|
|
};
|
|
}
|
|
|
|
function detectMissingCSVVariables(csvData) {
|
|
const missing = [];
|
|
|
|
if (!csvData.mcPlus1 || csvData.mcPlus1.split(',').length < 4) {
|
|
missing.push('MC+1 (insuffisant)');
|
|
}
|
|
if (!csvData.tPlus1 || csvData.tPlus1.split(',').length < 4) {
|
|
missing.push('T+1 (insuffisant)');
|
|
}
|
|
if (!csvData.lPlus1 || csvData.lPlus1.split(',').length < 4) {
|
|
missing.push('L+1 (insuffisant)');
|
|
}
|
|
|
|
return missing;
|
|
}
|
|
|
|
function assessGenerationQuality(generatedContent) {
|
|
let errorCount = 0;
|
|
let totalCount = Object.keys(generatedContent).length;
|
|
|
|
Object.values(generatedContent).forEach(content => {
|
|
if (content && (
|
|
content.includes('[ERREUR') ||
|
|
content.includes('ERROR') ||
|
|
content.length < 10
|
|
)) {
|
|
errorCount++;
|
|
}
|
|
});
|
|
|
|
return {
|
|
errorRate: totalCount > 0 ? errorCount / totalCount : 0,
|
|
totalGenerated: totalCount,
|
|
errorsFound: errorCount
|
|
};
|
|
}
|
|
|
|
// 🔄 NODE.JS : Email avec nodemailer (remplace MailApp)
|
|
async function sendErrorReport(report) {
|
|
try {
|
|
logSh('📧 Envoi rapport d\'erreur par email...', 'INFO'); // Using logSh instead of console.log
|
|
|
|
// Configuration nodemailer (Gmail par exemple)
|
|
const transporter = nodemailer.createTransport({
|
|
service: 'gmail',
|
|
auth: {
|
|
user: process.env.EMAIL_USER, // 'your-email@gmail.com'
|
|
pass: process.env.EMAIL_APP_PASSWORD // App password Google
|
|
}
|
|
});
|
|
|
|
const subject = `Erreur Workflow SEO Node.js - ${report.status} - ${report.csvData.mc0}`;
|
|
const htmlBody = createHTMLReport(report);
|
|
|
|
const mailOptions = {
|
|
from: process.env.EMAIL_USER,
|
|
to: 'alexistrouve.pro@gmail.com',
|
|
subject: subject,
|
|
html: htmlBody,
|
|
attachments: [{
|
|
filename: `error-report-${Date.now()}.json`,
|
|
content: JSON.stringify(report, null, 2),
|
|
contentType: 'application/json'
|
|
}]
|
|
};
|
|
|
|
await transporter.sendMail(mailOptions);
|
|
logSh('✅ Rapport d\'erreur envoyé par email', 'INFO'); // Using logSh instead of console.log
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Échec envoi email: ${error.message}`, 'ERROR'); // Using logSh instead of console.error
|
|
}
|
|
}
|
|
|
|
// ============= HTML REPORT - IDENTIQUE =============
|
|
|
|
function createHTMLReport(report) {
|
|
const statusColor = report.status === 'SUCCESS' ? '#28a745' : '#dc3545';
|
|
|
|
let html = `
|
|
<div style="font-family: Arial, sans-serif; max-width: 800px; margin: 0 auto;">
|
|
<h1 style="color: ${statusColor};">Rapport Workflow SEO Automatisé (Node.js)</h1>
|
|
|
|
<div style="background: #f8f9fa; padding: 15px; border-radius: 5px; margin: 20px 0;">
|
|
<h2>Résumé Exécutif</h2>
|
|
<p><strong>Statut:</strong> <span style="color: ${statusColor};">${report.status}</span></p>
|
|
<p><strong>Article:</strong> ${report.csvData.t0}</p>
|
|
<p><strong>Mot-clé:</strong> ${report.csvData.mc0}</p>
|
|
<p><strong>Taux de réussite:</strong> ${report.stats.successRate}%</p>
|
|
<p><strong>Timestamp:</strong> ${report.timestamp}</p>
|
|
<p><strong>Plateforme:</strong> Node.js Server</p>
|
|
</div>`;
|
|
|
|
if (report.errors.length > 0) {
|
|
html += `<div style="background: #f8d7da; padding: 15px; border-radius: 5px; margin: 20px 0;">
|
|
<h2>Erreurs Critiques (${report.errors.length})</h2>`;
|
|
|
|
report.errors.forEach((error, i) => {
|
|
html += `
|
|
<div style="margin: 10px 0; padding: 10px; border-left: 3px solid #dc3545;">
|
|
<h4>${i + 1}. ${error.type}</h4>
|
|
<p><strong>Message:</strong> ${error.message}</p>
|
|
<p><strong>Impact:</strong> ${error.impact}</p>
|
|
${error.suggestion ? `<p><strong>Solution:</strong> ${error.suggestion}</p>` : ''}
|
|
</div>`;
|
|
});
|
|
|
|
html += `</div>`;
|
|
}
|
|
|
|
if (report.warnings.length > 0) {
|
|
html += `<div style="background: #fff3cd; padding: 15px; border-radius: 5px; margin: 20px 0;">
|
|
<h2>Avertissements (${report.warnings.length})</h2>`;
|
|
|
|
report.warnings.forEach((warning, i) => {
|
|
html += `
|
|
<div style="margin: 10px 0; padding: 10px; border-left: 3px solid #ffc107;">
|
|
<h4>${i + 1}. ${warning.type}</h4>
|
|
<p>${warning.message}</p>
|
|
</div>`;
|
|
});
|
|
|
|
html += `</div>`;
|
|
}
|
|
|
|
html += `
|
|
<div style="background: #e9ecef; padding: 15px; border-radius: 5px; margin: 20px 0;">
|
|
<h2>Statistiques Détaillées</h2>
|
|
<ul>
|
|
<li>Éléments extraits: ${report.stats.elementsExtracted}</li>
|
|
<li>Contenus générés: ${report.stats.contentGenerated}</li>
|
|
<li>Tags remplacés: ${report.stats.tagsReplaced}</li>
|
|
<li>Tags restants: ${report.stats.tagsRemaining}</li>
|
|
</ul>
|
|
</div>
|
|
|
|
<div style="background: #d1ecf1; padding: 15px; border-radius: 5px; margin: 20px 0;">
|
|
<h2>Informations Système</h2>
|
|
<ul>
|
|
<li>Plateforme: Node.js</li>
|
|
<li>Version: ${process.version}</li>
|
|
<li>Mémoire: ${Math.round(process.memoryUsage().heapUsed / 1024 / 1024)}MB</li>
|
|
<li>Uptime: ${Math.round(process.uptime())}s</li>
|
|
</ul>
|
|
</div>
|
|
</div>`;
|
|
|
|
return html;
|
|
}
|
|
|
|
// 🔄 NODE.JS EXPORTS
|
|
module.exports = {
|
|
logSh,
|
|
cleanLogSheet,
|
|
validateWorkflowIntegrity,
|
|
detectDuplicateTags,
|
|
detectMissingCSVVariables,
|
|
assessGenerationQuality,
|
|
sendErrorReport,
|
|
createHTMLReport
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/BrainConfig.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: BrainConfig.js - Version Node.js
|
|
// Description: Configuration cerveau + sélection personnalité IA
|
|
// ========================================
|
|
|
|
require('dotenv').config();
|
|
const axios = require('axios');
|
|
const fs = require('fs').promises;
|
|
const path = require('path');
|
|
|
|
// Import de la fonction logSh (assumant qu'elle existe dans votre projet Node.js)
|
|
const { logSh } = require('./ErrorReporting');
|
|
|
|
// Configuration
|
|
const CONFIG = {
|
|
openai: {
|
|
apiKey: process.env.OPENAI_API_KEY || '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',
|
|
endpoint: 'https://api.openai.com/v1/chat/completions'
|
|
},
|
|
dataSource: {
|
|
type: process.env.DATA_SOURCE_TYPE || 'json', // 'json', 'csv', 'database'
|
|
instructionsPath: './data/instructions.json',
|
|
personalitiesPath: './data/personalities.json'
|
|
}
|
|
};
|
|
|
|
/**
|
|
* FONCTION PRINCIPALE - Équivalent getBrainConfig()
|
|
* @param {number|object} data - Numéro de ligne ou données directes
|
|
* @returns {object} Configuration avec données CSV + personnalité
|
|
*/
|
|
async function getBrainConfig(data) {
|
|
try {
|
|
logSh("🧠 Début getBrainConfig Node.js", "INFO");
|
|
|
|
// 1. RÉCUPÉRER LES DONNÉES CSV
|
|
let csvData;
|
|
if (typeof data === 'number') {
|
|
// Numéro de ligne fourni - lire depuis fichier
|
|
csvData = await readInstructionsData(data);
|
|
} else if (typeof data === 'object' && data.rowNumber) {
|
|
csvData = await readInstructionsData(data.rowNumber);
|
|
} else {
|
|
// Données déjà fournies
|
|
csvData = data;
|
|
}
|
|
|
|
logSh(`✅ CSV récupéré: ${csvData.mc0}`, "INFO");
|
|
|
|
// 2. RÉCUPÉRER LES PERSONNALITÉS
|
|
const personalities = await getPersonalities();
|
|
logSh(`✅ ${personalities.length} personnalités chargées`, "INFO");
|
|
|
|
// 3. SÉLECTIONNER LA MEILLEURE PERSONNALITÉ VIA IA
|
|
const selectedPersonality = await selectPersonalityWithAI(
|
|
csvData.mc0,
|
|
csvData.t0,
|
|
personalities
|
|
);
|
|
|
|
logSh(`✅ Personnalité sélectionnée: ${selectedPersonality.nom}`, "INFO");
|
|
|
|
return {
|
|
success: true,
|
|
data: {
|
|
...csvData,
|
|
personality: selectedPersonality,
|
|
timestamp: new Date().toISOString()
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur getBrainConfig: ${error.message}`, "ERROR");
|
|
return {
|
|
success: false,
|
|
error: error.message
|
|
};
|
|
}
|
|
}
|
|
|
|
/**
|
|
* LIRE DONNÉES INSTRUCTIONS depuis Google Sheets DIRECTEMENT
|
|
* @param {number} rowNumber - Numéro de ligne (2 = première ligne de données)
|
|
* @returns {object} Données CSV parsées
|
|
*/
|
|
async function readInstructionsData(rowNumber = 2) {
|
|
try {
|
|
logSh(`📊 Lecture Google Sheet ligne ${rowNumber}...`, 'INFO');
|
|
|
|
// NOUVEAU : Lecture directe depuis Google Sheets
|
|
const { google } = require('googleapis');
|
|
|
|
// Configuration auth Google Sheets - FORCE utilisation fichier JSON pour éviter problème TLS
|
|
const keyFilePath = path.join(__dirname, '..', 'seo-generator-470715-85d4a971c1af.json');
|
|
const auth = new google.auth.GoogleAuth({
|
|
keyFile: keyFilePath,
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets.readonly']
|
|
});
|
|
logSh('🔑 Utilisation fichier JSON pour contourner problème TLS OAuth', 'INFO');
|
|
|
|
const sheets = google.sheets({ version: 'v4', auth });
|
|
const SHEET_ID = process.env.GOOGLE_SHEETS_ID || '1iA2GvWeUxX-vpnAMfVm3ZMG9LhaC070SdGssEcXAh2c';
|
|
|
|
// Récupérer la ligne spécifique (A à I au minimum)
|
|
const response = await sheets.spreadsheets.values.get({
|
|
spreadsheetId: SHEET_ID,
|
|
range: `Instructions!A${rowNumber}:I${rowNumber}` // Ligne spécifique A-I
|
|
});
|
|
|
|
if (!response.data.values || response.data.values.length === 0) {
|
|
throw new Error(`Ligne ${rowNumber} non trouvée dans Google Sheet`);
|
|
}
|
|
|
|
const row = response.data.values[0];
|
|
logSh(`✅ Ligne ${rowNumber} récupérée: ${row.length} colonnes`, 'INFO');
|
|
|
|
const xmlTemplateValue = row[8] || '';
|
|
let xmlTemplate = xmlTemplateValue;
|
|
let xmlFileName = null;
|
|
|
|
// Si c'est un nom de fichier, garder le nom ET utiliser un template par défaut
|
|
if (xmlTemplateValue && xmlTemplateValue.endsWith('.xml') && xmlTemplateValue.length < 100) {
|
|
logSh(`🔧 XML filename detected (${xmlTemplateValue}), keeping filename for Digital Ocean`, 'INFO');
|
|
xmlFileName = xmlTemplateValue; // Garder le nom du fichier pour Digital Ocean
|
|
xmlTemplate = createDefaultXMLTemplate(); // Template par défaut pour le processing
|
|
}
|
|
|
|
return {
|
|
rowNumber: rowNumber,
|
|
slug: row[0] || '', // Colonne A
|
|
t0: row[1] || '', // Colonne B
|
|
mc0: row[2] || '', // Colonne C
|
|
tMinus1: row[3] || '', // Colonne D
|
|
lMinus1: row[4] || '', // Colonne E
|
|
mcPlus1: row[5] || '', // Colonne F
|
|
tPlus1: row[6] || '', // Colonne G
|
|
lPlus1: row[7] || '', // Colonne H
|
|
xmlTemplate: xmlTemplate, // XML template pour processing
|
|
xmlFileName: xmlFileName // Nom fichier pour Digital Ocean (si applicable)
|
|
};
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur lecture Google Sheet: ${error.message}`, "ERROR");
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* RÉCUPÉRER PERSONNALITÉS depuis l'onglet "Personnalites" du Google Sheet
|
|
* @returns {Array} Liste des personnalités disponibles
|
|
*/
|
|
async function getPersonalities() {
|
|
try {
|
|
logSh('📊 Lecture personnalités depuis Google Sheet (onglet Personnalites)...', 'INFO');
|
|
|
|
// Configuration auth Google Sheets - FORCE utilisation fichier JSON pour éviter problème TLS
|
|
const { google } = require('googleapis');
|
|
const keyFilePath = path.join(__dirname, '..', 'seo-generator-470715-85d4a971c1af.json');
|
|
const auth = new google.auth.GoogleAuth({
|
|
keyFile: keyFilePath,
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets.readonly']
|
|
});
|
|
logSh('🔑 Utilisation fichier JSON pour contourner problème TLS OAuth (personnalités)', 'INFO');
|
|
|
|
const sheets = google.sheets({ version: 'v4', auth });
|
|
const SHEET_ID = process.env.GOOGLE_SHEETS_ID || '1iA2GvWeUxX-vpnAMfVm3ZMG9LhaC070SdGssEcXAh2c';
|
|
|
|
// Récupérer toutes les personnalités (après la ligne d'en-tête)
|
|
const response = await sheets.spreadsheets.values.get({
|
|
spreadsheetId: SHEET_ID,
|
|
range: 'Personnalites!A2:O' // Colonnes A à O pour inclure les nouvelles colonnes IA
|
|
});
|
|
|
|
if (!response.data.values || response.data.values.length === 0) {
|
|
throw new Error('Aucune personnalité trouvée dans l\'onglet Personnalites');
|
|
}
|
|
|
|
const personalities = [];
|
|
|
|
// Traiter chaque ligne de personnalité
|
|
response.data.values.forEach((row, index) => {
|
|
if (row[0] && row[0].toString().trim() !== '') { // Si nom existe (colonne A)
|
|
const personality = {
|
|
nom: row[0]?.toString().trim() || '',
|
|
description: row[1]?.toString().trim() || 'Expert généraliste',
|
|
style: row[2]?.toString().trim() || 'professionnel',
|
|
|
|
// Configuration avancée depuis colonnes Google Sheet
|
|
motsClesSecteurs: parseCSVField(row[3]),
|
|
vocabulairePref: parseCSVField(row[4]),
|
|
connecteursPref: parseCSVField(row[5]),
|
|
erreursTypiques: parseCSVField(row[6]),
|
|
longueurPhrases: row[7]?.toString().trim() || 'moyennes',
|
|
niveauTechnique: row[8]?.toString().trim() || 'moyen',
|
|
ctaStyle: parseCSVField(row[9]),
|
|
defautsSimules: parseCSVField(row[10]),
|
|
|
|
// NOUVEAU: Configuration IA par étape depuis Google Sheets (colonnes L-O)
|
|
aiEtape1Base: row[11]?.toString().trim().toLowerCase() || '',
|
|
aiEtape2Technique: row[12]?.toString().trim().toLowerCase() || '',
|
|
aiEtape3Transitions: row[13]?.toString().trim().toLowerCase() || '',
|
|
aiEtape4Style: row[14]?.toString().trim().toLowerCase() || '',
|
|
|
|
// Backward compatibility
|
|
motsCles: parseCSVField(row[3] || '') // Utilise motsClesSecteurs
|
|
};
|
|
|
|
personalities.push(personality);
|
|
logSh(`✓ Personnalité chargée: ${personality.nom} (${personality.style})`, 'DEBUG');
|
|
}
|
|
});
|
|
|
|
logSh(`📊 ${personalities.length} personnalités chargées depuis Google Sheet`, "INFO");
|
|
|
|
return personalities;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ ÉCHEC: Impossible de récupérer les personnalités Google Sheets - ${error.message}`, "ERROR");
|
|
throw new Error(`FATAL: Personnalités Google Sheets inaccessibles - arrêt du workflow: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* PARSER CHAMP CSV - Helper function
|
|
* @param {string} field - Champ à parser
|
|
* @returns {Array} Liste des éléments parsés
|
|
*/
|
|
function parseCSVField(field) {
|
|
if (!field || field.toString().trim() === '') return [];
|
|
|
|
return field.toString()
|
|
.split(',')
|
|
.map(item => item.trim())
|
|
.filter(item => item.length > 0);
|
|
}
|
|
|
|
/**
|
|
* Sélectionner un sous-ensemble aléatoire de personnalités
|
|
* @param {Array} allPersonalities - Liste complète des personnalités
|
|
* @param {number} percentage - Pourcentage à garder (0.6 = 60%)
|
|
* @returns {Array} Sous-ensemble aléatoire
|
|
*/
|
|
function selectRandomPersonalities(allPersonalities, percentage = 0.6) {
|
|
const count = Math.ceil(allPersonalities.length * percentage);
|
|
|
|
// Mélanger avec Fisher-Yates shuffle (meilleur que sort())
|
|
const shuffled = [...allPersonalities];
|
|
for (let i = shuffled.length - 1; i > 0; i--) {
|
|
const j = Math.floor(Math.random() * (i + 1));
|
|
[shuffled[i], shuffled[j]] = [shuffled[j], shuffled[i]];
|
|
}
|
|
|
|
return shuffled.slice(0, count);
|
|
}
|
|
|
|
/**
|
|
* NOUVELLE FONCTION: Sélection de 4 personnalités complémentaires pour le pipeline multi-AI
|
|
* @param {string} mc0 - Mot-clé principal
|
|
* @param {string} t0 - Titre principal
|
|
* @param {Array} personalities - Liste des personnalités
|
|
* @returns {Array} 4 personnalités sélectionnées pour chaque étape
|
|
*/
|
|
async function selectMultiplePersonalitiesWithAI(mc0, t0, personalities) {
|
|
try {
|
|
logSh(`🎭 Sélection MULTI-personnalités IA pour: ${mc0}`, "INFO");
|
|
|
|
// Sélection aléatoire de 80% des personnalités (plus large pour 4 choix)
|
|
const randomPersonalities = selectRandomPersonalities(personalities, 0.8);
|
|
const totalCount = personalities.length;
|
|
const selectedCount = randomPersonalities.length;
|
|
|
|
logSh(`🎲 Pool aléatoire: ${selectedCount}/${totalCount} personnalités disponibles`, "DEBUG");
|
|
logSh(`📋 Personnalités dans le pool: ${randomPersonalities.map(p => p.nom).join(', ')}`, "DEBUG");
|
|
|
|
const prompt = `Choisis 4 personnalités COMPLÉMENTAIRES pour générer du contenu sur "${mc0}":
|
|
|
|
OBJECTIF: Créer une équipe de 4 rédacteurs avec styles différents mais cohérents
|
|
|
|
PERSONNALITÉS DISPONIBLES:
|
|
${randomPersonalities.map(p => `- ${p.nom}: ${p.description} (Style: ${p.style})`).join('\n')}
|
|
|
|
RÔLES À ATTRIBUER:
|
|
1. GÉNÉRATEUR BASE: Personnalité technique/experte pour la génération initiale
|
|
2. ENHANCER TECHNIQUE: Personnalité commerciale/précise pour améliorer les termes techniques
|
|
3. FLUIDITÉ: Personnalité créative/littéraire pour améliorer les transitions
|
|
4. STYLE FINAL: Personnalité terrain/accessible pour le style final
|
|
|
|
CRITÈRES:
|
|
- 4 personnalités aux styles DIFFÉRENTS mais complémentaires
|
|
- Adapté au secteur: ${mc0}
|
|
- Variabilité maximale pour anti-détection
|
|
- Éviter les doublons de style
|
|
|
|
FORMAT DE RÉPONSE (EXACTEMENT 4 noms séparés par des virgules):
|
|
Nom1, Nom2, Nom3, Nom4`;
|
|
|
|
const requestData = {
|
|
model: "gpt-4o-mini",
|
|
messages: [{"role": "user", "content": prompt}],
|
|
max_tokens: 100,
|
|
temperature: 1.0
|
|
};
|
|
|
|
const response = await axios.post(CONFIG.openai.endpoint, requestData, {
|
|
headers: {
|
|
'Authorization': `Bearer ${CONFIG.openai.apiKey}`,
|
|
'Content-Type': 'application/json'
|
|
},
|
|
timeout: 300000
|
|
});
|
|
|
|
const selectedNames = response.data.choices[0].message.content.trim()
|
|
.split(',')
|
|
.map(name => name.trim());
|
|
|
|
logSh(`🔍 Noms retournés par IA: ${selectedNames.join(', ')}`, "DEBUG");
|
|
|
|
// Mapper aux vraies personnalités
|
|
const selectedPersonalities = [];
|
|
selectedNames.forEach(name => {
|
|
const personality = randomPersonalities.find(p => p.nom === name);
|
|
if (personality) {
|
|
selectedPersonalities.push(personality);
|
|
}
|
|
});
|
|
|
|
// Compléter si pas assez de personnalités trouvées (sécurité)
|
|
while (selectedPersonalities.length < 4 && randomPersonalities.length > selectedPersonalities.length) {
|
|
const remaining = randomPersonalities.filter(p =>
|
|
!selectedPersonalities.some(selected => selected.nom === p.nom)
|
|
);
|
|
if (remaining.length > 0) {
|
|
const randomIndex = Math.floor(Math.random() * remaining.length);
|
|
selectedPersonalities.push(remaining[randomIndex]);
|
|
} else {
|
|
break;
|
|
}
|
|
}
|
|
|
|
// Garantir exactement 4 personnalités
|
|
const final4Personalities = selectedPersonalities.slice(0, 4);
|
|
|
|
logSh(`✅ Équipe de 4 personnalités sélectionnée:`, "INFO");
|
|
final4Personalities.forEach((p, index) => {
|
|
const roles = ['BASE', 'TECHNIQUE', 'FLUIDITÉ', 'STYLE'];
|
|
logSh(` ${index + 1}. ${roles[index]}: ${p.nom} (${p.style})`, "INFO");
|
|
});
|
|
|
|
return final4Personalities;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ FATAL: Sélection multi-personnalités échouée: ${error.message}`, "ERROR");
|
|
throw new Error(`FATAL: Sélection multi-personnalités IA impossible - arrêt du workflow: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* FONCTION LEGACY: Sélection personnalité unique (maintenue pour compatibilité)
|
|
* @param {string} mc0 - Mot-clé principal
|
|
* @param {string} t0 - Titre principal
|
|
* @param {Array} personalities - Liste des personnalités
|
|
* @returns {object} Personnalité sélectionnée
|
|
*/
|
|
async function selectPersonalityWithAI(mc0, t0, personalities) {
|
|
try {
|
|
logSh(`🤖 Sélection personnalité IA UNIQUE pour: ${mc0}`, "DEBUG");
|
|
|
|
// Appeler la fonction multi et prendre seulement la première
|
|
const multiPersonalities = await selectMultiplePersonalitiesWithAI(mc0, t0, personalities);
|
|
const selectedPersonality = multiPersonalities[0];
|
|
|
|
logSh(`✅ Personnalité IA sélectionnée (mode legacy): ${selectedPersonality.nom}`, "INFO");
|
|
|
|
return selectedPersonality;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ FATAL: Sélection personnalité par IA échouée: ${error.message}`, "ERROR");
|
|
throw new Error(`FATAL: Sélection personnalité IA inaccessible - arrêt du workflow: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* CRÉER TEMPLATE XML PAR DÉFAUT quand colonne I contient un nom de fichier
|
|
* Utilise les données CSV disponibles pour créer un template robuste
|
|
*/
|
|
function createDefaultXMLTemplate() {
|
|
return `<?xml version="1.0" encoding="UTF-8"?>
|
|
<article>
|
|
<header>
|
|
<h1>|Titre_Principal{{T0}}{Rédige un titre H1 accrocheur de maximum 10 mots pour {{MC0}}. Style {{personality.style}}}|</h1>
|
|
<intro>|Introduction{{MC0}}{Rédige une introduction engageante de 2-3 phrases sur {{MC0}}. Ton {{personality.style}}, utilise {{personality.vocabulairePref}}}|</intro>
|
|
</header>
|
|
|
|
<main>
|
|
<section class="primary">
|
|
<h2>|Titre_H2_1{{MC+1_1}}{Crée un titre H2 informatif sur {{MC+1_1}}. Style {{personality.style}}}|</h2>
|
|
<p>|Paragraphe_1{{MC+1_1}}{Rédige un paragraphe détaillé de 4-5 phrases sur {{MC+1_1}}. Explique les avantages et caractéristiques. Ton {{personality.style}}}|</p>
|
|
</section>
|
|
|
|
<section class="secondary">
|
|
<h2>|Titre_H2_2{{MC+1_2}}{Titre H2 pour {{MC+1_2}}. Mets en valeur les points forts. Ton {{personality.style}}}|</h2>
|
|
<p>|Paragraphe_2{{MC+1_2}}{Paragraphe de 4-5 phrases sur {{MC+1_2}}. Détaille pourquoi c'est important pour {{MC0}}. Ton {{personality.style}}}|</p>
|
|
</section>
|
|
|
|
<section class="benefits">
|
|
<h2>|Titre_H2_3{{MC+1_3}}{Titre H2 sur les bénéfices de {{MC+1_3}}. Accrocheur et informatif}|</h2>
|
|
<p>|Paragraphe_3{{MC+1_3}}{Explique en 4-5 phrases les avantages de {{MC+1_3}} pour {{MC0}}. Ton {{personality.style}}}|</p>
|
|
</section>
|
|
</main>
|
|
|
|
<aside class="faq">
|
|
<h2>|FAQ_Titre{Titre de section FAQ accrocheur sur {{MC0}}}|</h2>
|
|
|
|
<div class="faq-item">
|
|
<h3>|Faq_q_1{{MC+1_1}}{Question fréquente sur {{MC+1_1}} et {{MC0}}}|</h3>
|
|
<p>|Faq_a_1{{MC+1_1}}{Réponse claire et précise. 2-3 phrases. Ton {{personality.style}}}|</p>
|
|
</div>
|
|
|
|
<div class="faq-item">
|
|
<h3>|Faq_q_2{{MC+1_2}}{Question pratique sur {{MC+1_2}} en lien avec {{MC0}}}|</h3>
|
|
<p>|Faq_a_2{{MC+1_2}}{Réponse détaillée et utile. 2-3 phrases explicatives. Ton {{personality.style}}}|</p>
|
|
</div>
|
|
|
|
<div class="faq-item">
|
|
<h3>|Faq_q_3{{MC+1_3}}{Question sur {{MC+1_3}} que se posent les clients}|</h3>
|
|
<p>|Faq_a_3{{MC+1_3}}{Réponse complète qui rassure et informe. 2-3 phrases. Ton {{personality.style}}}|</p>
|
|
</div>
|
|
</aside>
|
|
|
|
<footer>
|
|
<p>|Conclusion{{MC0}}{Conclusion engageante de 2 phrases sur {{MC0}}. Appel à l'action subtil. Ton {{personality.style}}}|</p>
|
|
</footer>
|
|
</article>`;
|
|
}
|
|
|
|
/**
|
|
* CRÉER FICHIERS DE DONNÉES D'EXEMPLE
|
|
* Fonction utilitaire pour initialiser les fichiers JSON
|
|
*/
|
|
async function createSampleDataFiles() {
|
|
try {
|
|
// Créer répertoire data s'il n'existe pas
|
|
await fs.mkdir('./data', { recursive: true });
|
|
|
|
// Exemple instructions.json
|
|
const sampleInstructions = [
|
|
{
|
|
slug: "plaque-test",
|
|
t0: "Plaque test signalétique",
|
|
mc0: "plaque signalétique",
|
|
"t-1": "Signalétique",
|
|
"l-1": "/signaletique/",
|
|
"mc+1": "plaque dibond, plaque aluminium, plaque PVC",
|
|
"t+1": "Plaque dibond, Plaque alu, Plaque PVC",
|
|
"l+1": "/plaque-dibond/, /plaque-aluminium/, /plaque-pvc/",
|
|
xmlFileName: "template-plaque.xml"
|
|
}
|
|
];
|
|
|
|
// Exemple personalities.json
|
|
const samplePersonalities = [
|
|
{
|
|
nom: "Marc",
|
|
description: "Expert technique en signalétique",
|
|
style: "professionnel et précis",
|
|
motsClesSecteurs: "technique,dibond,aluminium,impression",
|
|
vocabulairePref: "précision,qualité,expertise,performance",
|
|
connecteursPref: "par ailleurs,en effet,notamment,cependant",
|
|
erreursTypiques: "accord_proximite,repetition_legere",
|
|
longueurPhrases: "moyennes",
|
|
niveauTechnique: "élevé",
|
|
ctaStyle: "découvrir,choisir,commander",
|
|
defautsSimules: "fatigue_cognitive,hesitation_technique"
|
|
},
|
|
{
|
|
nom: "Sophie",
|
|
description: "Passionnée de décoration et design",
|
|
style: "familier et chaleureux",
|
|
motsClesSecteurs: "décoration,design,esthétique,tendances",
|
|
vocabulairePref: "joli,magnifique,tendance,style",
|
|
connecteursPref: "du coup,en fait,sinon,au fait",
|
|
erreursTypiques: "familiarite_excessive,expression_populaire",
|
|
longueurPhrases: "courtes",
|
|
niveauTechnique: "moyen",
|
|
ctaStyle: "craquer,adopter,foncer",
|
|
defautsSimules: "enthousiasme_variable,anecdote_personnelle"
|
|
}
|
|
];
|
|
|
|
// Écrire les fichiers
|
|
await fs.writeFile('./data/instructions.json', JSON.stringify(sampleInstructions, null, 2));
|
|
await fs.writeFile('./data/personalities.json', JSON.stringify(samplePersonalities, null, 2));
|
|
|
|
logSh('✅ Fichiers de données d\'exemple créés dans ./data/', "INFO");
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur création fichiers exemple: ${error.message}`, "ERROR");
|
|
}
|
|
}
|
|
|
|
// ============= EXPORTS NODE.JS =============
|
|
|
|
module.exports = {
|
|
getBrainConfig,
|
|
getPersonalities,
|
|
selectPersonalityWithAI,
|
|
selectMultiplePersonalitiesWithAI, // NOUVEAU: Export de la fonction multi-personnalités
|
|
selectRandomPersonalities,
|
|
parseCSVField,
|
|
readInstructionsData,
|
|
createSampleDataFiles,
|
|
createDefaultXMLTemplate,
|
|
CONFIG
|
|
};
|
|
|
|
// ============= TEST RAPIDE SI LANCÉ DIRECTEMENT =============
|
|
|
|
if (require.main === module) {
|
|
(async () => {
|
|
try {
|
|
logSh('🧪 Test BrainConfig Node.js...', "INFO");
|
|
|
|
// Créer fichiers exemple si nécessaire
|
|
try {
|
|
await fs.access('./data/instructions.json');
|
|
} catch {
|
|
await createSampleDataFiles();
|
|
}
|
|
|
|
// Test de la fonction principale
|
|
const result = await getBrainConfig(2);
|
|
|
|
if (result.success) {
|
|
logSh(`✅ Test réussi: ${result.data.personality.nom} pour ${result.data.mc0}`, "INFO");
|
|
} else {
|
|
logSh(`❌ Test échoué: ${result.error}`, "ERROR");
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur test: ${error.message}`, "ERROR");
|
|
}
|
|
})();
|
|
}
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/LLMManager.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: LLMManager.js
|
|
// Description: Hub central pour tous les appels LLM (Version Node.js)
|
|
// Support: Claude, OpenAI, Gemini, Deepseek, Moonshot, Mistral
|
|
// ========================================
|
|
|
|
const fetch = globalThis.fetch.bind(globalThis);
|
|
const { logSh } = require('./ErrorReporting');
|
|
|
|
// ============= CONFIGURATION CENTRALISÉE =============
|
|
|
|
const LLM_CONFIG = {
|
|
openai: {
|
|
apiKey: process.env.OPENAI_API_KEY || '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',
|
|
endpoint: 'https://api.openai.com/v1/chat/completions',
|
|
model: 'gpt-4o-mini',
|
|
headers: {
|
|
'Authorization': 'Bearer {API_KEY}',
|
|
'Content-Type': 'application/json'
|
|
},
|
|
temperature: 0.7,
|
|
timeout: 300000, // 5 minutes
|
|
retries: 3
|
|
},
|
|
|
|
claude: {
|
|
apiKey: process.env.CLAUDE_API_KEY || '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',
|
|
endpoint: 'https://api.anthropic.com/v1/messages',
|
|
model: 'claude-sonnet-4-20250514',
|
|
headers: {
|
|
'x-api-key': '{API_KEY}',
|
|
'Content-Type': 'application/json',
|
|
'anthropic-version': '2023-06-01'
|
|
},
|
|
temperature: 0.7,
|
|
timeout: 300000, // 5 minutes
|
|
retries: 6
|
|
},
|
|
|
|
gemini: {
|
|
apiKey: process.env.GEMINI_API_KEY || 'AIzaSyAMzmIGbW5nJlBG5Qyr35sdjb3U2bIBtoE',
|
|
endpoint: 'https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent',
|
|
model: 'gemini-2.5-flash',
|
|
headers: {
|
|
'Content-Type': 'application/json'
|
|
},
|
|
temperature: 0.7,
|
|
maxTokens: 6000,
|
|
timeout: 300000, // 5 minutes
|
|
retries: 3
|
|
},
|
|
|
|
deepseek: {
|
|
apiKey: process.env.DEEPSEEK_API_KEY || 'sk-6e02bc9513884bb8b92b9920524e17b5',
|
|
endpoint: 'https://api.deepseek.com/v1/chat/completions',
|
|
model: 'deepseek-chat',
|
|
headers: {
|
|
'Authorization': 'Bearer {API_KEY}',
|
|
'Content-Type': 'application/json'
|
|
},
|
|
temperature: 0.7,
|
|
timeout: 300000, // 5 minutes
|
|
retries: 3
|
|
},
|
|
|
|
moonshot: {
|
|
apiKey: process.env.MOONSHOT_API_KEY || 'sk-zU9gyNkux2zcsj61cdKfztuP1Jozr6lFJ9viUJRPD8p8owhL',
|
|
endpoint: 'https://api.moonshot.ai/v1/chat/completions',
|
|
model: 'moonshot-v1-32k',
|
|
headers: {
|
|
'Authorization': 'Bearer {API_KEY}',
|
|
'Content-Type': 'application/json'
|
|
},
|
|
temperature: 0.7,
|
|
timeout: 300000, // 5 minutes
|
|
retries: 3
|
|
},
|
|
|
|
mistral: {
|
|
apiKey: process.env.MISTRAL_API_KEY || 'wESikMCIuixajSH8WHCiOV2z5sevgmVF',
|
|
endpoint: 'https://api.mistral.ai/v1/chat/completions',
|
|
model: 'mistral-small-latest',
|
|
headers: {
|
|
'Authorization': 'Bearer {API_KEY}',
|
|
'Content-Type': 'application/json'
|
|
},
|
|
max_tokens: 5000,
|
|
temperature: 0.7,
|
|
timeout: 300000, // 5 minutes
|
|
retries: 3
|
|
}
|
|
};
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
const sleep = (ms) => new Promise(resolve => setTimeout(resolve, ms));
|
|
|
|
// ============= INTERFACE UNIVERSELLE =============
|
|
|
|
/**
|
|
* Fonction principale pour appeler n'importe quel LLM
|
|
* @param {string} llmProvider - claude|openai|gemini|deepseek|moonshot|mistral
|
|
* @param {string} prompt - Le prompt à envoyer
|
|
* @param {object} options - Options personnalisées (température, tokens, etc.)
|
|
* @param {object} personality - Personnalité pour contexte système
|
|
* @returns {Promise<string>} - Réponse générée
|
|
*/
|
|
async function callLLM(llmProvider, prompt, options = {}, personality = null) {
|
|
const startTime = Date.now();
|
|
|
|
try {
|
|
// Vérifier si le provider existe
|
|
if (!LLM_CONFIG[llmProvider]) {
|
|
throw new Error(`Provider LLM inconnu: ${llmProvider}`);
|
|
}
|
|
|
|
// Vérifier si l'API key est configurée
|
|
const config = LLM_CONFIG[llmProvider];
|
|
if (!config.apiKey || config.apiKey.startsWith('VOTRE_CLE_')) {
|
|
throw new Error(`Clé API manquante pour ${llmProvider}`);
|
|
}
|
|
|
|
logSh(`🤖 Appel LLM: ${llmProvider.toUpperCase()} (${config.model}) | Personnalité: ${personality?.nom || 'aucune'}`, 'DEBUG');
|
|
|
|
// 📢 AFFICHAGE PROMPT COMPLET POUR DEBUG AVEC INFO IA
|
|
logSh(`\n🔍 ===== PROMPT ENVOYÉ À ${llmProvider.toUpperCase()} (${config.model}) | PERSONNALITÉ: ${personality?.nom || 'AUCUNE'} =====`, 'PROMPT');
|
|
logSh(prompt, 'PROMPT');
|
|
|
|
// 📤 LOG LLM REQUEST COMPLET
|
|
logSh(`📤 LLM REQUEST [${llmProvider.toUpperCase()}] (${config.model}) | Personnalité: ${personality?.nom || 'AUCUNE'}`, 'LLM');
|
|
logSh(prompt, 'LLM');
|
|
|
|
// Préparer la requête selon le provider
|
|
const requestData = buildRequestData(llmProvider, prompt, options, personality);
|
|
|
|
// Effectuer l'appel avec retry logic
|
|
const response = await callWithRetry(llmProvider, requestData, config);
|
|
|
|
// Parser la réponse selon le format du provider
|
|
const content = parseResponse(llmProvider, response);
|
|
|
|
// 📥 LOG LLM RESPONSE COMPLET
|
|
logSh(`📥 LLM RESPONSE [${llmProvider.toUpperCase()}] (${config.model}) | Durée: ${Date.now() - startTime}ms`, 'LLM');
|
|
logSh(content, 'LLM');
|
|
|
|
const duration = Date.now() - startTime;
|
|
logSh(`✅ ${llmProvider.toUpperCase()} (${personality?.nom || 'sans personnalité'}) réponse en ${duration}ms`, 'INFO');
|
|
|
|
// Enregistrer les stats d'usage
|
|
await recordUsageStats(llmProvider, prompt.length, content.length, duration);
|
|
|
|
return content;
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ Erreur ${llmProvider.toUpperCase()} (${personality?.nom || 'sans personnalité'}): ${error.toString()}`, 'ERROR');
|
|
|
|
// Enregistrer l'échec
|
|
await recordUsageStats(llmProvider, prompt.length, 0, duration, error.toString());
|
|
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
// ============= CONSTRUCTION DES REQUÊTES =============
|
|
|
|
function buildRequestData(provider, prompt, options, personality) {
|
|
const config = LLM_CONFIG[provider];
|
|
const temperature = options.temperature || config.temperature;
|
|
const maxTokens = options.maxTokens || config.maxTokens;
|
|
|
|
// Construire le système prompt si personnalité fournie
|
|
const systemPrompt = personality ?
|
|
`Tu es ${personality.nom}. ${personality.description}. Style: ${personality.style}` :
|
|
'Tu es un assistant expert.';
|
|
|
|
switch (provider) {
|
|
case 'openai':
|
|
case 'deepseek':
|
|
case 'moonshot':
|
|
case 'mistral':
|
|
return {
|
|
model: config.model,
|
|
messages: [
|
|
{ role: 'system', content: systemPrompt },
|
|
{ role: 'user', content: prompt }
|
|
],
|
|
max_tokens: maxTokens,
|
|
temperature: temperature,
|
|
stream: false
|
|
};
|
|
|
|
case 'claude':
|
|
return {
|
|
model: config.model,
|
|
max_tokens: maxTokens,
|
|
temperature: temperature,
|
|
system: systemPrompt,
|
|
messages: [
|
|
{ role: 'user', content: prompt }
|
|
]
|
|
};
|
|
|
|
case 'gemini':
|
|
return {
|
|
contents: [{
|
|
parts: [{
|
|
text: `${systemPrompt}\n\n${prompt}`
|
|
}]
|
|
}],
|
|
generationConfig: {
|
|
temperature: temperature,
|
|
maxOutputTokens: maxTokens
|
|
}
|
|
};
|
|
|
|
default:
|
|
throw new Error(`Format de requête non supporté pour ${provider}`);
|
|
}
|
|
}
|
|
|
|
// ============= APPELS AVEC RETRY =============
|
|
|
|
async function callWithRetry(provider, requestData, config) {
|
|
let lastError;
|
|
|
|
for (let attempt = 1; attempt <= config.retries; attempt++) {
|
|
try {
|
|
logSh(`🔄 Tentative ${attempt}/${config.retries} pour ${provider.toUpperCase()}`, 'DEBUG');
|
|
|
|
// Préparer les headers avec la clé API
|
|
const headers = {};
|
|
Object.keys(config.headers).forEach(key => {
|
|
headers[key] = config.headers[key].replace('{API_KEY}', config.apiKey);
|
|
});
|
|
|
|
// URL avec clé API pour Gemini (cas spécial)
|
|
let url = config.endpoint;
|
|
if (provider === 'gemini') {
|
|
url += `?key=${config.apiKey}`;
|
|
}
|
|
|
|
const options = {
|
|
method: 'POST',
|
|
headers: headers,
|
|
body: JSON.stringify(requestData),
|
|
timeout: config.timeout
|
|
};
|
|
|
|
const response = await fetch(url, options);
|
|
const responseText = await response.text();
|
|
|
|
if (response.ok) {
|
|
return JSON.parse(responseText);
|
|
} else if (response.status === 429) {
|
|
// Rate limiting - attendre plus longtemps
|
|
const waitTime = Math.pow(2, attempt) * 1000; // Exponential backoff
|
|
logSh(`⏳ Rate limit ${provider.toUpperCase()}, attente ${waitTime}ms`, 'WARNING');
|
|
await sleep(waitTime);
|
|
continue;
|
|
} else {
|
|
throw new Error(`HTTP ${response.status}: ${responseText}`);
|
|
}
|
|
|
|
} catch (error) {
|
|
lastError = error;
|
|
|
|
if (attempt < config.retries) {
|
|
const waitTime = 1000 * attempt;
|
|
logSh(`⚠ Erreur tentative ${attempt}: ${error.toString()}, retry dans ${waitTime}ms`, 'WARNING');
|
|
await sleep(waitTime);
|
|
}
|
|
}
|
|
}
|
|
|
|
throw new Error(`Échec après ${config.retries} tentatives: ${lastError.toString()}`);
|
|
}
|
|
|
|
// ============= PARSING DES RÉPONSES =============
|
|
|
|
function parseResponse(provider, responseData) {
|
|
try {
|
|
switch (provider) {
|
|
case 'openai':
|
|
case 'deepseek':
|
|
case 'moonshot':
|
|
case 'mistral':
|
|
return responseData.choices[0].message.content.trim();
|
|
|
|
case 'claude':
|
|
return responseData.content[0].text.trim();
|
|
|
|
case 'gemini':
|
|
const candidate = responseData.candidates[0];
|
|
|
|
// Vérifications multiples pour Gemini 2.5
|
|
if (candidate && candidate.content && candidate.content.parts && candidate.content.parts[0] && candidate.content.parts[0].text) {
|
|
return candidate.content.parts[0].text.trim();
|
|
} else if (candidate && candidate.text) {
|
|
return candidate.text.trim();
|
|
} else if (candidate && candidate.content && candidate.content.text) {
|
|
return candidate.content.text.trim();
|
|
} else {
|
|
// Debug : logger la structure complète
|
|
logSh('Gemini structure complète: ' + JSON.stringify(responseData), 'DEBUG');
|
|
return '[Gemini: pas de texte généré - problème modèle]';
|
|
}
|
|
default:
|
|
throw new Error(`Parser non supporté pour ${provider}`);
|
|
}
|
|
} catch (error) {
|
|
logSh(`❌ Erreur parsing ${provider}: ${error.toString()}`, 'ERROR');
|
|
logSh(`Response brute: ${JSON.stringify(responseData)}`, 'DEBUG');
|
|
throw new Error(`Impossible de parser la réponse ${provider}: ${error.toString()}`);
|
|
}
|
|
}
|
|
|
|
// ============= GESTION DES STATISTIQUES =============
|
|
|
|
async function recordUsageStats(provider, promptTokens, responseTokens, duration, error = null) {
|
|
try {
|
|
// TODO: Adapter selon votre système de stockage Node.js
|
|
// Peut être une base de données, un fichier, MongoDB, etc.
|
|
const statsData = {
|
|
timestamp: new Date(),
|
|
provider: provider,
|
|
model: LLM_CONFIG[provider].model,
|
|
promptTokens: promptTokens,
|
|
responseTokens: responseTokens,
|
|
duration: duration,
|
|
error: error || ''
|
|
};
|
|
|
|
// Exemple: log vers console ou fichier
|
|
logSh(`📊 Stats: ${JSON.stringify(statsData)}`, 'DEBUG');
|
|
|
|
// TODO: Implémenter sauvegarde réelle (DB, fichier, etc.)
|
|
|
|
} catch (statsError) {
|
|
// Ne pas faire planter le workflow si les stats échouent
|
|
logSh(`⚠ Erreur enregistrement stats: ${statsError.toString()}`, 'WARNING');
|
|
}
|
|
}
|
|
|
|
// ============= FONCTIONS UTILITAIRES =============
|
|
|
|
/**
|
|
* Tester la connectivité de tous les LLMs
|
|
*/
|
|
async function testAllLLMs() {
|
|
const testPrompt = "Dis bonjour en 5 mots maximum.";
|
|
const results = {};
|
|
|
|
const allProviders = Object.keys(LLM_CONFIG);
|
|
|
|
for (const provider of allProviders) {
|
|
try {
|
|
logSh(`🧪 Test ${provider}...`, 'INFO');
|
|
|
|
const response = await callLLM(provider, testPrompt);
|
|
results[provider] = {
|
|
status: 'SUCCESS',
|
|
response: response,
|
|
model: LLM_CONFIG[provider].model
|
|
};
|
|
|
|
} catch (error) {
|
|
results[provider] = {
|
|
status: 'ERROR',
|
|
error: error.toString(),
|
|
model: LLM_CONFIG[provider].model
|
|
};
|
|
}
|
|
|
|
// Petit délai entre tests
|
|
await sleep(500);
|
|
}
|
|
|
|
logSh(`📊 Tests terminés: ${JSON.stringify(results, null, 2)}`, 'INFO');
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Obtenir les providers disponibles (avec clés API valides)
|
|
*/
|
|
function getAvailableProviders() {
|
|
const available = [];
|
|
|
|
Object.keys(LLM_CONFIG).forEach(provider => {
|
|
const config = LLM_CONFIG[provider];
|
|
if (config.apiKey && !config.apiKey.startsWith('VOTRE_CLE_')) {
|
|
available.push(provider);
|
|
}
|
|
});
|
|
|
|
return available;
|
|
}
|
|
|
|
/**
|
|
* Obtenir des statistiques d'usage par provider
|
|
*/
|
|
async function getUsageStats() {
|
|
try {
|
|
// TODO: Adapter selon votre système de stockage
|
|
// Pour l'instant retourne un message par défaut
|
|
return { message: 'Statistiques non implémentées en Node.js' };
|
|
|
|
} catch (error) {
|
|
return { error: error.toString() };
|
|
}
|
|
}
|
|
|
|
// ============= MIGRATION DE L'ANCIEN CODE =============
|
|
|
|
/**
|
|
* Fonction de compatibilité pour remplacer votre ancien callOpenAI()
|
|
* Maintient la même signature pour ne pas casser votre code existant
|
|
*/
|
|
async function callOpenAI(prompt, personality) {
|
|
return await callLLM('openai', prompt, {}, personality);
|
|
}
|
|
|
|
// ============= EXPORTS POUR TESTS =============
|
|
|
|
/**
|
|
* Fonction de test rapide
|
|
*/
|
|
async function testLLMManager() {
|
|
logSh('🚀 Test du LLM Manager Node.js...', 'INFO');
|
|
|
|
// Test des providers disponibles
|
|
const available = getAvailableProviders();
|
|
logSh('Providers disponibles: ' + available.join(', ') + ' (' + available.length + '/6)', 'INFO');
|
|
|
|
// Test d'appel simple sur chaque provider disponible
|
|
for (const provider of available) {
|
|
try {
|
|
logSh(`🧪 Test ${provider}...`, 'DEBUG');
|
|
const startTime = Date.now();
|
|
|
|
const response = await callLLM(provider, 'Dis juste "Test OK"');
|
|
const duration = Date.now() - startTime;
|
|
|
|
logSh(`✅ Test ${provider} réussi: "${response}" (${duration}ms)`, 'INFO');
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Test ${provider} échoué: ${error.toString()}`, 'ERROR');
|
|
}
|
|
|
|
// Petit délai pour éviter rate limits
|
|
await sleep(500);
|
|
}
|
|
|
|
// Test spécifique OpenAI (compatibilité avec ancien code)
|
|
try {
|
|
logSh('🎯 Test spécifique OpenAI (compatibilité)...', 'DEBUG');
|
|
const response = await callLLM('openai', 'Dis juste "Test OK"');
|
|
logSh('✅ Test OpenAI compatibilité: ' + response, 'INFO');
|
|
} catch (error) {
|
|
logSh('❌ Test OpenAI compatibilité échoué: ' + error.toString(), 'ERROR');
|
|
}
|
|
|
|
// Afficher les stats d'usage
|
|
try {
|
|
logSh('📊 Récupération statistiques d\'usage...', 'DEBUG');
|
|
const stats = await getUsageStats();
|
|
|
|
if (stats.error) {
|
|
logSh('⚠ Erreur récupération stats: ' + stats.error, 'WARNING');
|
|
} else if (stats.message) {
|
|
logSh('📊 Stats: ' + stats.message, 'INFO');
|
|
} else {
|
|
// Formatter les stats pour les logs
|
|
Object.keys(stats).forEach(provider => {
|
|
const s = stats[provider];
|
|
logSh(`📈 ${provider}: ${s.calls} appels, ${s.successRate}% succès, ${s.avgDuration}ms moyen`, 'INFO');
|
|
});
|
|
}
|
|
} catch (error) {
|
|
logSh('❌ Erreur lors de la récupération des stats: ' + error.toString(), 'ERROR');
|
|
}
|
|
|
|
// Résumé final
|
|
const workingCount = available.length;
|
|
const totalProviders = Object.keys(LLM_CONFIG).length;
|
|
|
|
if (workingCount === totalProviders) {
|
|
logSh(`✅ Test LLM Manager COMPLET: ${workingCount}/${totalProviders} providers opérationnels`, 'INFO');
|
|
} else if (workingCount >= 2) {
|
|
logSh(`✅ Test LLM Manager PARTIEL: ${workingCount}/${totalProviders} providers opérationnels (suffisant pour DNA Mixing)`, 'INFO');
|
|
} else {
|
|
logSh(`❌ Test LLM Manager INSUFFISANT: ${workingCount}/${totalProviders} providers opérationnels (minimum 2 requis)`, 'ERROR');
|
|
}
|
|
|
|
logSh('🏁 Test LLM Manager terminé', 'INFO');
|
|
}
|
|
|
|
/**
|
|
* Version complète avec test de tous les providers (même non configurés)
|
|
*/
|
|
async function testLLMManagerComplete() {
|
|
logSh('🚀 Test COMPLET du LLM Manager (tous providers)...', 'INFO');
|
|
|
|
const allProviders = Object.keys(LLM_CONFIG);
|
|
logSh(`Providers configurés: ${allProviders.join(', ')}`, 'INFO');
|
|
|
|
const results = {
|
|
configured: 0,
|
|
working: 0,
|
|
failed: 0
|
|
};
|
|
|
|
for (const provider of allProviders) {
|
|
const config = LLM_CONFIG[provider];
|
|
|
|
// Vérifier si configuré
|
|
if (!config.apiKey || config.apiKey.startsWith('VOTRE_CLE_')) {
|
|
logSh(`⚙️ ${provider}: NON CONFIGURÉ (clé API manquante)`, 'WARNING');
|
|
continue;
|
|
}
|
|
|
|
results.configured++;
|
|
|
|
try {
|
|
logSh(`🧪 Test ${provider} (${config.model})...`, 'DEBUG');
|
|
const startTime = Date.now();
|
|
|
|
const response = await callLLM(provider, 'Réponds "OK" seulement.', { maxTokens: 100 });
|
|
const duration = Date.now() - startTime;
|
|
|
|
results.working++;
|
|
logSh(`✅ ${provider}: "${response.trim()}" (${duration}ms)`, 'INFO');
|
|
|
|
} catch (error) {
|
|
results.failed++;
|
|
logSh(`❌ ${provider}: ${error.toString()}`, 'ERROR');
|
|
}
|
|
|
|
// Délai entre tests
|
|
await sleep(700);
|
|
}
|
|
|
|
// Résumé final complet
|
|
logSh(`📊 RÉSUMÉ FINAL:`, 'INFO');
|
|
logSh(` • Providers total: ${allProviders.length}`, 'INFO');
|
|
logSh(` • Configurés: ${results.configured}`, 'INFO');
|
|
logSh(` • Fonctionnels: ${results.working}`, 'INFO');
|
|
logSh(` • En échec: ${results.failed}`, 'INFO');
|
|
|
|
const status = results.working >= 4 ? 'EXCELLENT' :
|
|
results.working >= 2 ? 'BON' : 'INSUFFISANT';
|
|
|
|
logSh(`🏆 STATUS: ${status} (${results.working} LLMs opérationnels)`,
|
|
status === 'INSUFFISANT' ? 'ERROR' : 'INFO');
|
|
|
|
logSh('🏁 Test LLM Manager COMPLET terminé', 'INFO');
|
|
|
|
return {
|
|
total: allProviders.length,
|
|
configured: results.configured,
|
|
working: results.working,
|
|
failed: results.failed,
|
|
status: status
|
|
};
|
|
}
|
|
|
|
// ============= EXPORTS MODULE =============
|
|
|
|
module.exports = {
|
|
callLLM,
|
|
callOpenAI,
|
|
testAllLLMs,
|
|
getAvailableProviders,
|
|
getUsageStats,
|
|
testLLMManager,
|
|
testLLMManagerComplete,
|
|
LLM_CONFIG
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/ElementExtraction.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: lib/element-extraction.js - CONVERTI POUR NODE.JS
|
|
// Description: Extraction et parsing des éléments XML
|
|
// ========================================
|
|
|
|
// 🔄 NODE.JS IMPORTS
|
|
const { logSh } = require('./ErrorReporting');
|
|
|
|
// ============= EXTRACTION PRINCIPALE =============
|
|
|
|
async function extractElements(xmlTemplate, csvData) {
|
|
try {
|
|
await logSh('Extraction éléments avec séparation tag/contenu...', 'DEBUG');
|
|
|
|
const regex = /\|([^|]+)\|/g;
|
|
const elements = [];
|
|
let match;
|
|
|
|
while ((match = regex.exec(xmlTemplate)) !== null) {
|
|
const fullMatch = match[1]; // Ex: "Titre_H1_1{{T0}}" ou "Titre_H3_3{{MC+1_3}}"
|
|
|
|
// Séparer nom du tag et variables
|
|
const nameMatch = fullMatch.match(/^([^{]+)/);
|
|
const variablesMatch = fullMatch.match(/\{\{([^}]+)\}\}/g);
|
|
|
|
// FIX REGEX INSTRUCTIONS - Enlever d'abord les {{variables}} puis chercher {instructions}
|
|
const withoutVariables = fullMatch.replace(/\{\{[^}]+\}\}/g, '');
|
|
const instructionsMatch = withoutVariables.match(/\{([^}]+)\}/);
|
|
|
|
const tagName = nameMatch ? nameMatch[1].trim() : fullMatch.split('{')[0];
|
|
|
|
// TAG PUR (sans variables)
|
|
const pureTag = `|${tagName}|`;
|
|
|
|
// RÉSOUDRE le contenu des variables
|
|
const resolvedContent = resolveVariablesContent(variablesMatch, csvData);
|
|
|
|
elements.push({
|
|
originalTag: pureTag, // ← TAG PUR : |Titre_H3_3|
|
|
name: tagName, // ← Titre_H3_3
|
|
variables: variablesMatch || [], // ← [{{MC+1_3}}]
|
|
resolvedContent: resolvedContent, // ← "Plaque de rue en aluminium"
|
|
instructions: instructionsMatch ? instructionsMatch[1] : null,
|
|
type: getElementType(tagName),
|
|
originalFullMatch: fullMatch // ← Backup si besoin
|
|
});
|
|
|
|
await logSh(`Tag séparé: ${pureTag} → "${resolvedContent}"`, 'DEBUG');
|
|
}
|
|
|
|
await logSh(`${elements.length} éléments extraits avec séparation`, 'INFO');
|
|
return elements;
|
|
|
|
} catch (error) {
|
|
await logSh(`Erreur extractElements: ${error}`, 'ERROR');
|
|
return [];
|
|
}
|
|
}
|
|
|
|
// ============= RÉSOLUTION VARIABLES - IDENTIQUE =============
|
|
|
|
function resolveVariablesContent(variablesMatch, csvData) {
|
|
if (!variablesMatch || variablesMatch.length === 0) {
|
|
return ""; // Pas de variables à résoudre
|
|
}
|
|
|
|
let resolvedContent = "";
|
|
|
|
variablesMatch.forEach(variable => {
|
|
const cleanVar = variable.replace(/[{}]/g, ''); // Enlever {{ }}
|
|
|
|
switch (cleanVar) {
|
|
case 'T0':
|
|
resolvedContent += csvData.t0;
|
|
break;
|
|
case 'MC0':
|
|
resolvedContent += csvData.mc0;
|
|
break;
|
|
case 'T-1':
|
|
resolvedContent += csvData.tMinus1;
|
|
break;
|
|
case 'L-1':
|
|
resolvedContent += csvData.lMinus1;
|
|
break;
|
|
default:
|
|
// Gérer MC+1_1, MC+1_2, etc.
|
|
if (cleanVar.startsWith('MC+1_')) {
|
|
const index = parseInt(cleanVar.split('_')[1]) - 1;
|
|
const mcPlus1 = csvData.mcPlus1.split(',').map(s => s.trim());
|
|
resolvedContent += mcPlus1[index] || `[${cleanVar} non défini]`;
|
|
}
|
|
else if (cleanVar.startsWith('T+1_')) {
|
|
const index = parseInt(cleanVar.split('_')[1]) - 1;
|
|
const tPlus1 = csvData.tPlus1.split(',').map(s => s.trim());
|
|
resolvedContent += tPlus1[index] || `[${cleanVar} non défini]`;
|
|
}
|
|
else if (cleanVar.startsWith('L+1_')) {
|
|
const index = parseInt(cleanVar.split('_')[1]) - 1;
|
|
const lPlus1 = csvData.lPlus1.split(',').map(s => s.trim());
|
|
resolvedContent += lPlus1[index] || `[${cleanVar} non défini]`;
|
|
}
|
|
else {
|
|
resolvedContent += `[${cleanVar} non résolu]`;
|
|
}
|
|
break;
|
|
}
|
|
});
|
|
|
|
return resolvedContent;
|
|
}
|
|
|
|
// ============= CLASSIFICATION ÉLÉMENTS - IDENTIQUE =============
|
|
|
|
function getElementType(name) {
|
|
if (name.includes('Titre_H1')) return 'titre_h1';
|
|
if (name.includes('Titre_H2')) return 'titre_h2';
|
|
if (name.includes('Titre_H3')) return 'titre_h3';
|
|
if (name.includes('Intro_')) return 'intro';
|
|
if (name.includes('Txt_')) return 'texte';
|
|
if (name.includes('Faq_q')) return 'faq_question';
|
|
if (name.includes('Faq_a')) return 'faq_reponse';
|
|
if (name.includes('Faq_H3')) return 'faq_titre';
|
|
return 'autre';
|
|
}
|
|
|
|
// ============= GÉNÉRATION SÉQUENTIELLE - ADAPTÉE =============
|
|
|
|
async function generateAllContent(elements, csvData, xmlTemplate) {
|
|
await logSh(`Début génération pour ${elements.length} éléments`, 'INFO');
|
|
|
|
const generatedContent = {};
|
|
|
|
for (let index = 0; index < elements.length; index++) {
|
|
const element = elements[index];
|
|
|
|
try {
|
|
await logSh(`Élément ${index + 1}/${elements.length}: ${element.name}`, 'DEBUG');
|
|
|
|
const prompt = createPromptForElement(element, csvData);
|
|
await logSh(`Prompt créé: ${prompt}`, 'DEBUG');
|
|
|
|
// 🔄 NODE.JS : Import callOpenAI depuis LLM manager
|
|
const { callLLM } = require('./LLMManager');
|
|
const content = await callLLM('openai', prompt, {}, csvData.personality);
|
|
|
|
await logSh(`Contenu reçu: ${content}`, 'DEBUG');
|
|
|
|
generatedContent[element.originalTag] = content;
|
|
|
|
// 🔄 NODE.JS : Pas de Utilities.sleep(), les appels API gèrent leur rate limiting
|
|
|
|
} catch (error) {
|
|
await logSh(`ERREUR élément ${element.name}: ${error.toString()}`, 'ERROR');
|
|
generatedContent[element.originalTag] = `[Erreur génération: ${element.name}]`;
|
|
}
|
|
}
|
|
|
|
await logSh(`Génération terminée. ${Object.keys(generatedContent).length} éléments`, 'INFO');
|
|
return generatedContent;
|
|
}
|
|
|
|
// ============= PARSING STRUCTURE - IDENTIQUE =============
|
|
|
|
function parseElementStructure(element) {
|
|
// NETTOYER le nom : enlever <strong>, </strong>, {{...}}, {...}
|
|
let cleanName = element.name
|
|
.replace(/<\/?strong>/g, '') // ← ENLEVER <strong>
|
|
.replace(/\{\{[^}]*\}\}/g, '') // Enlever {{MC0}}
|
|
.replace(/\{[^}]*\}/g, ''); // Enlever {instructions}
|
|
|
|
const parts = cleanName.split('_');
|
|
|
|
return {
|
|
type: parts[0],
|
|
level: parts[1],
|
|
indices: parts.slice(2).map(Number),
|
|
hierarchyPath: parts.slice(1).join('_'),
|
|
originalElement: element,
|
|
variables: element.variables || [],
|
|
instructions: element.instructions
|
|
};
|
|
}
|
|
|
|
// ============= HIÉRARCHIE INTELLIGENTE - ADAPTÉE =============
|
|
|
|
async function buildSmartHierarchy(elements) {
|
|
const hierarchy = {};
|
|
|
|
elements.forEach(element => {
|
|
const structure = parseElementStructure(element);
|
|
const path = structure.hierarchyPath;
|
|
|
|
if (!hierarchy[path]) {
|
|
hierarchy[path] = {
|
|
title: null,
|
|
text: null,
|
|
questions: [],
|
|
children: {}
|
|
};
|
|
}
|
|
|
|
// Associer intelligemment
|
|
if (structure.type === 'Titre') {
|
|
hierarchy[path].title = structure; // Tout l'objet avec variables + instructions
|
|
} else if (structure.type === 'Txt') {
|
|
hierarchy[path].text = structure;
|
|
} else if (structure.type === 'Intro') {
|
|
hierarchy[path].text = structure;
|
|
} else if (structure.type === 'Faq') {
|
|
hierarchy[path].questions.push(structure);
|
|
}
|
|
});
|
|
|
|
// ← LIGNE COMPILÉE
|
|
const mappingSummary = Object.keys(hierarchy).map(path => {
|
|
const section = hierarchy[path];
|
|
return `${path}:[T:${section.title ? '✓' : '✗'} Txt:${section.text ? '✓' : '✗'} FAQ:${section.questions.length}]`;
|
|
}).join(' | ');
|
|
|
|
await logSh('Correspondances: ' + mappingSummary, 'DEBUG');
|
|
|
|
return hierarchy;
|
|
}
|
|
|
|
// ============= PARSERS RÉPONSES - ADAPTÉS =============
|
|
|
|
async function parseTitlesResponse(response, allTitles) {
|
|
const results = {};
|
|
|
|
// Utiliser regex pour extraire [TAG] contenu
|
|
const regex = /\[([^\]]+)\]\s*\n([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
const content = match[2].trim();
|
|
|
|
// Nettoyer le contenu (enlever # et balises HTML si présentes)
|
|
const cleanContent = content
|
|
.replace(/^#+\s*/, '') // Enlever # du début
|
|
.replace(/<\/?[^>]+(>|$)/g, ""); // Enlever balises HTML
|
|
|
|
results[`|${tag}|`] = cleanContent;
|
|
|
|
await logSh(`✓ Titre parsé [${tag}]: "${cleanContent}"`, 'DEBUG');
|
|
}
|
|
|
|
// Fallback si parsing échoue
|
|
if (Object.keys(results).length === 0) {
|
|
await logSh('Parsing titres échoué, fallback ligne par ligne', 'WARNING');
|
|
const lines = response.split('\n').filter(line => line.trim());
|
|
|
|
allTitles.forEach((titleInfo, index) => {
|
|
if (lines[index]) {
|
|
results[titleInfo.tag] = lines[index].trim();
|
|
}
|
|
});
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
async function parseTextsResponse(response, allTexts) {
|
|
const results = {};
|
|
|
|
await logSh('Parsing réponse textes avec vrais tags...', 'DEBUG');
|
|
|
|
// Utiliser regex pour extraire [TAG] contenu avec les vrais noms
|
|
const regex = /\[([^\]]+)\]\s*\n([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
const content = match[2].trim();
|
|
|
|
// Nettoyer le contenu
|
|
const cleanContent = content.replace(/^#+\s*/, '').replace(/<\/?[^>]+(>|$)/g, "");
|
|
|
|
results[`|${tag}|`] = cleanContent;
|
|
|
|
await logSh(`✓ Texte parsé [${tag}]: "${cleanContent}"`, 'DEBUG');
|
|
}
|
|
|
|
// Fallback si parsing échoue - mapper par position
|
|
if (Object.keys(results).length === 0) {
|
|
await logSh('Parsing textes échoué, fallback ligne par ligne', 'WARNING');
|
|
|
|
const lines = response.split('\n')
|
|
.map(line => line.trim())
|
|
.filter(line => line.length > 0 && !line.startsWith('['));
|
|
|
|
for (let index = 0; index < allTexts.length; index++) {
|
|
const textInfo = allTexts[index];
|
|
if (index < lines.length) {
|
|
let content = lines[index];
|
|
content = content.replace(/^\d+\.\s*/, ''); // Enlever "1. " si présent
|
|
results[textInfo.tag] = content;
|
|
|
|
await logSh(`✓ Texte fallback ${index + 1} → ${textInfo.tag}: "${content}"`, 'DEBUG');
|
|
} else {
|
|
await logSh(`✗ Pas assez de lignes pour ${textInfo.tag}`, 'WARNING');
|
|
results[textInfo.tag] = `[Texte manquant ${index + 1}]`;
|
|
}
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= PARSER FAQ SPÉCIALISÉ - ADAPTÉ =============
|
|
|
|
async function parseFAQPairsResponse(response, faqPairs) {
|
|
const results = {};
|
|
|
|
await logSh('Parsing réponse paires FAQ...', 'DEBUG');
|
|
|
|
// Parser avec regex pour capturer question + réponse
|
|
const regex = /\[([^\]]+)\]\s*\n([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
const content = match[2].trim();
|
|
|
|
const cleanContent = content.replace(/^#+\s*/, '').replace(/<\/?[^>]+(>|$)/g, "");
|
|
|
|
parsedItems[tag] = cleanContent;
|
|
|
|
await logSh(`✓ Item FAQ parsé [${tag}]: "${cleanContent}"`, 'DEBUG');
|
|
}
|
|
|
|
// Mapper aux tags originaux avec |
|
|
Object.keys(parsedItems).forEach(cleanTag => {
|
|
const content = parsedItems[cleanTag];
|
|
results[`|${cleanTag}|`] = content;
|
|
});
|
|
|
|
// Vérification de cohérence paires
|
|
let pairsCompletes = 0;
|
|
for (const pair of faqPairs) {
|
|
const hasQuestion = results[pair.question.tag];
|
|
const hasAnswer = results[pair.answer.tag];
|
|
|
|
if (hasQuestion && hasAnswer) {
|
|
pairsCompletes++;
|
|
await logSh(`✓ Paire FAQ ${pair.number} complète: Q+R`, 'DEBUG');
|
|
} else {
|
|
await logSh(`⚠ Paire FAQ ${pair.number} incomplète: Q=${!!hasQuestion} R=${!!hasAnswer}`, 'WARNING');
|
|
}
|
|
}
|
|
|
|
await logSh(`${pairsCompletes}/${faqPairs.length} paires FAQ complètes`, 'INFO');
|
|
|
|
// FATAL si paires FAQ manquantes
|
|
if (pairsCompletes < faqPairs.length) {
|
|
const manquantes = faqPairs.length - pairsCompletes;
|
|
await logSh(`❌ FATAL: ${manquantes} paires FAQ manquantes sur ${faqPairs.length}`, 'ERROR');
|
|
throw new Error(`FATAL: Génération FAQ incomplète (${manquantes}/${faqPairs.length} manquantes) - arrêt du workflow`);
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
async function parseOtherElementsResponse(response, allOtherElements) {
|
|
const results = {};
|
|
|
|
await logSh('Parsing réponse autres éléments...', 'DEBUG');
|
|
|
|
const regex = /\[([^\]]+)\]\s*\n([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
const content = match[2].trim();
|
|
|
|
const cleanContent = content.replace(/^#+\s*/, '').replace(/<\/?[^>]+(>|$)/g, "");
|
|
|
|
results[`|${tag}|`] = cleanContent;
|
|
|
|
await logSh(`✓ Autre élément parsé [${tag}]: "${cleanContent}"`, 'DEBUG');
|
|
}
|
|
|
|
// Fallback si parsing partiel
|
|
if (Object.keys(results).length < allOtherElements.length) {
|
|
await logSh('Parsing autres éléments partiel, complétion fallback', 'WARNING');
|
|
|
|
const lines = response.split('\n')
|
|
.map(line => line.trim())
|
|
.filter(line => line.length > 0 && !line.startsWith('['));
|
|
|
|
allOtherElements.forEach((element, index) => {
|
|
if (!results[element.tag] && lines[index]) {
|
|
results[element.tag] = lines[index];
|
|
}
|
|
});
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS - ADAPTÉES =============
|
|
|
|
function createPromptForElement(element, csvData) {
|
|
// Cette fonction sera probablement définie dans content-generation.js
|
|
// Pour l'instant, retour basique
|
|
return `Génère du contenu pour ${element.type}: ${element.resolvedContent}`;
|
|
}
|
|
|
|
|
|
// 🔄 NODE.JS EXPORTS
|
|
module.exports = {
|
|
extractElements,
|
|
resolveVariablesContent,
|
|
getElementType,
|
|
generateAllContent,
|
|
parseElementStructure,
|
|
buildSmartHierarchy,
|
|
parseTitlesResponse,
|
|
parseTextsResponse,
|
|
parseFAQPairsResponse,
|
|
parseOtherElementsResponse,
|
|
createPromptForElement
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/MissingKeywords.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: MissingKeywords.js - Version Node.js
|
|
// Description: Génération automatique des mots-clés manquants
|
|
// ========================================
|
|
|
|
const { logSh } = require('./ErrorReporting');
|
|
const { callLLM } = require('./LLMManager');
|
|
|
|
/**
|
|
* Génère automatiquement les mots-clés manquants pour les éléments non définis
|
|
* @param {Array} elements - Liste des éléments extraits
|
|
* @param {Object} csvData - Données CSV avec personnalité
|
|
* @returns {Object} Éléments mis à jour avec nouveaux mots-clés
|
|
*/
|
|
async function generateMissingKeywords(elements, csvData) {
|
|
logSh('>>> GÉNÉRATION MOTS-CLÉS MANQUANTS <<<', 'INFO');
|
|
|
|
// 1. IDENTIFIER tous les éléments manquants
|
|
const missingElements = [];
|
|
elements.forEach(element => {
|
|
if (element.resolvedContent.includes('non défini') ||
|
|
element.resolvedContent.includes('non résolu') ||
|
|
element.resolvedContent.trim() === '') {
|
|
|
|
missingElements.push({
|
|
tag: element.originalTag,
|
|
name: element.name,
|
|
type: element.type,
|
|
currentContent: element.resolvedContent,
|
|
context: getElementContext(element, elements, csvData)
|
|
});
|
|
}
|
|
});
|
|
|
|
if (missingElements.length === 0) {
|
|
logSh('Aucun mot-clé manquant détecté', 'INFO');
|
|
return {};
|
|
}
|
|
|
|
logSh(`${missingElements.length} mots-clés manquants détectés`, 'INFO');
|
|
|
|
// 2. ANALYSER le contexte global disponible
|
|
const contextAnalysis = analyzeAvailableContext(elements, csvData);
|
|
|
|
// 3. GÉNÉRER tous les manquants en UN SEUL appel IA
|
|
const generatedKeywords = await callOpenAIForMissingKeywords(missingElements, contextAnalysis, csvData);
|
|
|
|
// 4. METTRE À JOUR les éléments avec les nouveaux mots-clés
|
|
const updatedElements = updateElementsWithKeywords(elements, generatedKeywords);
|
|
|
|
logSh(`Mots-clés manquants générés: ${Object.keys(generatedKeywords).length}`, 'INFO');
|
|
return updatedElements;
|
|
}
|
|
|
|
/**
|
|
* Analyser le contexte disponible pour guider la génération
|
|
* @param {Array} elements - Tous les éléments
|
|
* @param {Object} csvData - Données CSV
|
|
* @returns {Object} Analyse contextuelle
|
|
*/
|
|
function analyzeAvailableContext(elements, csvData) {
|
|
const availableKeywords = [];
|
|
const availableContent = [];
|
|
|
|
// Récupérer tous les mots-clés/contenu déjà disponibles
|
|
elements.forEach(element => {
|
|
if (element.resolvedContent &&
|
|
!element.resolvedContent.includes('non défini') &&
|
|
!element.resolvedContent.includes('non résolu') &&
|
|
element.resolvedContent.trim() !== '') {
|
|
|
|
if (element.type.includes('titre')) {
|
|
availableKeywords.push(element.resolvedContent);
|
|
} else {
|
|
availableContent.push(element.resolvedContent.substring(0, 100));
|
|
}
|
|
}
|
|
});
|
|
|
|
return {
|
|
mainKeyword: csvData.mc0,
|
|
mainTitle: csvData.t0,
|
|
availableKeywords: availableKeywords,
|
|
availableContent: availableContent,
|
|
theme: csvData.mc0, // Thème principal
|
|
businessContext: "Autocollant.fr - signalétique personnalisée, plaques"
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Obtenir le contexte spécifique d'un élément
|
|
* @param {Object} element - Élément à analyser
|
|
* @param {Array} allElements - Tous les éléments
|
|
* @param {Object} csvData - Données CSV
|
|
* @returns {Object} Contexte de l'élément
|
|
*/
|
|
function getElementContext(element, allElements, csvData) {
|
|
const context = {
|
|
elementType: element.type,
|
|
hierarchyLevel: element.name,
|
|
nearbyElements: []
|
|
};
|
|
|
|
// Trouver les éléments proches dans la hiérarchie
|
|
const elementParts = element.name.split('_');
|
|
if (elementParts.length >= 2) {
|
|
const baseLevel = elementParts.slice(0, 2).join('_'); // Ex: "Titre_H3"
|
|
|
|
allElements.forEach(otherElement => {
|
|
if (otherElement.name.startsWith(baseLevel) &&
|
|
otherElement.resolvedContent &&
|
|
!otherElement.resolvedContent.includes('non défini')) {
|
|
|
|
context.nearbyElements.push(otherElement.resolvedContent);
|
|
}
|
|
});
|
|
}
|
|
|
|
return context;
|
|
}
|
|
|
|
/**
|
|
* Appel IA pour générer tous les mots-clés manquants en un seul batch
|
|
* @param {Array} missingElements - Éléments manquants
|
|
* @param {Object} contextAnalysis - Analyse contextuelle
|
|
* @param {Object} csvData - Données CSV avec personnalité
|
|
* @returns {Object} Mots-clés générés
|
|
*/
|
|
async function callOpenAIForMissingKeywords(missingElements, contextAnalysis, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `Tu es ${personality.nom} (${personality.description}). Style: ${personality.style}
|
|
|
|
MISSION: GÉNÈRE ${missingElements.length} MOTS-CLÉS/EXPRESSIONS MANQUANTS pour ${contextAnalysis.mainKeyword}
|
|
|
|
CONTEXTE:
|
|
- Sujet: ${contextAnalysis.mainKeyword}
|
|
- Entreprise: Autocollant.fr (signalétique)
|
|
- Mots-clés existants: ${contextAnalysis.availableKeywords.slice(0, 3).join(', ')}
|
|
|
|
ÉLÉMENTS MANQUANTS:
|
|
`;
|
|
|
|
missingElements.forEach((missing, index) => {
|
|
prompt += `${index + 1}. [${missing.name}] → Mot-clé SEO\n`;
|
|
});
|
|
|
|
prompt += `\nCONSIGNES:
|
|
- Thème: ${contextAnalysis.mainKeyword}
|
|
- Mots-clés SEO naturels
|
|
- Varie les termes
|
|
- Évite répétitions
|
|
|
|
FORMAT:
|
|
[${missingElements[0].name}]
|
|
mot-clé
|
|
|
|
[${missingElements[1] ? missingElements[1].name : 'exemple'}]
|
|
mot-clé
|
|
|
|
etc...`;
|
|
|
|
try {
|
|
logSh('Génération mots-clés manquants...', 'DEBUG');
|
|
|
|
// Utilisation du LLM Manager avec fallback
|
|
const response = await callLLM('openai', prompt, {
|
|
temperature: 0.7,
|
|
maxTokens: 2000
|
|
}, personality);
|
|
|
|
// Parser la réponse
|
|
const generatedKeywords = parseMissingKeywordsResponse(response, missingElements);
|
|
|
|
return generatedKeywords;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ FATAL: Génération mots-clés manquants échouée: ${error}`, 'ERROR');
|
|
throw new Error(`FATAL: Génération mots-clés LLM impossible - arrêt du workflow: ${error}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Parser la réponse IA pour extraire les mots-clés générés
|
|
* @param {string} response - Réponse de l'IA
|
|
* @param {Array} missingElements - Éléments manquants
|
|
* @returns {Object} Mots-clés parsés
|
|
*/
|
|
function parseMissingKeywordsResponse(response, missingElements) {
|
|
const results = {};
|
|
|
|
const regex = /\[([^\]]+)\]\s*\n([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const elementName = match[1].trim();
|
|
const generatedKeyword = match[2].trim();
|
|
|
|
results[elementName] = generatedKeyword;
|
|
|
|
logSh(`✓ Mot-clé généré [${elementName}]: "${generatedKeyword}"`, 'DEBUG');
|
|
}
|
|
|
|
// FATAL si parsing partiel
|
|
if (Object.keys(results).length < missingElements.length) {
|
|
const manquants = missingElements.length - Object.keys(results).length;
|
|
logSh(`❌ FATAL: Parsing mots-clés partiel - ${manquants}/${missingElements.length} manquants`, 'ERROR');
|
|
throw new Error(`FATAL: Parsing mots-clés incomplet (${manquants}/${missingElements.length} manquants) - arrêt du workflow`);
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Mettre à jour les éléments avec les nouveaux mots-clés générés
|
|
* @param {Array} elements - Éléments originaux
|
|
* @param {Object} generatedKeywords - Nouveaux mots-clés
|
|
* @returns {Array} Éléments mis à jour
|
|
*/
|
|
function updateElementsWithKeywords(elements, generatedKeywords) {
|
|
const updatedElements = elements.map(element => {
|
|
const newKeyword = generatedKeywords[element.name];
|
|
|
|
if (newKeyword) {
|
|
return {
|
|
...element,
|
|
resolvedContent: newKeyword
|
|
};
|
|
}
|
|
|
|
return element;
|
|
});
|
|
|
|
logSh('Éléments mis à jour avec nouveaux mots-clés', 'INFO');
|
|
return updatedElements;
|
|
}
|
|
|
|
// Exports CommonJS
|
|
module.exports = {
|
|
generateMissingKeywords
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/trace.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// lib/trace.js
|
|
const { AsyncLocalStorage } = require('node:async_hooks');
|
|
const { randomUUID } = require('node:crypto');
|
|
const { logSh } = require('./ErrorReporting');
|
|
|
|
const als = new AsyncLocalStorage();
|
|
|
|
function now() { return performance.now(); }
|
|
function dur(ms) {
|
|
if (ms < 1e3) return `${ms.toFixed(1)}ms`;
|
|
const s = ms / 1e3;
|
|
return s < 60 ? `${s.toFixed(2)}s` : `${(s/60).toFixed(2)}m`;
|
|
}
|
|
|
|
class Span {
|
|
constructor({ name, parent = null, attrs = {} }) {
|
|
this.id = randomUUID();
|
|
this.name = name;
|
|
this.parent = parent;
|
|
this.children = [];
|
|
this.attrs = attrs;
|
|
this.start = now();
|
|
this.end = null;
|
|
this.status = 'ok';
|
|
this.error = null;
|
|
}
|
|
pathNames() {
|
|
const names = [];
|
|
let cur = this;
|
|
while (cur) { names.unshift(cur.name); cur = cur.parent; }
|
|
return names.join(' > ');
|
|
}
|
|
finish() { this.end = now(); }
|
|
duration() { return (this.end ?? now()) - this.start; }
|
|
}
|
|
|
|
class Tracer {
|
|
constructor() {
|
|
this.rootSpans = [];
|
|
}
|
|
current() { return als.getStore(); }
|
|
|
|
async startSpan(name, attrs = {}) {
|
|
const parent = this.current();
|
|
const span = new Span({ name, parent, attrs });
|
|
if (parent) parent.children.push(span);
|
|
else this.rootSpans.push(span);
|
|
|
|
// Formater les paramètres pour affichage
|
|
const paramsStr = this.formatParams(attrs);
|
|
await logSh(`▶ ${name}${paramsStr}`, 'TRACE');
|
|
return span;
|
|
}
|
|
|
|
async run(name, fn, attrs = {}) {
|
|
const parent = this.current();
|
|
const span = await this.startSpan(name, attrs);
|
|
return await als.run(span, async () => {
|
|
try {
|
|
const res = await fn();
|
|
span.finish();
|
|
const paramsStr = this.formatParams(span.attrs);
|
|
await logSh(`✔ ${name}${paramsStr} (${dur(span.duration())})`, 'TRACE');
|
|
return res;
|
|
} catch (err) {
|
|
span.status = 'error';
|
|
span.error = { message: err?.message, stack: err?.stack };
|
|
span.finish();
|
|
const paramsStr = this.formatParams(span.attrs);
|
|
await logSh(`✖ ${name}${paramsStr} FAILED (${dur(span.duration())})`, 'ERROR');
|
|
await logSh(`Stack trace: ${span.error.message}`, 'ERROR');
|
|
if (span.error.stack) {
|
|
const stackLines = span.error.stack.split('\n').slice(1, 6); // Première 5 lignes du stack
|
|
for (const line of stackLines) {
|
|
await logSh(` ${line.trim()}`, 'ERROR');
|
|
}
|
|
}
|
|
throw err;
|
|
}
|
|
});
|
|
}
|
|
|
|
async event(msg, extra = {}) {
|
|
const span = this.current();
|
|
const data = { trace: true, evt: 'span.event', ...extra };
|
|
if (span) {
|
|
data.span = span.id;
|
|
data.path = span.pathNames();
|
|
data.since_ms = +( (now() - span.start).toFixed(1) );
|
|
}
|
|
await logSh(`• ${msg}`, 'TRACE');
|
|
}
|
|
|
|
async annotate(fields = {}) {
|
|
const span = this.current();
|
|
if (span) Object.assign(span.attrs, fields);
|
|
await logSh('… annotate', 'TRACE');
|
|
}
|
|
|
|
formatParams(attrs = {}) {
|
|
const params = Object.entries(attrs)
|
|
.filter(([key, value]) => value !== undefined && value !== null)
|
|
.map(([key, value]) => {
|
|
// Tronquer les valeurs trop longues
|
|
const strValue = String(value);
|
|
const truncated = strValue.length > 50 ? strValue.substring(0, 47) + '...' : strValue;
|
|
return `${key}=${truncated}`;
|
|
});
|
|
|
|
return params.length > 0 ? `(${params.join(', ')})` : '';
|
|
}
|
|
|
|
printSummary() {
|
|
const lines = [];
|
|
const draw = (node, depth = 0) => {
|
|
const pad = ' '.repeat(depth);
|
|
const icon = node.status === 'error' ? '✖' : '✔';
|
|
lines.push(`${pad}${icon} ${node.name} (${dur(node.duration())})`);
|
|
if (Object.keys(node.attrs ?? {}).length) {
|
|
lines.push(`${pad} attrs: ${JSON.stringify(node.attrs)}`);
|
|
}
|
|
for (const ch of node.children) draw(ch, depth + 1);
|
|
if (node.status === 'error' && node.error?.message) {
|
|
lines.push(`${pad} error: ${node.error.message}`);
|
|
if (node.error.stack) {
|
|
const stackLines = String(node.error.stack || '').split('\n').slice(1, 4).map(s => s.trim());
|
|
if (stackLines.length) {
|
|
lines.push(`${pad} stack:`);
|
|
stackLines.forEach(line => {
|
|
if (line) lines.push(`${pad} ${line}`);
|
|
});
|
|
}
|
|
}
|
|
}
|
|
};
|
|
for (const r of this.rootSpans) draw(r, 0);
|
|
const summary = lines.join('\n');
|
|
logSh(`\n—— TRACE SUMMARY ——\n${summary}\n—— END TRACE ——`, 'INFO');
|
|
return summary;
|
|
}
|
|
}
|
|
|
|
const tracer = new Tracer();
|
|
|
|
module.exports = {
|
|
Span,
|
|
Tracer,
|
|
tracer
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/generation/InitialGeneration.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ÉTAPE 1: GÉNÉRATION INITIALE
|
|
// Responsabilité: Créer le contenu de base avec Claude uniquement
|
|
// LLM: Claude Sonnet (température 0.7)
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - GÉNÉRATION INITIALE
|
|
* Input: { content: {}, csvData: {}, context: {} }
|
|
* Output: { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function generateInitialContent(input) {
|
|
return await tracer.run('InitialGeneration.generateInitialContent()', async () => {
|
|
const { hierarchy, csvData, context = {} } = input;
|
|
|
|
await tracer.annotate({
|
|
step: '1/4',
|
|
llmProvider: 'claude',
|
|
elementsCount: Object.keys(hierarchy).length,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🚀 ÉTAPE 1/4: Génération initiale (Claude)`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(hierarchy).length} éléments à générer`, 'INFO');
|
|
|
|
try {
|
|
// Collecter tous les éléments dans l'ordre XML
|
|
const allElements = collectElementsInXMLOrder(hierarchy);
|
|
|
|
// Séparer FAQ pairs et autres éléments
|
|
const { faqPairs, otherElements } = separateElementTypes(allElements);
|
|
|
|
// Générer en chunks pour éviter timeouts
|
|
const results = {};
|
|
|
|
// 1. Générer éléments normaux (titres, textes, intro)
|
|
if (otherElements.length > 0) {
|
|
const normalResults = await generateNormalElements(otherElements, csvData);
|
|
Object.assign(results, normalResults);
|
|
}
|
|
|
|
// 2. Générer paires FAQ si présentes
|
|
if (faqPairs.length > 0) {
|
|
const faqResults = await generateFAQPairs(faqPairs, csvData);
|
|
Object.assign(results, faqResults);
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: Object.keys(results).length,
|
|
generated: Object.keys(results).length,
|
|
faqPairs: faqPairs.length,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ ÉTAPE 1/4 TERMINÉE: ${stats.generated} éléments générés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event(`Génération initiale terminée`, stats);
|
|
|
|
return {
|
|
content: results,
|
|
stats,
|
|
debug: {
|
|
llmProvider: 'claude',
|
|
step: 1,
|
|
elementsGenerated: Object.keys(results)
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ÉTAPE 1/4 ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`InitialGeneration failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Générer éléments normaux (titres, textes, intro) en chunks
|
|
*/
|
|
async function generateNormalElements(elements, csvData) {
|
|
logSh(`📝 Génération éléments normaux: ${elements.length} éléments`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(elements, 4); // Chunks de 4 pour éviter timeouts
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
logSh(` 📦 Chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
try {
|
|
const prompt = createBatchPrompt(chunk, csvData);
|
|
|
|
const response = await callLLM('claude', prompt, {
|
|
temperature: 0.7,
|
|
maxTokens: 2000 * chunk.length
|
|
}, csvData.personality);
|
|
|
|
const chunkResults = parseBatchResponse(response, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk ${chunkIndex + 1}: ${Object.keys(chunkResults).length} éléments générés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Générer paires FAQ cohérentes
|
|
*/
|
|
async function generateFAQPairs(faqPairs, csvData) {
|
|
logSh(`❓ Génération paires FAQ: ${faqPairs.length} paires`, 'DEBUG');
|
|
|
|
const prompt = createFAQPairsPrompt(faqPairs, csvData);
|
|
|
|
const response = await callLLM('claude', prompt, {
|
|
temperature: 0.8,
|
|
maxTokens: 3000
|
|
}, csvData.personality);
|
|
|
|
return parseFAQResponse(response, faqPairs);
|
|
}
|
|
|
|
/**
|
|
* Créer prompt batch pour éléments normaux
|
|
*/
|
|
function createBatchPrompt(elements, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `=== GÉNÉRATION CONTENU INITIAL ===
|
|
Entreprise: Autocollant.fr - signalétique personnalisée
|
|
Sujet: ${csvData.mc0}
|
|
Rédacteur: ${personality.nom} (${personality.style})
|
|
|
|
ÉLÉMENTS À GÉNÉRER:
|
|
|
|
`;
|
|
|
|
elements.forEach((elementInfo, index) => {
|
|
const cleanTag = elementInfo.tag.replace(/\|/g, '');
|
|
prompt += `${index + 1}. [${cleanTag}] - ${getElementDescription(elementInfo)}\n`;
|
|
});
|
|
|
|
prompt += `
|
|
STYLE ${personality.nom.toUpperCase()}:
|
|
- Vocabulaire: ${personality.vocabulairePref}
|
|
- Phrases: ${personality.longueurPhrases}
|
|
- Niveau: ${personality.niveauTechnique}
|
|
|
|
CONSIGNES:
|
|
- Contenu SEO optimisé pour ${csvData.mc0}
|
|
- Style ${personality.style} naturel
|
|
- Pas de références techniques dans contenu
|
|
- RÉPONSE DIRECTE par le contenu
|
|
|
|
FORMAT:
|
|
[${elements[0].tag.replace(/\|/g, '')}]
|
|
Contenu généré...
|
|
|
|
[${elements[1] ? elements[1].tag.replace(/\|/g, '') : 'element2'}]
|
|
Contenu généré...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse batch
|
|
*/
|
|
function parseBatchResponse(response, elements) {
|
|
const results = {};
|
|
const regex = /\[([^\]]+)\]\s*([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
const content = cleanGeneratedContent(match[2].trim());
|
|
parsedItems[tag] = content;
|
|
}
|
|
|
|
// Mapper aux vrais tags
|
|
elements.forEach(element => {
|
|
const cleanTag = element.tag.replace(/\|/g, '');
|
|
if (parsedItems[cleanTag] && parsedItems[cleanTag].length > 10) {
|
|
results[element.tag] = parsedItems[cleanTag];
|
|
} else {
|
|
results[element.tag] = `Contenu professionnel pour ${element.element.name || cleanTag}`;
|
|
logSh(`⚠️ Fallback pour [${cleanTag}]`, 'WARNING');
|
|
}
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt pour paires FAQ
|
|
*/
|
|
function createFAQPairsPrompt(faqPairs, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `=== GÉNÉRATION PAIRES FAQ ===
|
|
Sujet: ${csvData.mc0}
|
|
Rédacteur: ${personality.nom} (${personality.style})
|
|
|
|
PAIRES À GÉNÉRER:
|
|
`;
|
|
|
|
faqPairs.forEach((pair, index) => {
|
|
const qTag = pair.question.tag.replace(/\|/g, '');
|
|
const aTag = pair.answer.tag.replace(/\|/g, '');
|
|
prompt += `${index + 1}. [${qTag}] + [${aTag}]\n`;
|
|
});
|
|
|
|
prompt += `
|
|
CONSIGNES:
|
|
- Questions naturelles de clients
|
|
- Réponses expertes ${personality.style}
|
|
- Couvrir: prix, livraison, personnalisation
|
|
|
|
FORMAT:
|
|
[${faqPairs[0].question.tag.replace(/\|/g, '')}]
|
|
Question client naturelle ?
|
|
|
|
[${faqPairs[0].answer.tag.replace(/\|/g, '')}]
|
|
Réponse utile et rassurante.`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse FAQ
|
|
*/
|
|
function parseFAQResponse(response, faqPairs) {
|
|
const results = {};
|
|
const regex = /\[([^\]]+)\]\s*([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
const content = cleanGeneratedContent(match[2].trim());
|
|
parsedItems[tag] = content;
|
|
}
|
|
|
|
// Mapper aux paires FAQ
|
|
faqPairs.forEach(pair => {
|
|
const qCleanTag = pair.question.tag.replace(/\|/g, '');
|
|
const aCleanTag = pair.answer.tag.replace(/\|/g, '');
|
|
|
|
if (parsedItems[qCleanTag]) results[pair.question.tag] = parsedItems[qCleanTag];
|
|
if (parsedItems[aCleanTag]) results[pair.answer.tag] = parsedItems[aCleanTag];
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
function collectElementsInXMLOrder(hierarchy) {
|
|
const allElements = [];
|
|
|
|
Object.keys(hierarchy).forEach(path => {
|
|
const section = hierarchy[path];
|
|
|
|
if (section.title) {
|
|
allElements.push({
|
|
tag: section.title.originalElement.originalTag,
|
|
element: section.title.originalElement,
|
|
type: section.title.originalElement.type
|
|
});
|
|
}
|
|
|
|
if (section.text) {
|
|
allElements.push({
|
|
tag: section.text.originalElement.originalTag,
|
|
element: section.text.originalElement,
|
|
type: section.text.originalElement.type
|
|
});
|
|
}
|
|
|
|
section.questions.forEach(q => {
|
|
allElements.push({
|
|
tag: q.originalElement.originalTag,
|
|
element: q.originalElement,
|
|
type: q.originalElement.type
|
|
});
|
|
});
|
|
});
|
|
|
|
return allElements;
|
|
}
|
|
|
|
function separateElementTypes(allElements) {
|
|
const faqPairs = [];
|
|
const otherElements = [];
|
|
const faqQuestions = {};
|
|
const faqAnswers = {};
|
|
|
|
// Collecter FAQ questions et answers
|
|
allElements.forEach(element => {
|
|
if (element.type === 'faq_question') {
|
|
const numberMatch = element.tag.match(/(\d+)/);
|
|
const faqNumber = numberMatch ? numberMatch[1] : '1';
|
|
faqQuestions[faqNumber] = element;
|
|
} else if (element.type === 'faq_reponse') {
|
|
const numberMatch = element.tag.match(/(\d+)/);
|
|
const faqNumber = numberMatch ? numberMatch[1] : '1';
|
|
faqAnswers[faqNumber] = element;
|
|
} else {
|
|
otherElements.push(element);
|
|
}
|
|
});
|
|
|
|
// Créer paires FAQ
|
|
Object.keys(faqQuestions).forEach(number => {
|
|
const question = faqQuestions[number];
|
|
const answer = faqAnswers[number];
|
|
|
|
if (question && answer) {
|
|
faqPairs.push({ number, question, answer });
|
|
} else if (question) {
|
|
otherElements.push(question);
|
|
} else if (answer) {
|
|
otherElements.push(answer);
|
|
}
|
|
});
|
|
|
|
return { faqPairs, otherElements };
|
|
}
|
|
|
|
function getElementDescription(elementInfo) {
|
|
switch (elementInfo.type) {
|
|
case 'titre_h1': return 'Titre principal accrocheur';
|
|
case 'titre_h2': return 'Titre de section';
|
|
case 'titre_h3': return 'Sous-titre';
|
|
case 'intro': return 'Introduction engageante';
|
|
case 'texte': return 'Paragraphe informatif';
|
|
default: return 'Contenu pertinent';
|
|
}
|
|
}
|
|
|
|
function cleanGeneratedContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?Titre_[HU]\d+_\d+[.,\s]*/gi, '');
|
|
content = content.replace(/\*\*[^*]+\*\*/g, '');
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
module.exports = {
|
|
generateInitialContent, // ← MAIN ENTRY POINT
|
|
generateNormalElements,
|
|
generateFAQPairs,
|
|
createBatchPrompt,
|
|
parseBatchResponse,
|
|
collectElementsInXMLOrder,
|
|
separateElementTypes
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/generation/TechnicalEnhancement.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ÉTAPE 2: ENHANCEMENT TECHNIQUE
|
|
// Responsabilité: Améliorer la précision technique avec GPT-4
|
|
// LLM: GPT-4o-mini (température 0.4)
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - ENHANCEMENT TECHNIQUE
|
|
* Input: { content: {}, csvData: {}, context: {} }
|
|
* Output: { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function enhanceTechnicalTerms(input) {
|
|
return await tracer.run('TechnicalEnhancement.enhanceTechnicalTerms()', async () => {
|
|
const { content, csvData, context = {} } = input;
|
|
|
|
await tracer.annotate({
|
|
step: '2/4',
|
|
llmProvider: 'gpt4',
|
|
elementsCount: Object.keys(content).length,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🔧 ÉTAPE 2/4: Enhancement technique (GPT-4)`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à analyser`, 'INFO');
|
|
|
|
try {
|
|
// 1. Analyser tous les éléments pour détecter termes techniques
|
|
const technicalAnalysis = await analyzeTechnicalTerms(content, csvData);
|
|
|
|
// 2. Filter les éléments qui ont besoin d'enhancement
|
|
const elementsNeedingEnhancement = technicalAnalysis.filter(item => item.needsEnhancement);
|
|
|
|
logSh(` 📋 Analyse: ${elementsNeedingEnhancement.length}/${Object.keys(content).length} éléments nécessitent enhancement`, 'INFO');
|
|
|
|
if (elementsNeedingEnhancement.length === 0) {
|
|
logSh(`✅ ÉTAPE 2/4: Aucun enhancement nécessaire`, 'INFO');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration: Date.now() - startTime },
|
|
debug: { llmProvider: 'gpt4', step: 2, enhancementsApplied: [] }
|
|
};
|
|
}
|
|
|
|
// 3. Améliorer les éléments sélectionnés
|
|
const enhancedResults = await enhanceSelectedElements(elementsNeedingEnhancement, csvData);
|
|
|
|
// 4. Merger avec contenu original
|
|
const finalContent = { ...content };
|
|
let actuallyEnhanced = 0;
|
|
|
|
Object.keys(enhancedResults).forEach(tag => {
|
|
if (enhancedResults[tag] !== content[tag]) {
|
|
finalContent[tag] = enhancedResults[tag];
|
|
actuallyEnhanced++;
|
|
}
|
|
});
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: Object.keys(content).length,
|
|
enhanced: actuallyEnhanced,
|
|
candidate: elementsNeedingEnhancement.length,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ ÉTAPE 2/4 TERMINÉE: ${stats.enhanced} éléments améliorés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event(`Enhancement technique terminé`, stats);
|
|
|
|
return {
|
|
content: finalContent,
|
|
stats,
|
|
debug: {
|
|
llmProvider: 'gpt4',
|
|
step: 2,
|
|
enhancementsApplied: Object.keys(enhancedResults),
|
|
technicalTermsFound: elementsNeedingEnhancement.map(e => e.technicalTerms)
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ÉTAPE 2/4 ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`TechnicalEnhancement failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Analyser tous les éléments pour détecter termes techniques
|
|
*/
|
|
async function analyzeTechnicalTerms(content, csvData) {
|
|
logSh(`🔍 Analyse termes techniques batch`, 'DEBUG');
|
|
|
|
const contentEntries = Object.keys(content);
|
|
|
|
const analysisPrompt = `MISSION: Analyser ces ${contentEntries.length} contenus et identifier leurs termes techniques.
|
|
|
|
CONTEXTE: ${csvData.mc0} - Secteur: signalétique/impression
|
|
|
|
CONTENUS À ANALYSER:
|
|
|
|
${contentEntries.map((tag, i) => `[${i + 1}] TAG: ${tag}
|
|
CONTENU: "${content[tag]}"`).join('\n\n')}
|
|
|
|
CONSIGNES:
|
|
- Identifie UNIQUEMENT les vrais termes techniques métier/industrie
|
|
- Évite mots génériques (qualité, service, pratique, personnalisé)
|
|
- Focus: matériaux, procédés, normes, dimensions, technologies
|
|
- Si aucun terme technique → "AUCUN"
|
|
|
|
EXEMPLES VALIDES: dibond, impression UV, fraisage CNC, épaisseur 3mm
|
|
EXEMPLES INVALIDES: durable, pratique, personnalisé, moderne
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] dibond, impression UV OU AUCUN
|
|
[2] AUCUN
|
|
[3] aluminium, fraisage CNC OU AUCUN
|
|
etc...`;
|
|
|
|
try {
|
|
const analysisResponse = await callLLM('gpt4', analysisPrompt, {
|
|
temperature: 0.3,
|
|
maxTokens: 2000
|
|
}, csvData.personality);
|
|
|
|
return parseAnalysisResponse(analysisResponse, content, contentEntries);
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Analyse termes techniques échouée: ${error.message}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Améliorer les éléments sélectionnés
|
|
*/
|
|
async function enhanceSelectedElements(elementsNeedingEnhancement, csvData) {
|
|
logSh(`🛠️ Enhancement ${elementsNeedingEnhancement.length} éléments`, 'DEBUG');
|
|
|
|
const enhancementPrompt = `MISSION: Améliore UNIQUEMENT la précision technique de ces contenus.
|
|
|
|
CONTEXTE: ${csvData.mc0} - Secteur signalétique/impression
|
|
PERSONNALITÉ: ${csvData.personality?.nom} (${csvData.personality?.style})
|
|
|
|
CONTENUS À AMÉLIORER:
|
|
|
|
${elementsNeedingEnhancement.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
CONTENU: "${item.content}"
|
|
TERMES TECHNIQUES: ${item.technicalTerms.join(', ')}`).join('\n\n')}
|
|
|
|
CONSIGNES:
|
|
- GARDE même longueur, structure et ton ${csvData.personality?.style}
|
|
- Intègre naturellement les termes techniques listés
|
|
- NE CHANGE PAS le fond du message
|
|
- Vocabulaire expert mais accessible
|
|
- Termes secteur: dibond, aluminium, impression UV, fraisage, PMMA
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu avec amélioration technique
|
|
[2] Contenu avec amélioration technique
|
|
etc...`;
|
|
|
|
try {
|
|
const enhancedResponse = await callLLM('gpt4', enhancementPrompt, {
|
|
temperature: 0.4,
|
|
maxTokens: 5000
|
|
}, csvData.personality);
|
|
|
|
return parseEnhancementResponse(enhancedResponse, elementsNeedingEnhancement);
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Enhancement éléments échoué: ${error.message}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Parser réponse analyse
|
|
*/
|
|
function parseAnalysisResponse(response, content, contentEntries) {
|
|
const results = [];
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\[\d+\]|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const index = parseInt(match[1]) - 1;
|
|
const termsText = match[2].trim();
|
|
parsedItems[index] = termsText;
|
|
}
|
|
|
|
contentEntries.forEach((tag, index) => {
|
|
const termsText = parsedItems[index] || 'AUCUN';
|
|
const hasTerms = !termsText.toUpperCase().includes('AUCUN');
|
|
|
|
const technicalTerms = hasTerms ?
|
|
termsText.split(',').map(t => t.trim()).filter(t => t.length > 0) :
|
|
[];
|
|
|
|
results.push({
|
|
tag,
|
|
content: content[tag],
|
|
technicalTerms,
|
|
needsEnhancement: hasTerms && technicalTerms.length > 0
|
|
});
|
|
|
|
logSh(`🔍 [${tag}]: ${hasTerms ? technicalTerms.join(', ') : 'aucun terme technique'}`, 'DEBUG');
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse enhancement
|
|
*/
|
|
function parseEnhancementResponse(response, elementsNeedingEnhancement) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < elementsNeedingEnhancement.length) {
|
|
let enhancedContent = match[2].trim();
|
|
const element = elementsNeedingEnhancement[index];
|
|
|
|
// Nettoyer le contenu généré
|
|
enhancedContent = cleanEnhancedContent(enhancedContent);
|
|
|
|
if (enhancedContent && enhancedContent.length > 10) {
|
|
results[element.tag] = enhancedContent;
|
|
logSh(`✅ Enhanced [${element.tag}]: "${enhancedContent.substring(0, 100)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content;
|
|
logSh(`⚠️ Fallback [${element.tag}]: contenu invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < elementsNeedingEnhancement.length) {
|
|
const element = elementsNeedingEnhancement[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Nettoyer contenu amélioré
|
|
*/
|
|
function cleanEnhancedContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?pour\s+/gi, '');
|
|
content = content.replace(/\*\*[^*]+\*\*/g, '');
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
module.exports = {
|
|
enhanceTechnicalTerms, // ← MAIN ENTRY POINT
|
|
analyzeTechnicalTerms,
|
|
enhanceSelectedElements,
|
|
parseAnalysisResponse,
|
|
parseEnhancementResponse
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/generation/TransitionEnhancement.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ÉTAPE 3: ENHANCEMENT TRANSITIONS
|
|
// Responsabilité: Améliorer la fluidité avec Gemini
|
|
// LLM: Gemini (température 0.6)
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - ENHANCEMENT TRANSITIONS
|
|
* Input: { content: {}, csvData: {}, context: {} }
|
|
* Output: { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function enhanceTransitions(input) {
|
|
return await tracer.run('TransitionEnhancement.enhanceTransitions()', async () => {
|
|
const { content, csvData, context = {} } = input;
|
|
|
|
await tracer.annotate({
|
|
step: '3/4',
|
|
llmProvider: 'gemini',
|
|
elementsCount: Object.keys(content).length,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🔗 ÉTAPE 3/4: Enhancement transitions (Gemini)`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à analyser`, 'INFO');
|
|
|
|
try {
|
|
// 1. Analyser quels éléments ont besoin d'amélioration transitions
|
|
const elementsNeedingTransitions = analyzeTransitionNeeds(content);
|
|
|
|
logSh(` 📋 Analyse: ${elementsNeedingTransitions.length}/${Object.keys(content).length} éléments nécessitent fluidité`, 'INFO');
|
|
|
|
if (elementsNeedingTransitions.length === 0) {
|
|
logSh(`✅ ÉTAPE 3/4: Transitions déjà optimales`, 'INFO');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration: Date.now() - startTime },
|
|
debug: { llmProvider: 'gemini', step: 3, enhancementsApplied: [] }
|
|
};
|
|
}
|
|
|
|
// 2. Améliorer en chunks pour Gemini
|
|
const improvedResults = await improveTransitionsInChunks(elementsNeedingTransitions, csvData);
|
|
|
|
// 3. Merger avec contenu original
|
|
const finalContent = { ...content };
|
|
let actuallyImproved = 0;
|
|
|
|
Object.keys(improvedResults).forEach(tag => {
|
|
if (improvedResults[tag] !== content[tag]) {
|
|
finalContent[tag] = improvedResults[tag];
|
|
actuallyImproved++;
|
|
}
|
|
});
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: Object.keys(content).length,
|
|
enhanced: actuallyImproved,
|
|
candidate: elementsNeedingTransitions.length,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ ÉTAPE 3/4 TERMINÉE: ${stats.enhanced} éléments fluidifiés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event(`Enhancement transitions terminé`, stats);
|
|
|
|
return {
|
|
content: finalContent,
|
|
stats,
|
|
debug: {
|
|
llmProvider: 'gemini',
|
|
step: 3,
|
|
enhancementsApplied: Object.keys(improvedResults),
|
|
transitionIssues: elementsNeedingTransitions.map(e => e.issues)
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ÉTAPE 3/4 ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: retourner contenu original si Gemini indisponible
|
|
logSh(`🔄 Fallback: contenu original conservé`, 'WARNING');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration },
|
|
debug: { llmProvider: 'gemini', step: 3, error: error.message, fallback: true }
|
|
};
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Analyser besoin d'amélioration transitions
|
|
*/
|
|
function analyzeTransitionNeeds(content) {
|
|
const elementsNeedingTransitions = [];
|
|
|
|
Object.keys(content).forEach(tag => {
|
|
const text = content[tag];
|
|
|
|
// Filtrer les éléments longs (>150 chars) qui peuvent bénéficier d'améliorations
|
|
if (text.length > 150) {
|
|
const needsTransitions = evaluateTransitionQuality(text);
|
|
|
|
if (needsTransitions.needsImprovement) {
|
|
elementsNeedingTransitions.push({
|
|
tag,
|
|
content: text,
|
|
issues: needsTransitions.issues,
|
|
score: needsTransitions.score
|
|
});
|
|
|
|
logSh(` 🔍 [${tag}]: Score=${needsTransitions.score.toFixed(2)}, Issues: ${needsTransitions.issues.join(', ')}`, 'DEBUG');
|
|
}
|
|
} else {
|
|
logSh(` ⏭️ [${tag}]: Trop court (${text.length}c), ignoré`, 'DEBUG');
|
|
}
|
|
});
|
|
|
|
// Trier par score (plus problématique en premier)
|
|
elementsNeedingTransitions.sort((a, b) => a.score - b.score);
|
|
|
|
return elementsNeedingTransitions;
|
|
}
|
|
|
|
/**
|
|
* Évaluer qualité transitions d'un texte
|
|
*/
|
|
function evaluateTransitionQuality(text) {
|
|
const sentences = text.split(/[.!?]+/).filter(s => s.trim().length > 10);
|
|
|
|
if (sentences.length < 2) {
|
|
return { needsImprovement: false, score: 1.0, issues: [] };
|
|
}
|
|
|
|
const issues = [];
|
|
let score = 1.0; // Score parfait = 1.0, problématique = 0.0
|
|
|
|
// Analyse 1: Connecteurs répétitifs
|
|
const repetitiveConnectors = analyzeRepetitiveConnectors(text);
|
|
if (repetitiveConnectors > 0.3) {
|
|
issues.push('connecteurs_répétitifs');
|
|
score -= 0.3;
|
|
}
|
|
|
|
// Analyse 2: Transitions abruptes
|
|
const abruptTransitions = analyzeAbruptTransitions(sentences);
|
|
if (abruptTransitions > 0.4) {
|
|
issues.push('transitions_abruptes');
|
|
score -= 0.4;
|
|
}
|
|
|
|
// Analyse 3: Manque de variété dans longueurs
|
|
const sentenceVariety = analyzeSentenceVariety(sentences);
|
|
if (sentenceVariety < 0.3) {
|
|
issues.push('phrases_uniformes');
|
|
score -= 0.2;
|
|
}
|
|
|
|
// Analyse 4: Trop formel ou trop familier
|
|
const formalityIssues = analyzeFormalityBalance(text);
|
|
if (formalityIssues > 0.5) {
|
|
issues.push('formalité_déséquilibrée');
|
|
score -= 0.1;
|
|
}
|
|
|
|
return {
|
|
needsImprovement: score < 0.6,
|
|
score: Math.max(0, score),
|
|
issues
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Améliorer transitions en chunks
|
|
*/
|
|
async function improveTransitionsInChunks(elementsNeedingTransitions, csvData) {
|
|
logSh(`🔄 Amélioration transitions: ${elementsNeedingTransitions.length} éléments`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(elementsNeedingTransitions, 6); // Chunks plus petits pour Gemini
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(` 📦 Chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
const improvementPrompt = createTransitionImprovementPrompt(chunk, csvData);
|
|
|
|
const improvedResponse = await callLLM('gemini', improvementPrompt, {
|
|
temperature: 0.6,
|
|
maxTokens: 2500
|
|
}, csvData.personality);
|
|
|
|
const chunkResults = parseTransitionResponse(improvedResponse, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk ${chunkIndex + 1}: ${Object.keys(chunkResults).length} améliorés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: garder contenu original pour ce chunk
|
|
chunk.forEach(element => {
|
|
results[element.tag] = element.content;
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt amélioration transitions
|
|
*/
|
|
function createTransitionImprovementPrompt(chunk, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `MISSION: Améliore UNIQUEMENT les transitions et fluidité de ces contenus.
|
|
|
|
CONTEXTE: Article SEO ${csvData.mc0}
|
|
PERSONNALITÉ: ${personality?.nom} (${personality?.style} web professionnel)
|
|
CONNECTEURS PRÉFÉRÉS: ${personality?.connecteursPref}
|
|
|
|
CONTENUS À FLUIDIFIER:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
PROBLÈMES: ${item.issues.join(', ')}
|
|
CONTENU: "${item.content}"`).join('\n\n')}
|
|
|
|
OBJECTIFS:
|
|
- Connecteurs plus naturels et variés: ${personality?.connecteursPref}
|
|
- Transitions fluides entre idées
|
|
- ÉVITE répétitions excessives ("du coup", "franchement", "par ailleurs")
|
|
- Style ${personality?.style} mais professionnel web
|
|
|
|
CONSIGNES STRICTES:
|
|
- NE CHANGE PAS le fond du message
|
|
- GARDE même structure et longueur
|
|
- Améliore SEULEMENT la fluidité
|
|
- RESPECTE le style ${personality?.nom}
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu avec transitions améliorées
|
|
[2] Contenu avec transitions améliorées
|
|
etc...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse amélioration transitions
|
|
*/
|
|
function parseTransitionResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let improvedContent = match[2].trim();
|
|
const element = chunk[index];
|
|
|
|
// Nettoyer le contenu amélioré
|
|
improvedContent = cleanImprovedContent(improvedContent);
|
|
|
|
if (improvedContent && improvedContent.length > 10) {
|
|
results[element.tag] = improvedContent;
|
|
logSh(`✅ Improved [${element.tag}]: "${improvedContent.substring(0, 100)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content;
|
|
logSh(`⚠️ Fallback [${element.tag}]: amélioration invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < chunk.length) {
|
|
const element = chunk[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
function analyzeRepetitiveConnectors(content) {
|
|
const connectors = ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc'];
|
|
let totalConnectors = 0;
|
|
let repetitions = 0;
|
|
|
|
connectors.forEach(connector => {
|
|
const matches = (content.match(new RegExp(`\\b${connector}\\b`, 'gi')) || []);
|
|
totalConnectors += matches.length;
|
|
if (matches.length > 1) repetitions += matches.length - 1;
|
|
});
|
|
|
|
return totalConnectors > 0 ? repetitions / totalConnectors : 0;
|
|
}
|
|
|
|
function analyzeAbruptTransitions(sentences) {
|
|
if (sentences.length < 2) return 0;
|
|
|
|
let abruptCount = 0;
|
|
|
|
for (let i = 1; i < sentences.length; i++) {
|
|
const current = sentences[i].trim();
|
|
const hasConnector = hasTransitionWord(current);
|
|
|
|
if (!hasConnector && current.length > 30) {
|
|
abruptCount++;
|
|
}
|
|
}
|
|
|
|
return abruptCount / (sentences.length - 1);
|
|
}
|
|
|
|
function analyzeSentenceVariety(sentences) {
|
|
if (sentences.length < 2) return 1;
|
|
|
|
const lengths = sentences.map(s => s.trim().length);
|
|
const avgLength = lengths.reduce((a, b) => a + b, 0) / lengths.length;
|
|
const variance = lengths.reduce((acc, len) => acc + Math.pow(len - avgLength, 2), 0) / lengths.length;
|
|
const stdDev = Math.sqrt(variance);
|
|
|
|
return Math.min(1, stdDev / avgLength);
|
|
}
|
|
|
|
function analyzeFormalityBalance(content) {
|
|
const formalIndicators = ['il convient de', 'par conséquent', 'néanmoins', 'toutefois'];
|
|
const casualIndicators = ['du coup', 'bon', 'franchement', 'nickel'];
|
|
|
|
let formalCount = 0;
|
|
let casualCount = 0;
|
|
|
|
formalIndicators.forEach(indicator => {
|
|
if (content.toLowerCase().includes(indicator)) formalCount++;
|
|
});
|
|
|
|
casualIndicators.forEach(indicator => {
|
|
if (content.toLowerCase().includes(indicator)) casualCount++;
|
|
});
|
|
|
|
const total = formalCount + casualCount;
|
|
if (total === 0) return 0;
|
|
|
|
// Déséquilibre si trop d'un côté
|
|
const balance = Math.abs(formalCount - casualCount) / total;
|
|
return balance;
|
|
}
|
|
|
|
function hasTransitionWord(sentence) {
|
|
const connectors = ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc', 'ensuite', 'puis', 'également', 'aussi'];
|
|
return connectors.some(connector => sentence.toLowerCase().includes(connector));
|
|
}
|
|
|
|
function cleanImprovedContent(content) {
|
|
if (!content) return content;
|
|
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?/, '');
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
module.exports = {
|
|
enhanceTransitions, // ← MAIN ENTRY POINT
|
|
analyzeTransitionNeeds,
|
|
evaluateTransitionQuality,
|
|
improveTransitionsInChunks,
|
|
createTransitionImprovementPrompt,
|
|
parseTransitionResponse
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/generation/StyleEnhancement.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ÉTAPE 4: ENHANCEMENT STYLE PERSONNALITÉ
|
|
// Responsabilité: Appliquer le style personnalité avec Mistral
|
|
// LLM: Mistral (température 0.8)
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - ENHANCEMENT STYLE
|
|
* Input: { content: {}, csvData: {}, context: {} }
|
|
* Output: { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function applyPersonalityStyle(input) {
|
|
return await tracer.run('StyleEnhancement.applyPersonalityStyle()', async () => {
|
|
const { content, csvData, context = {} } = input;
|
|
|
|
await tracer.annotate({
|
|
step: '4/4',
|
|
llmProvider: 'mistral',
|
|
elementsCount: Object.keys(content).length,
|
|
personality: csvData.personality?.nom,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎭 ÉTAPE 4/4: Enhancement style ${csvData.personality?.nom} (Mistral)`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à styliser`, 'INFO');
|
|
|
|
try {
|
|
const personality = csvData.personality;
|
|
|
|
if (!personality) {
|
|
logSh(`⚠️ ÉTAPE 4/4: Aucune personnalité définie, style standard`, 'WARNING');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration: Date.now() - startTime },
|
|
debug: { llmProvider: 'mistral', step: 4, personalityApplied: 'none' }
|
|
};
|
|
}
|
|
|
|
// 1. Préparer éléments pour stylisation
|
|
const styleElements = prepareElementsForStyling(content);
|
|
|
|
// 2. Appliquer style en chunks
|
|
const styledResults = await applyStyleInChunks(styleElements, csvData);
|
|
|
|
// 3. Merger résultats
|
|
const finalContent = { ...content };
|
|
let actuallyStyled = 0;
|
|
|
|
Object.keys(styledResults).forEach(tag => {
|
|
if (styledResults[tag] !== content[tag]) {
|
|
finalContent[tag] = styledResults[tag];
|
|
actuallyStyled++;
|
|
}
|
|
});
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: Object.keys(content).length,
|
|
enhanced: actuallyStyled,
|
|
personality: personality.nom,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ ÉTAPE 4/4 TERMINÉE: ${stats.enhanced} éléments stylisés ${personality.nom} (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event(`Enhancement style terminé`, stats);
|
|
|
|
return {
|
|
content: finalContent,
|
|
stats,
|
|
debug: {
|
|
llmProvider: 'mistral',
|
|
step: 4,
|
|
personalityApplied: personality.nom,
|
|
styleCharacteristics: {
|
|
vocabulaire: personality.vocabulairePref,
|
|
connecteurs: personality.connecteursPref,
|
|
style: personality.style
|
|
}
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ÉTAPE 4/4 ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: retourner contenu original si Mistral indisponible
|
|
logSh(`🔄 Fallback: contenu original conservé`, 'WARNING');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration },
|
|
debug: { llmProvider: 'mistral', step: 4, error: error.message, fallback: true }
|
|
};
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Préparer éléments pour stylisation
|
|
*/
|
|
function prepareElementsForStyling(content) {
|
|
const styleElements = [];
|
|
|
|
Object.keys(content).forEach(tag => {
|
|
const text = content[tag];
|
|
|
|
// Tous les éléments peuvent bénéficier d'adaptation personnalité
|
|
// Même les courts (titres) peuvent être adaptés au style
|
|
styleElements.push({
|
|
tag,
|
|
content: text,
|
|
priority: calculateStylePriority(text, tag)
|
|
});
|
|
});
|
|
|
|
// Trier par priorité (titres d'abord, puis textes longs)
|
|
styleElements.sort((a, b) => b.priority - a.priority);
|
|
|
|
return styleElements;
|
|
}
|
|
|
|
/**
|
|
* Calculer priorité de stylisation
|
|
*/
|
|
function calculateStylePriority(text, tag) {
|
|
let priority = 1.0;
|
|
|
|
// Titres = haute priorité (plus visible)
|
|
if (tag.includes('Titre') || tag.includes('H1') || tag.includes('H2')) {
|
|
priority += 0.5;
|
|
}
|
|
|
|
// Textes longs = priorité selon longueur
|
|
if (text.length > 200) {
|
|
priority += 0.3;
|
|
} else if (text.length > 100) {
|
|
priority += 0.2;
|
|
}
|
|
|
|
// Introduction = haute priorité
|
|
if (tag.includes('intro') || tag.includes('Introduction')) {
|
|
priority += 0.4;
|
|
}
|
|
|
|
return priority;
|
|
}
|
|
|
|
/**
|
|
* Appliquer style en chunks
|
|
*/
|
|
async function applyStyleInChunks(styleElements, csvData) {
|
|
logSh(`🎨 Stylisation: ${styleElements.length} éléments selon ${csvData.personality.nom}`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(styleElements, 8); // Chunks de 8 pour Mistral
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(` 📦 Chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
const stylePrompt = createStylePrompt(chunk, csvData);
|
|
|
|
const styledResponse = await callLLM('mistral', stylePrompt, {
|
|
temperature: 0.8,
|
|
maxTokens: 3000
|
|
}, csvData.personality);
|
|
|
|
const chunkResults = parseStyleResponse(styledResponse, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk ${chunkIndex + 1}: ${Object.keys(chunkResults).length} stylisés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: garder contenu original
|
|
chunk.forEach(element => {
|
|
results[element.tag] = element.content;
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt de stylisation
|
|
*/
|
|
function createStylePrompt(chunk, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `MISSION: Adapte UNIQUEMENT le style de ces contenus selon ${personality.nom}.
|
|
|
|
CONTEXTE: Article SEO e-commerce ${csvData.mc0}
|
|
PERSONNALITÉ: ${personality.nom}
|
|
DESCRIPTION: ${personality.description}
|
|
STYLE: ${personality.style} adapté web professionnel
|
|
VOCABULAIRE: ${personality.vocabulairePref}
|
|
CONNECTEURS: ${personality.connecteursPref}
|
|
NIVEAU TECHNIQUE: ${personality.niveauTechnique}
|
|
LONGUEUR PHRASES: ${personality.longueurPhrases}
|
|
|
|
CONTENUS À STYLISER:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag} (Priorité: ${item.priority.toFixed(1)})
|
|
CONTENU: "${item.content}"`).join('\n\n')}
|
|
|
|
OBJECTIFS STYLISATION ${personality.nom.toUpperCase()}:
|
|
- Adapte le TON selon ${personality.style}
|
|
- Vocabulaire: ${personality.vocabulairePref}
|
|
- Connecteurs variés: ${personality.connecteursPref}
|
|
- Phrases: ${personality.longueurPhrases}
|
|
- Niveau: ${personality.niveauTechnique}
|
|
|
|
CONSIGNES STRICTES:
|
|
- GARDE le même contenu informatif et technique
|
|
- Adapte SEULEMENT ton, expressions, vocabulaire selon ${personality.nom}
|
|
- RESPECTE longueur approximative (±20%)
|
|
- ÉVITE répétitions excessives
|
|
- Style ${personality.nom} reconnaissable mais NATUREL web
|
|
- PAS de messages d'excuse
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu stylisé selon ${personality.nom}
|
|
[2] Contenu stylisé selon ${personality.nom}
|
|
etc...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse stylisation
|
|
*/
|
|
function parseStyleResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let styledContent = match[2].trim();
|
|
const element = chunk[index];
|
|
|
|
// Nettoyer le contenu stylisé
|
|
styledContent = cleanStyledContent(styledContent);
|
|
|
|
if (styledContent && styledContent.length > 10) {
|
|
results[element.tag] = styledContent;
|
|
logSh(`✅ Styled [${element.tag}]: "${styledContent.substring(0, 100)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content;
|
|
logSh(`⚠️ Fallback [${element.tag}]: stylisation invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < chunk.length) {
|
|
const element = chunk[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Nettoyer contenu stylisé
|
|
*/
|
|
function cleanStyledContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?voici\s+/gi, '');
|
|
content = content.replace(/^pour\s+ce\s+contenu[,\s]*/gi, '');
|
|
content = content.replace(/\*\*[^*]+\*\*/g, '');
|
|
|
|
// Réduire répétitions excessives mais garder le style personnalité
|
|
content = content.replace(/(du coup[,\s]+){4,}/gi, 'du coup ');
|
|
content = content.replace(/(bon[,\s]+){4,}/gi, 'bon ');
|
|
content = content.replace(/(franchement[,\s]+){3,}/gi, 'franchement ');
|
|
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
/**
|
|
* Obtenir instructions de style dynamiques
|
|
*/
|
|
function getPersonalityStyleInstructions(personality) {
|
|
if (!personality) return "Style professionnel standard";
|
|
|
|
return `STYLE ${personality.nom.toUpperCase()} (${personality.style}):
|
|
- Description: ${personality.description}
|
|
- Vocabulaire: ${personality.vocabulairePref || 'professionnel'}
|
|
- Connecteurs: ${personality.connecteursPref || 'par ailleurs, en effet'}
|
|
- Mots-clés: ${personality.motsClesSecteurs || 'technique, qualité'}
|
|
- Phrases: ${personality.longueurPhrases || 'Moyennes'}
|
|
- Niveau: ${personality.niveauTechnique || 'Accessible'}
|
|
- CTA: ${personality.ctaStyle || 'Professionnel'}`;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
module.exports = {
|
|
applyPersonalityStyle, // ← MAIN ENTRY POINT
|
|
prepareElementsForStyling,
|
|
calculateStylePriority,
|
|
applyStyleInChunks,
|
|
createStylePrompt,
|
|
parseStyleResponse,
|
|
getPersonalityStyleInstructions
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/post-processing/SentenceVariation.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// PATTERN BREAKING - TECHNIQUE 1: SENTENCE VARIATION
|
|
// Responsabilité: Varier les longueurs de phrases pour casser l'uniformité
|
|
// Anti-détection: Éviter patterns syntaxiques réguliers des LLMs
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - VARIATION LONGUEUR PHRASES
|
|
* @param {Object} input - { content: {}, config: {}, context: {} }
|
|
* @returns {Object} - { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function applySentenceVariation(input) {
|
|
return await tracer.run('SentenceVariation.applySentenceVariation()', async () => {
|
|
const { content, config = {}, context = {} } = input;
|
|
|
|
const {
|
|
intensity = 0.3, // Probabilité de modification (30%)
|
|
splitThreshold = 100, // Chars pour split
|
|
mergeThreshold = 30, // Chars pour merge
|
|
preserveQuestions = true, // Préserver questions FAQ
|
|
preserveTitles = true // Préserver titres
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
technique: 'sentence_variation',
|
|
intensity,
|
|
elementsCount: Object.keys(content).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`📐 TECHNIQUE 1/3: Variation longueur phrases (intensité: ${intensity})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à analyser`, 'DEBUG');
|
|
|
|
try {
|
|
const results = {};
|
|
let totalProcessed = 0;
|
|
let totalModified = 0;
|
|
let modificationsDetails = [];
|
|
|
|
// Traiter chaque élément de contenu
|
|
for (const [tag, text] of Object.entries(content)) {
|
|
totalProcessed++;
|
|
|
|
// Skip certains éléments selon config
|
|
if (shouldSkipElement(tag, text, { preserveQuestions, preserveTitles })) {
|
|
results[tag] = text;
|
|
logSh(` ⏭️ [${tag}]: Préservé (${getSkipReason(tag, text)})`, 'DEBUG');
|
|
continue;
|
|
}
|
|
|
|
// Appliquer variation si éligible
|
|
const variationResult = varyTextStructure(text, {
|
|
intensity,
|
|
splitThreshold,
|
|
mergeThreshold,
|
|
tag
|
|
});
|
|
|
|
results[tag] = variationResult.text;
|
|
|
|
if (variationResult.modified) {
|
|
totalModified++;
|
|
modificationsDetails.push({
|
|
tag,
|
|
modifications: variationResult.modifications,
|
|
originalLength: text.length,
|
|
newLength: variationResult.text.length
|
|
});
|
|
|
|
logSh(` ✏️ [${tag}]: ${variationResult.modifications.length} modifications`, 'DEBUG');
|
|
} else {
|
|
logSh(` ➡️ [${tag}]: Aucune modification`, 'DEBUG');
|
|
}
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: totalProcessed,
|
|
modified: totalModified,
|
|
modificationRate: Math.round((totalModified / totalProcessed) * 100),
|
|
duration,
|
|
technique: 'sentence_variation'
|
|
};
|
|
|
|
logSh(`✅ VARIATION PHRASES: ${stats.modified}/${stats.processed} éléments modifiés (${stats.modificationRate}%) en ${duration}ms`, 'INFO');
|
|
|
|
await tracer.event('Sentence variation terminée', stats);
|
|
|
|
return {
|
|
content: results,
|
|
stats,
|
|
debug: {
|
|
technique: 'sentence_variation',
|
|
config: { intensity, splitThreshold, mergeThreshold },
|
|
modifications: modificationsDetails
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ VARIATION PHRASES échouée après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`SentenceVariation failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Appliquer variation structure à un texte
|
|
*/
|
|
function varyTextStructure(text, config) {
|
|
const { intensity, splitThreshold, mergeThreshold, tag } = config;
|
|
|
|
if (text.length < 50) {
|
|
return { text, modified: false, modifications: [] };
|
|
}
|
|
|
|
// Séparer en phrases
|
|
const sentences = splitIntoSentences(text);
|
|
|
|
if (sentences.length < 2) {
|
|
return { text, modified: false, modifications: [] };
|
|
}
|
|
|
|
let modifiedSentences = [...sentences];
|
|
const modifications = [];
|
|
|
|
// TECHNIQUE 1: SPLIT des phrases longues
|
|
for (let i = 0; i < modifiedSentences.length; i++) {
|
|
const sentence = modifiedSentences[i];
|
|
|
|
if (sentence.length > splitThreshold && Math.random() < intensity) {
|
|
const splitResult = splitLongSentence(sentence);
|
|
if (splitResult.success) {
|
|
modifiedSentences.splice(i, 1, splitResult.part1, splitResult.part2);
|
|
modifications.push({
|
|
type: 'split',
|
|
original: sentence.substring(0, 50) + '...',
|
|
result: `${splitResult.part1.substring(0, 25)}... | ${splitResult.part2.substring(0, 25)}...`
|
|
});
|
|
i++; // Skip la phrase suivante (qui est notre part2)
|
|
}
|
|
}
|
|
}
|
|
|
|
// TECHNIQUE 2: MERGE des phrases courtes
|
|
for (let i = 0; i < modifiedSentences.length - 1; i++) {
|
|
const current = modifiedSentences[i];
|
|
const next = modifiedSentences[i + 1];
|
|
|
|
if (current.length < mergeThreshold && next.length < mergeThreshold && Math.random() < intensity) {
|
|
const merged = mergeSentences(current, next);
|
|
if (merged.success) {
|
|
modifiedSentences.splice(i, 2, merged.result);
|
|
modifications.push({
|
|
type: 'merge',
|
|
original: `${current.substring(0, 20)}... + ${next.substring(0, 20)}...`,
|
|
result: merged.result.substring(0, 50) + '...'
|
|
});
|
|
}
|
|
}
|
|
}
|
|
|
|
const finalText = modifiedSentences.join(' ').trim();
|
|
|
|
return {
|
|
text: finalText,
|
|
modified: modifications.length > 0,
|
|
modifications
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Diviser texte en phrases
|
|
*/
|
|
function splitIntoSentences(text) {
|
|
// Regex plus sophistiquée pour gérer les abréviations
|
|
const sentences = text.split(/(?<![A-Z][a-z]\.)\s*[.!?]+\s+/)
|
|
.map(s => s.trim())
|
|
.filter(s => s.length > 5);
|
|
|
|
return sentences;
|
|
}
|
|
|
|
/**
|
|
* Diviser une phrase longue en deux
|
|
*/
|
|
function splitLongSentence(sentence) {
|
|
// Points de rupture naturels
|
|
const breakPoints = [
|
|
', et ',
|
|
', mais ',
|
|
', car ',
|
|
', donc ',
|
|
', ainsi ',
|
|
', alors ',
|
|
', tandis que ',
|
|
', bien que '
|
|
];
|
|
|
|
// Chercher le meilleur point de rupture proche du milieu
|
|
const idealBreak = sentence.length / 2;
|
|
let bestBreak = null;
|
|
let bestDistance = Infinity;
|
|
|
|
for (const breakPoint of breakPoints) {
|
|
const index = sentence.indexOf(breakPoint, idealBreak - 50);
|
|
if (index > 0 && index < sentence.length - 20) {
|
|
const distance = Math.abs(index - idealBreak);
|
|
if (distance < bestDistance) {
|
|
bestDistance = distance;
|
|
bestBreak = { index, breakPoint };
|
|
}
|
|
}
|
|
}
|
|
|
|
if (bestBreak) {
|
|
const part1 = sentence.substring(0, bestBreak.index + 1).trim();
|
|
const part2 = sentence.substring(bestBreak.index + bestBreak.breakPoint.length).trim();
|
|
|
|
// Assurer que part2 commence par une majuscule
|
|
const capitalizedPart2 = part2.charAt(0).toUpperCase() + part2.slice(1);
|
|
|
|
return {
|
|
success: true,
|
|
part1,
|
|
part2: capitalizedPart2
|
|
};
|
|
}
|
|
|
|
return { success: false };
|
|
}
|
|
|
|
/**
|
|
* Fusionner deux phrases courtes
|
|
*/
|
|
function mergeSentences(sentence1, sentence2) {
|
|
// Connecteurs pour fusion naturelle
|
|
const connectors = [
|
|
'et',
|
|
'puis',
|
|
'aussi',
|
|
'également',
|
|
'de plus'
|
|
];
|
|
|
|
// Choisir connecteur aléatoire
|
|
const connector = connectors[Math.floor(Math.random() * connectors.length)];
|
|
|
|
// Nettoyer les phrases
|
|
let cleaned1 = sentence1.replace(/[.!?]+$/, '').trim();
|
|
let cleaned2 = sentence2.trim();
|
|
|
|
// Mettre sentence2 en minuscule sauf si nom propre
|
|
if (!/^[A-Z][a-z]*\s+[A-Z]/.test(cleaned2)) {
|
|
cleaned2 = cleaned2.charAt(0).toLowerCase() + cleaned2.slice(1);
|
|
}
|
|
|
|
const merged = `${cleaned1}, ${connector} ${cleaned2}`;
|
|
|
|
return {
|
|
success: merged.length < 200, // Éviter phrases trop longues
|
|
result: merged
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Déterminer si un élément doit être skippé
|
|
*/
|
|
function shouldSkipElement(tag, text, config) {
|
|
// Skip titres si demandé
|
|
if (config.preserveTitles && (tag.includes('Titre') || tag.includes('H1') || tag.includes('H2'))) {
|
|
return true;
|
|
}
|
|
|
|
// Skip questions FAQ si demandé
|
|
if (config.preserveQuestions && (tag.includes('Faq_q') || text.includes('?'))) {
|
|
return true;
|
|
}
|
|
|
|
// Skip textes très courts
|
|
if (text.length < 50) {
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
/**
|
|
* Obtenir raison du skip pour debug
|
|
*/
|
|
function getSkipReason(tag, text) {
|
|
if (tag.includes('Titre') || tag.includes('H1') || tag.includes('H2')) return 'titre';
|
|
if (tag.includes('Faq_q') || text.includes('?')) return 'question';
|
|
if (text.length < 50) return 'trop court';
|
|
return 'autre';
|
|
}
|
|
|
|
/**
|
|
* Analyser les patterns de phrases d'un texte
|
|
*/
|
|
function analyzeSentencePatterns(text) {
|
|
const sentences = splitIntoSentences(text);
|
|
|
|
if (sentences.length < 2) {
|
|
return { needsVariation: false, patterns: [] };
|
|
}
|
|
|
|
const lengths = sentences.map(s => s.length);
|
|
const avgLength = lengths.reduce((a, b) => a + b, 0) / lengths.length;
|
|
|
|
// Calculer uniformité (variance faible = uniformité élevée)
|
|
const variance = lengths.reduce((acc, len) => acc + Math.pow(len - avgLength, 2), 0) / lengths.length;
|
|
const uniformity = 1 / (1 + Math.sqrt(variance) / avgLength); // 0-1, 1 = très uniforme
|
|
|
|
return {
|
|
needsVariation: uniformity > 0.7, // Seuil d'uniformité problématique
|
|
patterns: {
|
|
avgLength: Math.round(avgLength),
|
|
uniformity: Math.round(uniformity * 100),
|
|
sentenceCount: sentences.length,
|
|
variance: Math.round(variance)
|
|
}
|
|
};
|
|
}
|
|
|
|
module.exports = {
|
|
applySentenceVariation, // ← MAIN ENTRY POINT
|
|
varyTextStructure,
|
|
splitIntoSentences,
|
|
splitLongSentence,
|
|
mergeSentences,
|
|
analyzeSentencePatterns
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/post-processing/LLMFingerprintRemoval.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// PATTERN BREAKING - TECHNIQUE 2: LLM FINGERPRINT REMOVAL
|
|
// Responsabilité: Remplacer mots/expressions typiques des LLMs
|
|
// Anti-détection: Éviter vocabulaire détectable par les analyseurs IA
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* DICTIONNAIRE ANTI-DÉTECTION
|
|
* Mots/expressions LLM → Alternatives humaines naturelles
|
|
*/
|
|
const LLM_FINGERPRINTS = {
|
|
// Mots techniques/corporate typiques IA
|
|
'optimal': ['idéal', 'parfait', 'adapté', 'approprié', 'convenable'],
|
|
'optimale': ['idéale', 'parfaite', 'adaptée', 'appropriée', 'convenable'],
|
|
'comprehensive': ['complet', 'détaillé', 'exhaustif', 'approfondi', 'global'],
|
|
'seamless': ['fluide', 'naturel', 'sans accroc', 'harmonieux', 'lisse'],
|
|
'robust': ['solide', 'fiable', 'résistant', 'costaud', 'stable'],
|
|
'robuste': ['solide', 'fiable', 'résistant', 'costaud', 'stable'],
|
|
|
|
// Expressions trop formelles/IA
|
|
'il convient de noter': ['on remarque', 'il faut savoir', 'à noter', 'important'],
|
|
'il convient de': ['il faut', 'on doit', 'mieux vaut', 'il est bon de'],
|
|
'par conséquent': ['du coup', 'donc', 'résultat', 'ainsi'],
|
|
'néanmoins': ['cependant', 'mais', 'pourtant', 'malgré tout'],
|
|
'toutefois': ['cependant', 'mais', 'pourtant', 'quand même'],
|
|
'de surcroît': ['de plus', 'en plus', 'aussi', 'également'],
|
|
|
|
// Superlatifs excessifs typiques IA
|
|
'extrêmement': ['très', 'super', 'vraiment', 'particulièrement'],
|
|
'particulièrement': ['très', 'vraiment', 'spécialement', 'surtout'],
|
|
'remarquablement': ['très', 'vraiment', 'sacrément', 'fichement'],
|
|
'exceptionnellement': ['très', 'vraiment', 'super', 'incroyablement'],
|
|
|
|
// Mots de liaison trop mécaniques
|
|
'en définitive': ['au final', 'finalement', 'bref', 'en gros'],
|
|
'il s\'avère que': ['on voit que', 'il se trouve que', 'en fait'],
|
|
'force est de constater': ['on constate', 'on voit bien', 'c\'est clair'],
|
|
|
|
// Expressions commerciales robotiques
|
|
'solution innovante': ['nouveauté', 'innovation', 'solution moderne', 'nouvelle approche'],
|
|
'approche holistique': ['approche globale', 'vision d\'ensemble', 'approche complète'],
|
|
'expérience utilisateur': ['confort d\'utilisation', 'facilité d\'usage', 'ergonomie'],
|
|
'retour sur investissement': ['rentabilité', 'bénéfices', 'profits'],
|
|
|
|
// Adjectifs surutilisés par IA
|
|
'révolutionnaire': ['nouveau', 'moderne', 'innovant', 'original'],
|
|
'game-changer': ['nouveauté', 'innovation', 'changement', 'révolution'],
|
|
'cutting-edge': ['moderne', 'récent', 'nouveau', 'avancé'],
|
|
'state-of-the-art': ['moderne', 'récent', 'performant', 'haut de gamme']
|
|
};
|
|
|
|
/**
|
|
* EXPRESSIONS CONTEXTUELLES SECTEUR SIGNALÉTIQUE
|
|
* Adaptées au domaine métier pour plus de naturel
|
|
*/
|
|
const CONTEXTUAL_REPLACEMENTS = {
|
|
'solution': {
|
|
'signalétique': ['plaque', 'panneau', 'support', 'réalisation'],
|
|
'impression': ['tirage', 'print', 'production', 'fabrication'],
|
|
'default': ['option', 'possibilité', 'choix', 'alternative']
|
|
},
|
|
'produit': {
|
|
'signalétique': ['plaque', 'panneau', 'enseigne', 'support'],
|
|
'default': ['article', 'réalisation', 'création']
|
|
},
|
|
'service': {
|
|
'signalétique': ['prestation', 'réalisation', 'travail', 'création'],
|
|
'default': ['prestation', 'travail', 'aide']
|
|
}
|
|
};
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - SUPPRESSION EMPREINTES LLM
|
|
* @param {Object} input - { content: {}, config: {}, context: {} }
|
|
* @returns {Object} - { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function removeLLMFingerprints(input) {
|
|
return await tracer.run('LLMFingerprintRemoval.removeLLMFingerprints()', async () => {
|
|
const { content, config = {}, context = {} } = input;
|
|
|
|
const {
|
|
intensity = 1.0, // Probabilité de remplacement (100%)
|
|
preserveKeywords = true, // Préserver mots-clés SEO
|
|
contextualMode = true, // Mode contextuel métier
|
|
csvData = null // Pour contexte métier
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
technique: 'fingerprint_removal',
|
|
intensity,
|
|
elementsCount: Object.keys(content).length,
|
|
contextualMode
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🔍 TECHNIQUE 2/3: Suppression empreintes LLM (intensité: ${intensity})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à nettoyer`, 'DEBUG');
|
|
|
|
try {
|
|
const results = {};
|
|
let totalProcessed = 0;
|
|
let totalReplacements = 0;
|
|
let replacementDetails = [];
|
|
|
|
// Préparer contexte métier
|
|
const businessContext = extractBusinessContext(csvData);
|
|
|
|
// Traiter chaque élément de contenu
|
|
for (const [tag, text] of Object.entries(content)) {
|
|
totalProcessed++;
|
|
|
|
if (text.length < 20) {
|
|
results[tag] = text;
|
|
continue;
|
|
}
|
|
|
|
// Appliquer suppression des empreintes
|
|
const cleaningResult = cleanTextFingerprints(text, {
|
|
intensity,
|
|
preserveKeywords,
|
|
contextualMode,
|
|
businessContext,
|
|
tag
|
|
});
|
|
|
|
results[tag] = cleaningResult.text;
|
|
|
|
if (cleaningResult.replacements.length > 0) {
|
|
totalReplacements += cleaningResult.replacements.length;
|
|
replacementDetails.push({
|
|
tag,
|
|
replacements: cleaningResult.replacements,
|
|
fingerprintsFound: cleaningResult.fingerprintsDetected
|
|
});
|
|
|
|
logSh(` 🧹 [${tag}]: ${cleaningResult.replacements.length} remplacements`, 'DEBUG');
|
|
} else {
|
|
logSh(` ✅ [${tag}]: Aucune empreinte détectée`, 'DEBUG');
|
|
}
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: totalProcessed,
|
|
totalReplacements,
|
|
avgReplacementsPerElement: Math.round(totalReplacements / totalProcessed * 100) / 100,
|
|
elementsWithFingerprints: replacementDetails.length,
|
|
duration,
|
|
technique: 'fingerprint_removal'
|
|
};
|
|
|
|
logSh(`✅ NETTOYAGE EMPREINTES: ${stats.totalReplacements} remplacements sur ${stats.elementsWithFingerprints}/${stats.processed} éléments en ${duration}ms`, 'INFO');
|
|
|
|
await tracer.event('Fingerprint removal terminée', stats);
|
|
|
|
return {
|
|
content: results,
|
|
stats,
|
|
debug: {
|
|
technique: 'fingerprint_removal',
|
|
config: { intensity, preserveKeywords, contextualMode },
|
|
replacements: replacementDetails,
|
|
businessContext
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ NETTOYAGE EMPREINTES échoué après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`LLMFingerprintRemoval failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Nettoyer les empreintes LLM d'un texte
|
|
*/
|
|
function cleanTextFingerprints(text, config) {
|
|
const { intensity, preserveKeywords, contextualMode, businessContext, tag } = config;
|
|
|
|
let cleanedText = text;
|
|
const replacements = [];
|
|
const fingerprintsDetected = [];
|
|
|
|
// PHASE 1: Remplacements directs du dictionnaire
|
|
for (const [fingerprint, alternatives] of Object.entries(LLM_FINGERPRINTS)) {
|
|
const regex = new RegExp(`\\b${escapeRegex(fingerprint)}\\b`, 'gi');
|
|
const matches = text.match(regex);
|
|
|
|
if (matches) {
|
|
fingerprintsDetected.push(fingerprint);
|
|
|
|
// Appliquer remplacement selon intensité
|
|
if (Math.random() <= intensity) {
|
|
const alternative = selectBestAlternative(alternatives, businessContext, contextualMode);
|
|
|
|
cleanedText = cleanedText.replace(regex, (match) => {
|
|
// Préserver la casse originale
|
|
return preserveCase(match, alternative);
|
|
});
|
|
|
|
replacements.push({
|
|
type: 'direct',
|
|
original: fingerprint,
|
|
replacement: alternative,
|
|
occurrences: matches.length
|
|
});
|
|
}
|
|
}
|
|
}
|
|
|
|
// PHASE 2: Remplacements contextuels
|
|
if (contextualMode && businessContext) {
|
|
const contextualReplacements = applyContextualReplacements(cleanedText, businessContext);
|
|
cleanedText = contextualReplacements.text;
|
|
replacements.push(...contextualReplacements.replacements);
|
|
}
|
|
|
|
// PHASE 3: Détection patterns récurrents
|
|
const patternReplacements = replaceRecurringPatterns(cleanedText, intensity);
|
|
cleanedText = patternReplacements.text;
|
|
replacements.push(...patternReplacements.replacements);
|
|
|
|
return {
|
|
text: cleanedText,
|
|
replacements,
|
|
fingerprintsDetected
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Sélectionner la meilleure alternative selon le contexte
|
|
*/
|
|
function selectBestAlternative(alternatives, businessContext, contextualMode) {
|
|
if (!contextualMode || !businessContext) {
|
|
// Mode aléatoire simple
|
|
return alternatives[Math.floor(Math.random() * alternatives.length)];
|
|
}
|
|
|
|
// Mode contextuel : privilégier alternatives adaptées au métier
|
|
const contextualAlternatives = alternatives.filter(alt =>
|
|
isContextuallyAppropriate(alt, businessContext)
|
|
);
|
|
|
|
const finalAlternatives = contextualAlternatives.length > 0 ? contextualAlternatives : alternatives;
|
|
return finalAlternatives[Math.floor(Math.random() * finalAlternatives.length)];
|
|
}
|
|
|
|
/**
|
|
* Vérifier si une alternative est contextuelle appropriée
|
|
*/
|
|
function isContextuallyAppropriate(alternative, businessContext) {
|
|
const { sector, vocabulary } = businessContext;
|
|
|
|
// Signalétique : privilégier vocabulaire technique/artisanal
|
|
if (sector === 'signalétique') {
|
|
const technicalWords = ['solide', 'fiable', 'costaud', 'résistant', 'adapté'];
|
|
return technicalWords.includes(alternative);
|
|
}
|
|
|
|
return true; // Par défaut accepter
|
|
}
|
|
|
|
/**
|
|
* Appliquer remplacements contextuels
|
|
*/
|
|
function applyContextualReplacements(text, businessContext) {
|
|
let processedText = text;
|
|
const replacements = [];
|
|
|
|
for (const [word, contexts] of Object.entries(CONTEXTUAL_REPLACEMENTS)) {
|
|
const regex = new RegExp(`\\b${word}\\b`, 'gi');
|
|
const matches = processedText.match(regex);
|
|
|
|
if (matches) {
|
|
const contextAlternatives = contexts[businessContext.sector] || contexts.default;
|
|
const replacement = contextAlternatives[Math.floor(Math.random() * contextAlternatives.length)];
|
|
|
|
processedText = processedText.replace(regex, (match) => {
|
|
return preserveCase(match, replacement);
|
|
});
|
|
|
|
replacements.push({
|
|
type: 'contextual',
|
|
original: word,
|
|
replacement,
|
|
occurrences: matches.length,
|
|
context: businessContext.sector
|
|
});
|
|
}
|
|
}
|
|
|
|
return { text: processedText, replacements };
|
|
}
|
|
|
|
/**
|
|
* Remplacer patterns récurrents
|
|
*/
|
|
function replaceRecurringPatterns(text, intensity) {
|
|
let processedText = text;
|
|
const replacements = [];
|
|
|
|
// Pattern 1: "très + adjectif" → variantes
|
|
const veryPattern = /\btrès\s+(\w+)/gi;
|
|
const veryMatches = [...text.matchAll(veryPattern)];
|
|
|
|
if (veryMatches.length > 2 && Math.random() < intensity) {
|
|
// Remplacer certains "très" par des alternatives
|
|
const alternatives = ['super', 'vraiment', 'particulièrement', 'assez'];
|
|
|
|
veryMatches.slice(1).forEach((match, index) => {
|
|
if (Math.random() < 0.5) {
|
|
const alternative = alternatives[Math.floor(Math.random() * alternatives.length)];
|
|
const fullMatch = match[0];
|
|
const adjective = match[1];
|
|
const replacement = `${alternative} ${adjective}`;
|
|
|
|
processedText = processedText.replace(fullMatch, replacement);
|
|
|
|
replacements.push({
|
|
type: 'pattern',
|
|
pattern: '"très + adjectif"',
|
|
original: fullMatch,
|
|
replacement
|
|
});
|
|
}
|
|
});
|
|
}
|
|
|
|
return { text: processedText, replacements };
|
|
}
|
|
|
|
/**
|
|
* Extraire contexte métier des données CSV
|
|
*/
|
|
function extractBusinessContext(csvData) {
|
|
if (!csvData) {
|
|
return { sector: 'general', vocabulary: [] };
|
|
}
|
|
|
|
const mc0 = csvData.mc0?.toLowerCase() || '';
|
|
|
|
// Détection secteur
|
|
let sector = 'general';
|
|
if (mc0.includes('plaque') || mc0.includes('panneau') || mc0.includes('enseigne')) {
|
|
sector = 'signalétique';
|
|
} else if (mc0.includes('impression') || mc0.includes('print')) {
|
|
sector = 'impression';
|
|
}
|
|
|
|
// Extraction vocabulaire clé
|
|
const vocabulary = [csvData.mc0, csvData.t0, csvData.tMinus1].filter(Boolean);
|
|
|
|
return { sector, vocabulary };
|
|
}
|
|
|
|
/**
|
|
* Préserver la casse originale
|
|
*/
|
|
function preserveCase(original, replacement) {
|
|
if (original === original.toUpperCase()) {
|
|
return replacement.toUpperCase();
|
|
} else if (original[0] === original[0].toUpperCase()) {
|
|
return replacement.charAt(0).toUpperCase() + replacement.slice(1).toLowerCase();
|
|
} else {
|
|
return replacement.toLowerCase();
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Échapper caractères regex
|
|
*/
|
|
function escapeRegex(text) {
|
|
return text.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
|
|
}
|
|
|
|
/**
|
|
* Analyser les empreintes LLM dans un texte
|
|
*/
|
|
function analyzeLLMFingerprints(text) {
|
|
const detectedFingerprints = [];
|
|
let totalMatches = 0;
|
|
|
|
for (const fingerprint of Object.keys(LLM_FINGERPRINTS)) {
|
|
const regex = new RegExp(`\\b${escapeRegex(fingerprint)}\\b`, 'gi');
|
|
const matches = text.match(regex);
|
|
|
|
if (matches) {
|
|
detectedFingerprints.push({
|
|
fingerprint,
|
|
occurrences: matches.length,
|
|
category: categorizefingerprint(fingerprint)
|
|
});
|
|
totalMatches += matches.length;
|
|
}
|
|
}
|
|
|
|
return {
|
|
hasFingerprints: detectedFingerprints.length > 0,
|
|
fingerprints: detectedFingerprints,
|
|
totalMatches,
|
|
riskLevel: calculateRiskLevel(detectedFingerprints, text.length)
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Catégoriser une empreinte LLM
|
|
*/
|
|
function categorizefingerprint(fingerprint) {
|
|
const categories = {
|
|
'technical': ['optimal', 'comprehensive', 'robust', 'seamless'],
|
|
'formal': ['il convient de', 'néanmoins', 'par conséquent'],
|
|
'superlative': ['extrêmement', 'particulièrement', 'remarquablement'],
|
|
'commercial': ['solution innovante', 'game-changer', 'révolutionnaire']
|
|
};
|
|
|
|
for (const [category, words] of Object.entries(categories)) {
|
|
if (words.some(word => fingerprint.includes(word))) {
|
|
return category;
|
|
}
|
|
}
|
|
|
|
return 'other';
|
|
}
|
|
|
|
/**
|
|
* Calculer niveau de risque de détection
|
|
*/
|
|
function calculateRiskLevel(fingerprints, textLength) {
|
|
if (fingerprints.length === 0) return 'low';
|
|
|
|
const fingerprintDensity = fingerprints.reduce((sum, fp) => sum + fp.occurrences, 0) / (textLength / 100);
|
|
|
|
if (fingerprintDensity > 3) return 'high';
|
|
if (fingerprintDensity > 1.5) return 'medium';
|
|
return 'low';
|
|
}
|
|
|
|
module.exports = {
|
|
removeLLMFingerprints, // ← MAIN ENTRY POINT
|
|
cleanTextFingerprints,
|
|
analyzeLLMFingerprints,
|
|
LLM_FINGERPRINTS,
|
|
CONTEXTUAL_REPLACEMENTS,
|
|
extractBusinessContext
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/post-processing/TransitionHumanization.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// PATTERN BREAKING - TECHNIQUE 3: TRANSITION HUMANIZATION
|
|
// Responsabilité: Remplacer connecteurs mécaniques par transitions naturelles
|
|
// Anti-détection: Éviter patterns de liaison typiques des LLMs
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* DICTIONNAIRE CONNECTEURS HUMANISÉS
|
|
* Connecteurs LLM → Alternatives naturelles par contexte
|
|
*/
|
|
const TRANSITION_REPLACEMENTS = {
|
|
// Connecteurs trop formels → versions naturelles
|
|
'par ailleurs': {
|
|
alternatives: ['d\'ailleurs', 'au fait', 'soit dit en passant', 'à propos', 'sinon'],
|
|
weight: 0.8,
|
|
contexts: ['casual', 'conversational']
|
|
},
|
|
|
|
'en effet': {
|
|
alternatives: ['effectivement', 'c\'est vrai', 'tout à fait', 'absolument', 'exactement'],
|
|
weight: 0.9,
|
|
contexts: ['confirmative', 'agreement']
|
|
},
|
|
|
|
'de plus': {
|
|
alternatives: ['aussi', 'également', 'qui plus est', 'en plus', 'et puis'],
|
|
weight: 0.7,
|
|
contexts: ['additive', 'continuation']
|
|
},
|
|
|
|
'cependant': {
|
|
alternatives: ['mais', 'pourtant', 'néanmoins', 'malgré tout', 'quand même'],
|
|
weight: 0.6,
|
|
contexts: ['contrast', 'opposition']
|
|
},
|
|
|
|
'ainsi': {
|
|
alternatives: ['donc', 'du coup', 'comme ça', 'par conséquent', 'résultat'],
|
|
weight: 0.8,
|
|
contexts: ['consequence', 'result']
|
|
},
|
|
|
|
'donc': {
|
|
alternatives: ['du coup', 'alors', 'par conséquent', 'ainsi', 'résultat'],
|
|
weight: 0.5,
|
|
contexts: ['consequence', 'logical']
|
|
},
|
|
|
|
// Connecteurs de séquence
|
|
'ensuite': {
|
|
alternatives: ['puis', 'après', 'et puis', 'alors', 'du coup'],
|
|
weight: 0.6,
|
|
contexts: ['sequence', 'temporal']
|
|
},
|
|
|
|
'puis': {
|
|
alternatives: ['ensuite', 'après', 'et puis', 'alors'],
|
|
weight: 0.4,
|
|
contexts: ['sequence', 'temporal']
|
|
},
|
|
|
|
// Connecteurs d'emphase
|
|
'également': {
|
|
alternatives: ['aussi', 'de même', 'pareillement', 'en plus'],
|
|
weight: 0.6,
|
|
contexts: ['similarity', 'addition']
|
|
},
|
|
|
|
'aussi': {
|
|
alternatives: ['également', 'de même', 'en plus', 'pareillement'],
|
|
weight: 0.3,
|
|
contexts: ['similarity', 'addition']
|
|
},
|
|
|
|
// Connecteurs de conclusion
|
|
'enfin': {
|
|
alternatives: ['finalement', 'au final', 'pour finir', 'en dernier'],
|
|
weight: 0.5,
|
|
contexts: ['conclusion', 'final']
|
|
},
|
|
|
|
'finalement': {
|
|
alternatives: ['au final', 'en fin de compte', 'pour finir', 'enfin'],
|
|
weight: 0.4,
|
|
contexts: ['conclusion', 'final']
|
|
}
|
|
};
|
|
|
|
/**
|
|
* PATTERNS DE TRANSITION NATURELLE
|
|
* Selon le style de personnalité
|
|
*/
|
|
const PERSONALITY_TRANSITIONS = {
|
|
'décontracté': {
|
|
preferred: ['du coup', 'alors', 'bon', 'après', 'sinon'],
|
|
avoided: ['par conséquent', 'néanmoins', 'toutefois']
|
|
},
|
|
|
|
'technique': {
|
|
preferred: ['donc', 'ainsi', 'par conséquent', 'résultat'],
|
|
avoided: ['du coup', 'bon', 'franchement']
|
|
},
|
|
|
|
'commercial': {
|
|
preferred: ['aussi', 'de plus', 'également', 'qui plus est'],
|
|
avoided: ['du coup', 'bon', 'franchement']
|
|
},
|
|
|
|
'familier': {
|
|
preferred: ['du coup', 'bon', 'alors', 'après', 'franchement'],
|
|
avoided: ['par conséquent', 'néanmoins', 'de surcroît']
|
|
}
|
|
};
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - HUMANISATION TRANSITIONS
|
|
* @param {Object} input - { content: {}, config: {}, context: {} }
|
|
* @returns {Object} - { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function humanizeTransitions(input) {
|
|
return await tracer.run('TransitionHumanization.humanizeTransitions()', async () => {
|
|
const { content, config = {}, context = {} } = input;
|
|
|
|
const {
|
|
intensity = 0.6, // Probabilité de remplacement (60%)
|
|
personalityStyle = null, // Style de personnalité pour guidage
|
|
avoidRepetition = true, // Éviter répétitions excessives
|
|
preserveFormal = false, // Préserver style formel
|
|
csvData = null // Données pour personnalité
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
technique: 'transition_humanization',
|
|
intensity,
|
|
personalityStyle: personalityStyle || csvData?.personality?.style,
|
|
elementsCount: Object.keys(content).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🔗 TECHNIQUE 3/3: Humanisation transitions (intensité: ${intensity})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à humaniser`, 'DEBUG');
|
|
|
|
try {
|
|
const results = {};
|
|
let totalProcessed = 0;
|
|
let totalReplacements = 0;
|
|
let humanizationDetails = [];
|
|
|
|
// Extraire style de personnalité
|
|
const effectivePersonalityStyle = personalityStyle || csvData?.personality?.style || 'neutral';
|
|
|
|
// Analyser patterns globaux pour éviter répétitions
|
|
const globalPatterns = analyzeGlobalTransitionPatterns(content);
|
|
|
|
// Traiter chaque élément de contenu
|
|
for (const [tag, text] of Object.entries(content)) {
|
|
totalProcessed++;
|
|
|
|
if (text.length < 30) {
|
|
results[tag] = text;
|
|
continue;
|
|
}
|
|
|
|
// Appliquer humanisation des transitions
|
|
const humanizationResult = humanizeTextTransitions(text, {
|
|
intensity,
|
|
personalityStyle: effectivePersonalityStyle,
|
|
avoidRepetition,
|
|
preserveFormal,
|
|
globalPatterns,
|
|
tag
|
|
});
|
|
|
|
results[tag] = humanizationResult.text;
|
|
|
|
if (humanizationResult.replacements.length > 0) {
|
|
totalReplacements += humanizationResult.replacements.length;
|
|
humanizationDetails.push({
|
|
tag,
|
|
replacements: humanizationResult.replacements,
|
|
transitionsDetected: humanizationResult.transitionsFound
|
|
});
|
|
|
|
logSh(` 🔄 [${tag}]: ${humanizationResult.replacements.length} transitions humanisées`, 'DEBUG');
|
|
} else {
|
|
logSh(` ➡️ [${tag}]: Transitions déjà naturelles`, 'DEBUG');
|
|
}
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: totalProcessed,
|
|
totalReplacements,
|
|
avgReplacementsPerElement: Math.round(totalReplacements / totalProcessed * 100) / 100,
|
|
elementsWithTransitions: humanizationDetails.length,
|
|
personalityStyle: effectivePersonalityStyle,
|
|
duration,
|
|
technique: 'transition_humanization'
|
|
};
|
|
|
|
logSh(`✅ HUMANISATION TRANSITIONS: ${stats.totalReplacements} remplacements sur ${stats.elementsWithTransitions}/${stats.processed} éléments en ${duration}ms`, 'INFO');
|
|
|
|
await tracer.event('Transition humanization terminée', stats);
|
|
|
|
return {
|
|
content: results,
|
|
stats,
|
|
debug: {
|
|
technique: 'transition_humanization',
|
|
config: { intensity, personalityStyle: effectivePersonalityStyle, avoidRepetition },
|
|
humanizations: humanizationDetails,
|
|
globalPatterns
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ HUMANISATION TRANSITIONS échouée après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`TransitionHumanization failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Humaniser les transitions d'un texte
|
|
*/
|
|
function humanizeTextTransitions(text, config) {
|
|
const { intensity, personalityStyle, avoidRepetition, preserveFormal, globalPatterns, tag } = config;
|
|
|
|
let humanizedText = text;
|
|
const replacements = [];
|
|
const transitionsFound = [];
|
|
|
|
// Statistiques usage pour éviter répétitions
|
|
const usageStats = {};
|
|
|
|
// Traiter chaque connecteur du dictionnaire
|
|
for (const [transition, transitionData] of Object.entries(TRANSITION_REPLACEMENTS)) {
|
|
const { alternatives, weight, contexts } = transitionData;
|
|
|
|
// Rechercher occurrences (insensible à la casse, mais préserver limites mots)
|
|
const regex = new RegExp(`\\b${escapeRegex(transition)}\\b`, 'gi');
|
|
const matches = [...text.matchAll(regex)];
|
|
|
|
if (matches.length > 0) {
|
|
transitionsFound.push(transition);
|
|
|
|
// Décider si on remplace selon intensité et poids
|
|
const shouldReplace = Math.random() < (intensity * weight);
|
|
|
|
if (shouldReplace && !preserveFormal) {
|
|
// Sélectionner meilleure alternative
|
|
const selectedAlternative = selectBestTransitionAlternative(
|
|
alternatives,
|
|
personalityStyle,
|
|
usageStats,
|
|
avoidRepetition
|
|
);
|
|
|
|
// Appliquer remplacement en préservant la casse
|
|
humanizedText = humanizedText.replace(regex, (match) => {
|
|
return preserveCase(match, selectedAlternative);
|
|
});
|
|
|
|
// Enregistrer usage
|
|
usageStats[selectedAlternative] = (usageStats[selectedAlternative] || 0) + matches.length;
|
|
|
|
replacements.push({
|
|
original: transition,
|
|
replacement: selectedAlternative,
|
|
occurrences: matches.length,
|
|
contexts,
|
|
personalityMatch: isPersonalityAppropriate(selectedAlternative, personalityStyle)
|
|
});
|
|
}
|
|
}
|
|
}
|
|
|
|
// Post-processing : éviter accumulations
|
|
if (avoidRepetition) {
|
|
const repetitionCleaned = reduceTransitionRepetition(humanizedText, usageStats);
|
|
humanizedText = repetitionCleaned.text;
|
|
replacements.push(...repetitionCleaned.additionalChanges);
|
|
}
|
|
|
|
return {
|
|
text: humanizedText,
|
|
replacements,
|
|
transitionsFound
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Sélectionner meilleure alternative de transition
|
|
*/
|
|
function selectBestTransitionAlternative(alternatives, personalityStyle, usageStats, avoidRepetition) {
|
|
// Filtrer selon personnalité
|
|
const personalityFiltered = alternatives.filter(alt =>
|
|
isPersonalityAppropriate(alt, personalityStyle)
|
|
);
|
|
|
|
const candidateList = personalityFiltered.length > 0 ? personalityFiltered : alternatives;
|
|
|
|
if (!avoidRepetition) {
|
|
return candidateList[Math.floor(Math.random() * candidateList.length)];
|
|
}
|
|
|
|
// Éviter les alternatives déjà trop utilisées
|
|
const lessUsedAlternatives = candidateList.filter(alt =>
|
|
(usageStats[alt] || 0) < 2
|
|
);
|
|
|
|
const finalList = lessUsedAlternatives.length > 0 ? lessUsedAlternatives : candidateList;
|
|
return finalList[Math.floor(Math.random() * finalList.length)];
|
|
}
|
|
|
|
/**
|
|
* Vérifier si alternative appropriée pour personnalité
|
|
*/
|
|
function isPersonalityAppropriate(alternative, personalityStyle) {
|
|
if (!personalityStyle || personalityStyle === 'neutral') return true;
|
|
|
|
const styleMapping = {
|
|
'décontracté': PERSONALITY_TRANSITIONS.décontracté,
|
|
'technique': PERSONALITY_TRANSITIONS.technique,
|
|
'commercial': PERSONALITY_TRANSITIONS.commercial,
|
|
'familier': PERSONALITY_TRANSITIONS.familier
|
|
};
|
|
|
|
const styleConfig = styleMapping[personalityStyle.toLowerCase()];
|
|
if (!styleConfig) return true;
|
|
|
|
// Éviter les connecteurs inappropriés
|
|
if (styleConfig.avoided.includes(alternative)) return false;
|
|
|
|
// Privilégier les connecteurs préférés
|
|
if (styleConfig.preferred.includes(alternative)) return true;
|
|
|
|
return true;
|
|
}
|
|
|
|
/**
|
|
* Réduire répétitions excessives de transitions
|
|
*/
|
|
function reduceTransitionRepetition(text, usageStats) {
|
|
let processedText = text;
|
|
const additionalChanges = [];
|
|
|
|
// Identifier connecteurs surutilisés (>3 fois)
|
|
const overusedTransitions = Object.entries(usageStats)
|
|
.filter(([transition, count]) => count > 3)
|
|
.map(([transition]) => transition);
|
|
|
|
for (const overusedTransition of overusedTransitions) {
|
|
// Remplacer quelques occurrences par des alternatives
|
|
const regex = new RegExp(`\\b${escapeRegex(overusedTransition)}\\b`, 'g');
|
|
let replacements = 0;
|
|
|
|
processedText = processedText.replace(regex, (match, offset) => {
|
|
// Remplacer 1 occurrence sur 3 environ
|
|
if (Math.random() < 0.33 && replacements < 2) {
|
|
replacements++;
|
|
const alternatives = findAlternativesFor(overusedTransition);
|
|
const alternative = alternatives[Math.floor(Math.random() * alternatives.length)];
|
|
|
|
additionalChanges.push({
|
|
type: 'repetition_reduction',
|
|
original: overusedTransition,
|
|
replacement: alternative,
|
|
reason: 'overuse'
|
|
});
|
|
|
|
return preserveCase(match, alternative);
|
|
}
|
|
return match;
|
|
});
|
|
}
|
|
|
|
return { text: processedText, additionalChanges };
|
|
}
|
|
|
|
/**
|
|
* Trouver alternatives pour un connecteur donné
|
|
*/
|
|
function findAlternativesFor(transition) {
|
|
// Chercher dans le dictionnaire
|
|
for (const [key, data] of Object.entries(TRANSITION_REPLACEMENTS)) {
|
|
if (data.alternatives.includes(transition)) {
|
|
return data.alternatives.filter(alt => alt !== transition);
|
|
}
|
|
}
|
|
|
|
// Alternatives génériques
|
|
const genericAlternatives = {
|
|
'du coup': ['alors', 'donc', 'ainsi'],
|
|
'alors': ['du coup', 'donc', 'ensuite'],
|
|
'donc': ['du coup', 'alors', 'ainsi'],
|
|
'aussi': ['également', 'de plus', 'en plus'],
|
|
'mais': ['cependant', 'pourtant', 'néanmoins']
|
|
};
|
|
|
|
return genericAlternatives[transition] || ['donc', 'alors'];
|
|
}
|
|
|
|
/**
|
|
* Analyser patterns globaux de transitions
|
|
*/
|
|
function analyzeGlobalTransitionPatterns(content) {
|
|
const allText = Object.values(content).join(' ');
|
|
const transitionCounts = {};
|
|
const repetitionPatterns = [];
|
|
|
|
// Compter occurrences globales
|
|
for (const transition of Object.keys(TRANSITION_REPLACEMENTS)) {
|
|
const regex = new RegExp(`\\b${escapeRegex(transition)}\\b`, 'gi');
|
|
const matches = allText.match(regex);
|
|
if (matches) {
|
|
transitionCounts[transition] = matches.length;
|
|
}
|
|
}
|
|
|
|
// Identifier patterns de répétition problématiques
|
|
const sortedTransitions = Object.entries(transitionCounts)
|
|
.sort(([,a], [,b]) => b - a)
|
|
.slice(0, 5); // Top 5 plus utilisées
|
|
|
|
sortedTransitions.forEach(([transition, count]) => {
|
|
if (count > 5) {
|
|
repetitionPatterns.push({
|
|
transition,
|
|
count,
|
|
severity: count > 10 ? 'high' : count > 7 ? 'medium' : 'low'
|
|
});
|
|
}
|
|
});
|
|
|
|
return {
|
|
transitionCounts,
|
|
repetitionPatterns,
|
|
diversityScore: Object.keys(transitionCounts).length / Math.max(1, Object.values(transitionCounts).reduce((a,b) => a+b, 0))
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Préserver la casse originale
|
|
*/
|
|
function preserveCase(original, replacement) {
|
|
if (original === original.toUpperCase()) {
|
|
return replacement.toUpperCase();
|
|
} else if (original[0] === original[0].toUpperCase()) {
|
|
return replacement.charAt(0).toUpperCase() + replacement.slice(1).toLowerCase();
|
|
} else {
|
|
return replacement.toLowerCase();
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Échapper caractères regex
|
|
*/
|
|
function escapeRegex(text) {
|
|
return text.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
|
|
}
|
|
|
|
/**
|
|
* Analyser qualité des transitions d'un texte
|
|
*/
|
|
function analyzeTransitionQuality(text) {
|
|
const sentences = text.split(/[.!?]+/).filter(s => s.trim().length > 5);
|
|
|
|
if (sentences.length < 2) {
|
|
return { score: 100, issues: [], naturalness: 'high' };
|
|
}
|
|
|
|
let mechanicalTransitions = 0;
|
|
let totalTransitions = 0;
|
|
const issues = [];
|
|
|
|
// Analyser chaque transition
|
|
sentences.forEach((sentence, index) => {
|
|
if (index === 0) return;
|
|
|
|
const trimmed = sentence.trim();
|
|
const startsWithTransition = Object.keys(TRANSITION_REPLACEMENTS).some(transition =>
|
|
trimmed.toLowerCase().startsWith(transition.toLowerCase())
|
|
);
|
|
|
|
if (startsWithTransition) {
|
|
totalTransitions++;
|
|
|
|
// Vérifier si transition mécanique
|
|
const transition = Object.keys(TRANSITION_REPLACEMENTS).find(t =>
|
|
trimmed.toLowerCase().startsWith(t.toLowerCase())
|
|
);
|
|
|
|
if (transition && TRANSITION_REPLACEMENTS[transition].weight > 0.7) {
|
|
mechanicalTransitions++;
|
|
issues.push({
|
|
type: 'mechanical_transition',
|
|
transition,
|
|
suggestion: TRANSITION_REPLACEMENTS[transition].alternatives[0]
|
|
});
|
|
}
|
|
}
|
|
});
|
|
|
|
const mechanicalRatio = totalTransitions > 0 ? mechanicalTransitions / totalTransitions : 0;
|
|
const score = Math.max(0, 100 - (mechanicalRatio * 100));
|
|
|
|
let naturalness = 'high';
|
|
if (mechanicalRatio > 0.5) naturalness = 'low';
|
|
else if (mechanicalRatio > 0.25) naturalness = 'medium';
|
|
|
|
return { score: Math.round(score), issues, naturalness, mechanicalRatio };
|
|
}
|
|
|
|
module.exports = {
|
|
humanizeTransitions, // ← MAIN ENTRY POINT
|
|
humanizeTextTransitions,
|
|
analyzeTransitionQuality,
|
|
analyzeGlobalTransitionPatterns,
|
|
TRANSITION_REPLACEMENTS,
|
|
PERSONALITY_TRANSITIONS
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/post-processing/PatternBreaking.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ORCHESTRATEUR PATTERN BREAKING - NIVEAU 2
|
|
// Responsabilité: Coordonner les 3 techniques anti-détection
|
|
// Objectif: -20% détection IA vs Niveau 1
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
// Import des 3 techniques Pattern Breaking
|
|
const { applySentenceVariation } = require('./SentenceVariation');
|
|
const { removeLLMFingerprints } = require('./LLMFingerprintRemoval');
|
|
const { humanizeTransitions } = require('./TransitionHumanization');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - PATTERN BREAKING COMPLET
|
|
* @param {Object} input - { content: {}, csvData: {}, options: {} }
|
|
* @returns {Object} - { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function applyPatternBreaking(input) {
|
|
return await tracer.run('PatternBreaking.applyPatternBreaking()', async () => {
|
|
const { content, csvData, options = {} } = input;
|
|
|
|
const config = {
|
|
// Configuration globale
|
|
intensity: 0.6, // Intensité générale (60%)
|
|
|
|
// Contrôle par technique
|
|
sentenceVariation: true, // Activer variation phrases
|
|
fingerprintRemoval: true, // Activer suppression empreintes
|
|
transitionHumanization: true, // Activer humanisation transitions
|
|
|
|
// Configuration spécifique par technique
|
|
sentenceVariationConfig: {
|
|
intensity: 0.3,
|
|
splitThreshold: 100,
|
|
mergeThreshold: 30,
|
|
preserveQuestions: true,
|
|
preserveTitles: true
|
|
},
|
|
|
|
fingerprintRemovalConfig: {
|
|
intensity: 1.0,
|
|
preserveKeywords: true,
|
|
contextualMode: true,
|
|
csvData
|
|
},
|
|
|
|
transitionHumanizationConfig: {
|
|
intensity: 0.6,
|
|
personalityStyle: csvData?.personality?.style,
|
|
avoidRepetition: true,
|
|
preserveFormal: false,
|
|
csvData
|
|
},
|
|
|
|
// Options avancées
|
|
qualityPreservation: true, // Préserver qualité contenu
|
|
seoIntegrity: true, // Maintenir intégrité SEO
|
|
readabilityCheck: true, // Vérifier lisibilité
|
|
|
|
...options // Override avec options fournies
|
|
};
|
|
|
|
await tracer.annotate({
|
|
level: 2,
|
|
technique: 'pattern_breaking',
|
|
elementsCount: Object.keys(content).length,
|
|
personality: csvData?.personality?.nom,
|
|
config: {
|
|
sentenceVariation: config.sentenceVariation,
|
|
fingerprintRemoval: config.fingerprintRemoval,
|
|
transitionHumanization: config.transitionHumanization,
|
|
intensity: config.intensity
|
|
}
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎯 NIVEAU 2: PATTERN BREAKING (3 techniques)`, 'INFO');
|
|
logSh(` 🎭 Personnalité: ${csvData?.personality?.nom} (${csvData?.personality?.style})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à traiter`, 'INFO');
|
|
logSh(` ⚙️ Techniques actives: ${[config.sentenceVariation && 'Variation', config.fingerprintRemoval && 'Empreintes', config.transitionHumanization && 'Transitions'].filter(Boolean).join(' + ')}`, 'INFO');
|
|
|
|
try {
|
|
let currentContent = { ...content };
|
|
const pipelineStats = {
|
|
techniques: [],
|
|
totalDuration: 0,
|
|
qualityMetrics: {}
|
|
};
|
|
|
|
// Analyse initiale de qualité
|
|
if (config.qualityPreservation) {
|
|
pipelineStats.qualityMetrics.initial = analyzeContentQuality(currentContent);
|
|
}
|
|
|
|
// TECHNIQUE 1: VARIATION LONGUEUR PHRASES
|
|
if (config.sentenceVariation) {
|
|
const step1Result = await applySentenceVariation({
|
|
content: currentContent,
|
|
config: config.sentenceVariationConfig,
|
|
context: { step: 1, totalSteps: 3 }
|
|
});
|
|
|
|
currentContent = step1Result.content;
|
|
pipelineStats.techniques.push({
|
|
name: 'SentenceVariation',
|
|
...step1Result.stats,
|
|
qualityImpact: calculateQualityImpact(content, step1Result.content)
|
|
});
|
|
|
|
logSh(` ✅ 1/3: Variation phrases - ${step1Result.stats.modified}/${step1Result.stats.processed} éléments`, 'INFO');
|
|
}
|
|
|
|
// TECHNIQUE 2: SUPPRESSION EMPREINTES LLM
|
|
if (config.fingerprintRemoval) {
|
|
const step2Result = await removeLLMFingerprints({
|
|
content: currentContent,
|
|
config: config.fingerprintRemovalConfig,
|
|
context: { step: 2, totalSteps: 3 }
|
|
});
|
|
|
|
currentContent = step2Result.content;
|
|
pipelineStats.techniques.push({
|
|
name: 'FingerprintRemoval',
|
|
...step2Result.stats,
|
|
qualityImpact: calculateQualityImpact(content, step2Result.content)
|
|
});
|
|
|
|
logSh(` ✅ 2/3: Suppression empreintes - ${step2Result.stats.totalReplacements} remplacements`, 'INFO');
|
|
}
|
|
|
|
// TECHNIQUE 3: HUMANISATION TRANSITIONS
|
|
if (config.transitionHumanization) {
|
|
const step3Result = await humanizeTransitions({
|
|
content: currentContent,
|
|
config: config.transitionHumanizationConfig,
|
|
context: { step: 3, totalSteps: 3 }
|
|
});
|
|
|
|
currentContent = step3Result.content;
|
|
pipelineStats.techniques.push({
|
|
name: 'TransitionHumanization',
|
|
...step3Result.stats,
|
|
qualityImpact: calculateQualityImpact(content, step3Result.content)
|
|
});
|
|
|
|
logSh(` ✅ 3/3: Humanisation transitions - ${step3Result.stats.totalReplacements} améliorations`, 'INFO');
|
|
}
|
|
|
|
// POST-PROCESSING: Vérifications qualité
|
|
if (config.qualityPreservation || config.readabilityCheck) {
|
|
const qualityCheck = performQualityChecks(content, currentContent, config);
|
|
pipelineStats.qualityMetrics.final = qualityCheck;
|
|
|
|
// Rollback si qualité trop dégradée
|
|
if (qualityCheck.shouldRollback) {
|
|
logSh(`⚠️ ROLLBACK: Qualité dégradée, retour contenu original`, 'WARNING');
|
|
currentContent = content;
|
|
pipelineStats.rollback = true;
|
|
}
|
|
}
|
|
|
|
// RÉSULTATS FINAUX
|
|
const totalDuration = Date.now() - startTime;
|
|
pipelineStats.totalDuration = totalDuration;
|
|
|
|
const totalModifications = pipelineStats.techniques.reduce((sum, tech) => {
|
|
return sum + (tech.modified || tech.totalReplacements || 0);
|
|
}, 0);
|
|
|
|
const stats = {
|
|
level: 2,
|
|
technique: 'pattern_breaking',
|
|
processed: Object.keys(content).length,
|
|
totalModifications,
|
|
techniquesUsed: pipelineStats.techniques.length,
|
|
duration: totalDuration,
|
|
techniques: pipelineStats.techniques,
|
|
qualityPreserved: !pipelineStats.rollback,
|
|
rollback: pipelineStats.rollback || false
|
|
};
|
|
|
|
logSh(`🎯 NIVEAU 2 TERMINÉ: ${totalModifications} modifications sur ${stats.processed} éléments (${totalDuration}ms)`, 'INFO');
|
|
|
|
// Log détaillé par technique
|
|
pipelineStats.techniques.forEach(tech => {
|
|
const modificationsCount = tech.modified || tech.totalReplacements || 0;
|
|
logSh(` • ${tech.name}: ${modificationsCount} modifications (${tech.duration}ms)`, 'DEBUG');
|
|
});
|
|
|
|
await tracer.event('Pattern breaking terminé', stats);
|
|
|
|
return {
|
|
content: currentContent,
|
|
stats,
|
|
debug: {
|
|
level: 2,
|
|
technique: 'pattern_breaking',
|
|
config,
|
|
pipeline: pipelineStats,
|
|
qualityMetrics: pipelineStats.qualityMetrics
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const totalDuration = Date.now() - startTime;
|
|
logSh(`❌ NIVEAU 2 ÉCHOUÉ après ${totalDuration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: retourner contenu original
|
|
logSh(`🔄 Fallback: contenu original conservé`, 'WARNING');
|
|
|
|
await tracer.event('Pattern breaking échoué', {
|
|
error: error.message,
|
|
duration: totalDuration,
|
|
fallback: true
|
|
});
|
|
|
|
return {
|
|
content,
|
|
stats: {
|
|
level: 2,
|
|
technique: 'pattern_breaking',
|
|
processed: Object.keys(content).length,
|
|
totalModifications: 0,
|
|
duration: totalDuration,
|
|
error: error.message,
|
|
fallback: true
|
|
},
|
|
debug: { error: error.message, fallback: true }
|
|
};
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* MODE DIAGNOSTIC - Test individuel des techniques
|
|
*/
|
|
async function diagnosticPatternBreaking(content, csvData) {
|
|
logSh(`🔬 DIAGNOSTIC NIVEAU 2: Test individuel des techniques`, 'INFO');
|
|
|
|
const diagnostics = {
|
|
techniques: [],
|
|
errors: [],
|
|
performance: {},
|
|
recommendations: []
|
|
};
|
|
|
|
const techniques = [
|
|
{ name: 'SentenceVariation', func: applySentenceVariation },
|
|
{ name: 'FingerprintRemoval', func: removeLLMFingerprints },
|
|
{ name: 'TransitionHumanization', func: humanizeTransitions }
|
|
];
|
|
|
|
for (const technique of techniques) {
|
|
try {
|
|
const startTime = Date.now();
|
|
const result = await technique.func({
|
|
content,
|
|
config: { csvData },
|
|
context: { diagnostic: true }
|
|
});
|
|
|
|
diagnostics.techniques.push({
|
|
name: technique.name,
|
|
success: true,
|
|
duration: Date.now() - startTime,
|
|
stats: result.stats,
|
|
effectivenessScore: calculateEffectivenessScore(result.stats)
|
|
});
|
|
|
|
} catch (error) {
|
|
diagnostics.errors.push({
|
|
technique: technique.name,
|
|
error: error.message
|
|
});
|
|
diagnostics.techniques.push({
|
|
name: technique.name,
|
|
success: false,
|
|
error: error.message
|
|
});
|
|
}
|
|
}
|
|
|
|
// Générer recommandations
|
|
diagnostics.recommendations = generateRecommendations(diagnostics.techniques);
|
|
|
|
const successfulTechniques = diagnostics.techniques.filter(t => t.success);
|
|
diagnostics.performance.totalDuration = diagnostics.techniques.reduce((sum, t) => sum + (t.duration || 0), 0);
|
|
diagnostics.performance.successRate = Math.round((successfulTechniques.length / techniques.length) * 100);
|
|
|
|
logSh(`🔬 DIAGNOSTIC TERMINÉ: ${successfulTechniques.length}/${techniques.length} techniques opérationnelles`, 'INFO');
|
|
|
|
return diagnostics;
|
|
}
|
|
|
|
/**
|
|
* Analyser qualité du contenu
|
|
*/
|
|
function analyzeContentQuality(content) {
|
|
const allText = Object.values(content).join(' ');
|
|
const wordCount = allText.split(/\s+/).length;
|
|
const avgWordsPerElement = wordCount / Object.keys(content).length;
|
|
|
|
// Métrique de lisibilité approximative (Flesch simplifié)
|
|
const sentences = allText.split(/[.!?]+/).filter(s => s.trim().length > 5);
|
|
const avgWordsPerSentence = wordCount / Math.max(1, sentences.length);
|
|
const readabilityScore = Math.max(0, 100 - (avgWordsPerSentence * 1.5));
|
|
|
|
return {
|
|
wordCount,
|
|
elementCount: Object.keys(content).length,
|
|
avgWordsPerElement: Math.round(avgWordsPerElement),
|
|
avgWordsPerSentence: Math.round(avgWordsPerSentence),
|
|
readabilityScore: Math.round(readabilityScore),
|
|
sentenceCount: sentences.length
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Calculer impact qualité entre avant/après
|
|
*/
|
|
function calculateQualityImpact(originalContent, modifiedContent) {
|
|
const originalQuality = analyzeContentQuality(originalContent);
|
|
const modifiedQuality = analyzeContentQuality(modifiedContent);
|
|
|
|
const wordCountChange = ((modifiedQuality.wordCount - originalQuality.wordCount) / originalQuality.wordCount) * 100;
|
|
const readabilityChange = modifiedQuality.readabilityScore - originalQuality.readabilityScore;
|
|
|
|
return {
|
|
wordCountChange: Math.round(wordCountChange * 100) / 100,
|
|
readabilityChange: Math.round(readabilityChange),
|
|
severe: Math.abs(wordCountChange) > 10 || Math.abs(readabilityChange) > 15
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Effectuer vérifications qualité
|
|
*/
|
|
function performQualityChecks(originalContent, modifiedContent, config) {
|
|
const originalQuality = analyzeContentQuality(originalContent);
|
|
const modifiedQuality = analyzeContentQuality(modifiedContent);
|
|
|
|
const qualityThresholds = {
|
|
maxWordCountChange: 15, // % max changement nombre mots
|
|
minReadabilityScore: 50, // Score lisibilité minimum
|
|
maxReadabilityDrop: 20 // Baisse max lisibilité
|
|
};
|
|
|
|
const issues = [];
|
|
|
|
// Vérification nombre de mots
|
|
const wordCountChange = Math.abs(modifiedQuality.wordCount - originalQuality.wordCount) / originalQuality.wordCount * 100;
|
|
if (wordCountChange > qualityThresholds.maxWordCountChange) {
|
|
issues.push({
|
|
type: 'word_count_change',
|
|
severity: 'high',
|
|
change: wordCountChange,
|
|
threshold: qualityThresholds.maxWordCountChange
|
|
});
|
|
}
|
|
|
|
// Vérification lisibilité
|
|
if (modifiedQuality.readabilityScore < qualityThresholds.minReadabilityScore) {
|
|
issues.push({
|
|
type: 'low_readability',
|
|
severity: 'medium',
|
|
score: modifiedQuality.readabilityScore,
|
|
threshold: qualityThresholds.minReadabilityScore
|
|
});
|
|
}
|
|
|
|
const readabilityDrop = originalQuality.readabilityScore - modifiedQuality.readabilityScore;
|
|
if (readabilityDrop > qualityThresholds.maxReadabilityDrop) {
|
|
issues.push({
|
|
type: 'readability_drop',
|
|
severity: 'high',
|
|
drop: readabilityDrop,
|
|
threshold: qualityThresholds.maxReadabilityDrop
|
|
});
|
|
}
|
|
|
|
// Décision rollback
|
|
const highSeverityIssues = issues.filter(issue => issue.severity === 'high');
|
|
const shouldRollback = highSeverityIssues.length > 0 && config.qualityPreservation;
|
|
|
|
return {
|
|
originalQuality,
|
|
modifiedQuality,
|
|
issues,
|
|
shouldRollback,
|
|
qualityScore: calculateOverallQualityScore(issues, modifiedQuality)
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Calculer score de qualité global
|
|
*/
|
|
function calculateOverallQualityScore(issues, quality) {
|
|
let baseScore = 100;
|
|
|
|
issues.forEach(issue => {
|
|
const penalty = issue.severity === 'high' ? 30 : issue.severity === 'medium' ? 15 : 5;
|
|
baseScore -= penalty;
|
|
});
|
|
|
|
// Bonus pour bonne lisibilité
|
|
if (quality.readabilityScore > 70) baseScore += 10;
|
|
|
|
return Math.max(0, Math.min(100, baseScore));
|
|
}
|
|
|
|
/**
|
|
* Calculer score d'efficacité d'une technique
|
|
*/
|
|
function calculateEffectivenessScore(stats) {
|
|
if (!stats) return 0;
|
|
|
|
const modificationsCount = stats.modified || stats.totalReplacements || 0;
|
|
const processedCount = stats.processed || 1;
|
|
const modificationRate = (modificationsCount / processedCount) * 100;
|
|
|
|
// Score basé sur taux de modification et durée
|
|
const baseScore = Math.min(100, modificationRate * 2); // Max 50% modification = score 100
|
|
const durationPenalty = Math.max(0, (stats.duration - 1000) / 100); // Pénalité si > 1s
|
|
|
|
return Math.max(0, Math.round(baseScore - durationPenalty));
|
|
}
|
|
|
|
/**
|
|
* Générer recommandations basées sur diagnostic
|
|
*/
|
|
function generateRecommendations(techniqueResults) {
|
|
const recommendations = [];
|
|
|
|
techniqueResults.forEach(tech => {
|
|
if (!tech.success) {
|
|
recommendations.push({
|
|
type: 'error',
|
|
technique: tech.name,
|
|
message: `${tech.name} a échoué: ${tech.error}`,
|
|
action: 'Vérifier configuration et dépendances'
|
|
});
|
|
return;
|
|
}
|
|
|
|
const effectiveness = tech.effectivenessScore || 0;
|
|
|
|
if (effectiveness < 30) {
|
|
recommendations.push({
|
|
type: 'low_effectiveness',
|
|
technique: tech.name,
|
|
message: `${tech.name} peu efficace (score: ${effectiveness})`,
|
|
action: 'Augmenter intensité ou réviser configuration'
|
|
});
|
|
} else if (effectiveness > 80) {
|
|
recommendations.push({
|
|
type: 'high_effectiveness',
|
|
technique: tech.name,
|
|
message: `${tech.name} très efficace (score: ${effectiveness})`,
|
|
action: 'Configuration optimale'
|
|
});
|
|
}
|
|
|
|
if (tech.duration > 3000) {
|
|
recommendations.push({
|
|
type: 'performance',
|
|
technique: tech.name,
|
|
message: `${tech.name} lent (${tech.duration}ms)`,
|
|
action: 'Considérer réduction intensité ou optimisation'
|
|
});
|
|
}
|
|
});
|
|
|
|
return recommendations;
|
|
}
|
|
|
|
module.exports = {
|
|
applyPatternBreaking, // ← MAIN ENTRY POINT
|
|
diagnosticPatternBreaking, // ← Mode diagnostic
|
|
analyzeContentQuality,
|
|
performQualityChecks,
|
|
calculateQualityImpact,
|
|
calculateEffectivenessScore
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/ContentGeneration.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ORCHESTRATEUR GÉNÉRATION - ARCHITECTURE REFACTORISÉE
|
|
// Responsabilité: Coordonner les 4 étapes de génération
|
|
// ========================================
|
|
|
|
const { logSh } = require('./ErrorReporting');
|
|
const { tracer } = require('./trace');
|
|
|
|
// Import des 4 étapes séparées
|
|
const { generateInitialContent } = require('./generation/InitialGeneration');
|
|
const { enhanceTechnicalTerms } = require('./generation/TechnicalEnhancement');
|
|
const { enhanceTransitions } = require('./generation/TransitionEnhancement');
|
|
const { applyPersonalityStyle } = require('./generation/StyleEnhancement');
|
|
|
|
// Import Pattern Breaking (Niveau 2)
|
|
const { applyPatternBreaking } = require('./post-processing/PatternBreaking');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - GÉNÉRATION AVEC SELECTIVE ENHANCEMENT
|
|
* @param {Object} hierarchy - Hiérarchie des éléments extraits
|
|
* @param {Object} csvData - Données CSV avec personnalité
|
|
* @param {Object} options - Options de génération
|
|
* @returns {Object} - Contenu généré final
|
|
*/
|
|
async function generateWithContext(hierarchy, csvData, options = {}) {
|
|
return await tracer.run('ContentGeneration.generateWithContext()', async () => {
|
|
const startTime = Date.now();
|
|
|
|
const pipelineName = options.patternBreaking ? 'selective_enhancement_with_pattern_breaking' : 'selective_enhancement';
|
|
const totalSteps = options.patternBreaking ? 5 : 4;
|
|
|
|
await tracer.annotate({
|
|
pipeline: pipelineName,
|
|
elementsCount: Object.keys(hierarchy).length,
|
|
personality: csvData.personality?.nom,
|
|
mc0: csvData.mc0,
|
|
options,
|
|
totalSteps
|
|
});
|
|
|
|
logSh(`🚀 DÉBUT PIPELINE ${options.patternBreaking ? 'NIVEAU 2' : 'NIVEAU 1'}`, 'INFO');
|
|
logSh(` 🎭 Personnalité: ${csvData.personality?.nom} (${csvData.personality?.style})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(hierarchy).length} éléments à traiter`, 'INFO');
|
|
logSh(` 🔧 Options: ${JSON.stringify(options)}`, 'DEBUG');
|
|
|
|
try {
|
|
let pipelineResults = {
|
|
content: {},
|
|
stats: { stages: [], totalDuration: 0 },
|
|
debug: { pipeline: 'selective_enhancement', stages: [] }
|
|
};
|
|
|
|
// ÉTAPE 1: GÉNÉRATION INITIALE (Claude)
|
|
const step1Result = await generateInitialContent({
|
|
hierarchy,
|
|
csvData,
|
|
context: { step: 1, totalSteps, options }
|
|
});
|
|
|
|
pipelineResults.content = step1Result.content;
|
|
pipelineResults.stats.stages.push({ stage: 1, name: 'InitialGeneration', ...step1Result.stats });
|
|
pipelineResults.debug.stages.push(step1Result.debug);
|
|
|
|
// ÉTAPE 2: ENHANCEMENT TECHNIQUE (GPT-4) - Optionnel
|
|
if (!options.skipTechnical) {
|
|
const step2Result = await enhanceTechnicalTerms({
|
|
content: pipelineResults.content,
|
|
csvData,
|
|
context: { step: 2, totalSteps, options }
|
|
});
|
|
|
|
pipelineResults.content = step2Result.content;
|
|
pipelineResults.stats.stages.push({ stage: 2, name: 'TechnicalEnhancement', ...step2Result.stats });
|
|
pipelineResults.debug.stages.push(step2Result.debug);
|
|
} else {
|
|
logSh(`⏭️ ÉTAPE 2/4 IGNORÉE: Enhancement technique désactivé`, 'INFO');
|
|
}
|
|
|
|
// ÉTAPE 3: ENHANCEMENT TRANSITIONS (Gemini) - Optionnel
|
|
if (!options.skipTransitions) {
|
|
const step3Result = await enhanceTransitions({
|
|
content: pipelineResults.content,
|
|
csvData,
|
|
context: { step: 3, totalSteps, options }
|
|
});
|
|
|
|
pipelineResults.content = step3Result.content;
|
|
pipelineResults.stats.stages.push({ stage: 3, name: 'TransitionEnhancement', ...step3Result.stats });
|
|
pipelineResults.debug.stages.push(step3Result.debug);
|
|
} else {
|
|
logSh(`⏭️ ÉTAPE 3/4 IGNORÉE: Enhancement transitions désactivé`, 'INFO');
|
|
}
|
|
|
|
// ÉTAPE 4: ENHANCEMENT STYLE (Mistral) - Optionnel
|
|
if (!options.skipStyle) {
|
|
const step4Result = await applyPersonalityStyle({
|
|
content: pipelineResults.content,
|
|
csvData,
|
|
context: { step: 4, totalSteps, options }
|
|
});
|
|
|
|
pipelineResults.content = step4Result.content;
|
|
pipelineResults.stats.stages.push({ stage: 4, name: 'StyleEnhancement', ...step4Result.stats });
|
|
pipelineResults.debug.stages.push(step4Result.debug);
|
|
} else {
|
|
logSh(`⏭️ ÉTAPE 4/${totalSteps} IGNORÉE: Enhancement style désactivé`, 'INFO');
|
|
}
|
|
|
|
// ÉTAPE 5: PATTERN BREAKING (NIVEAU 2) - Optionnel
|
|
if (options.patternBreaking) {
|
|
const step5Result = await applyPatternBreaking({
|
|
content: pipelineResults.content,
|
|
csvData,
|
|
options: options.patternBreakingConfig || {}
|
|
});
|
|
|
|
pipelineResults.content = step5Result.content;
|
|
pipelineResults.stats.stages.push({ stage: 5, name: 'PatternBreaking', ...step5Result.stats });
|
|
pipelineResults.debug.stages.push(step5Result.debug);
|
|
} else if (totalSteps === 5) {
|
|
logSh(`⏭️ ÉTAPE 5/5 IGNORÉE: Pattern Breaking désactivé`, 'INFO');
|
|
}
|
|
|
|
// RÉSULTATS FINAUX
|
|
const totalDuration = Date.now() - startTime;
|
|
pipelineResults.stats.totalDuration = totalDuration;
|
|
|
|
const totalProcessed = pipelineResults.stats.stages.reduce((sum, stage) => sum + (stage.processed || 0), 0);
|
|
const totalEnhanced = pipelineResults.stats.stages.reduce((sum, stage) => sum + (stage.enhanced || 0), 0);
|
|
|
|
logSh(`✅ PIPELINE TERMINÉ: ${Object.keys(pipelineResults.content).length} éléments générés`, 'INFO');
|
|
logSh(` ⏱️ Durée totale: ${totalDuration}ms`, 'INFO');
|
|
logSh(` 📈 Enhancements: ${totalEnhanced} sur ${totalProcessed} éléments traités`, 'INFO');
|
|
|
|
// Log détaillé par étape
|
|
pipelineResults.stats.stages.forEach(stage => {
|
|
const enhancementRate = stage.processed > 0 ? Math.round((stage.enhanced / stage.processed) * 100) : 0;
|
|
logSh(` ${stage.stage}. ${stage.name}: ${stage.enhanced}/${stage.processed} (${enhancementRate}%) en ${stage.duration}ms`, 'DEBUG');
|
|
});
|
|
|
|
await tracer.event(`Pipeline ${pipelineName} terminé`, {
|
|
totalElements: Object.keys(pipelineResults.content).length,
|
|
totalEnhanced,
|
|
totalDuration,
|
|
stagesExecuted: pipelineResults.stats.stages.length
|
|
});
|
|
|
|
// Retourner uniquement le contenu pour compatibilité
|
|
return pipelineResults.content;
|
|
|
|
} catch (error) {
|
|
const totalDuration = Date.now() - startTime;
|
|
logSh(`❌ PIPELINE ÉCHOUÉ après ${totalDuration}ms: ${error.message}`, 'ERROR');
|
|
logSh(`❌ Stack trace: ${error.stack}`, 'DEBUG');
|
|
|
|
await tracer.event(`Pipeline ${pipelineName} échoué`, {
|
|
error: error.message,
|
|
duration: totalDuration
|
|
});
|
|
|
|
throw new Error(`ContentGeneration pipeline failed: ${error.message}`);
|
|
}
|
|
}, { hierarchy, csvData, options });
|
|
}
|
|
|
|
/**
|
|
* GÉNÉRATION SIMPLE (ÉTAPE 1 UNIQUEMENT)
|
|
* Pour tests ou fallback rapide
|
|
*/
|
|
async function generateSimple(hierarchy, csvData) {
|
|
logSh(`🔥 GÉNÉRATION SIMPLE: Claude uniquement`, 'INFO');
|
|
|
|
const result = await generateInitialContent({
|
|
hierarchy,
|
|
csvData,
|
|
context: { step: 1, totalSteps: 1, simple: true }
|
|
});
|
|
|
|
return result.content;
|
|
}
|
|
|
|
/**
|
|
* GÉNÉRATION AVANCÉE AVEC CONTRÔLE GRANULAIRE
|
|
* Permet de choisir exactement quelles étapes exécuter
|
|
*/
|
|
async function generateAdvanced(hierarchy, csvData, stageConfig = {}) {
|
|
const {
|
|
initial = true,
|
|
technical = true,
|
|
transitions = true,
|
|
style = true,
|
|
patternBreaking = false, // ✨ NOUVEAU: Niveau 2
|
|
patternBreakingConfig = {} // ✨ NOUVEAU: Config Pattern Breaking
|
|
} = stageConfig;
|
|
|
|
const options = {
|
|
skipTechnical: !technical,
|
|
skipTransitions: !transitions,
|
|
skipStyle: !style,
|
|
patternBreaking, // ✨ NOUVEAU
|
|
patternBreakingConfig // ✨ NOUVEAU
|
|
};
|
|
|
|
const activeStages = [
|
|
initial && 'Initial',
|
|
technical && 'Technical',
|
|
transitions && 'Transitions',
|
|
style && 'Style',
|
|
patternBreaking && 'PatternBreaking' // ✨ NOUVEAU
|
|
].filter(Boolean);
|
|
|
|
logSh(`🎛️ GÉNÉRATION AVANCÉE: ${activeStages.join(' + ')}`, 'INFO');
|
|
|
|
return await generateWithContext(hierarchy, csvData, options);
|
|
}
|
|
|
|
/**
|
|
* GÉNÉRATION NIVEAU 2 (AVEC PATTERN BREAKING)
|
|
* Shortcut pour activer Pattern Breaking facilement
|
|
*/
|
|
async function generateWithPatternBreaking(hierarchy, csvData, patternConfig = {}) {
|
|
logSh(`🎯 GÉNÉRATION NIVEAU 2: Pattern Breaking activé`, 'INFO');
|
|
|
|
const options = {
|
|
patternBreaking: true,
|
|
patternBreakingConfig: {
|
|
intensity: 0.6,
|
|
sentenceVariation: true,
|
|
fingerprintRemoval: true,
|
|
transitionHumanization: true,
|
|
...patternConfig
|
|
}
|
|
};
|
|
|
|
return await generateWithContext(hierarchy, csvData, options);
|
|
}
|
|
|
|
/**
|
|
* DIAGNOSTIC PIPELINE
|
|
* Exécute chaque étape avec mesures détaillées
|
|
*/
|
|
async function diagnosticPipeline(hierarchy, csvData) {
|
|
logSh(`🔬 MODE DIAGNOSTIC: Analyse détaillée pipeline`, 'INFO');
|
|
|
|
const diagnostics = {
|
|
stages: [],
|
|
errors: [],
|
|
performance: {},
|
|
content: {}
|
|
};
|
|
|
|
let currentContent = {};
|
|
|
|
try {
|
|
// Test étape 1
|
|
const step1Start = Date.now();
|
|
const step1Result = await generateInitialContent({ hierarchy, csvData });
|
|
diagnostics.stages.push({
|
|
stage: 1,
|
|
name: 'InitialGeneration',
|
|
success: true,
|
|
duration: Date.now() - step1Start,
|
|
elementsGenerated: Object.keys(step1Result.content).length,
|
|
stats: step1Result.stats
|
|
});
|
|
currentContent = step1Result.content;
|
|
|
|
} catch (error) {
|
|
diagnostics.errors.push({ stage: 1, error: error.message });
|
|
diagnostics.stages.push({ stage: 1, name: 'InitialGeneration', success: false });
|
|
return diagnostics;
|
|
}
|
|
|
|
// Test étapes 2-4 individuellement
|
|
const stages = [
|
|
{ stage: 2, name: 'TechnicalEnhancement', func: enhanceTechnicalTerms },
|
|
{ stage: 3, name: 'TransitionEnhancement', func: enhanceTransitions },
|
|
{ stage: 4, name: 'StyleEnhancement', func: applyPersonalityStyle }
|
|
];
|
|
|
|
for (const stageInfo of stages) {
|
|
try {
|
|
const stageStart = Date.now();
|
|
const stageResult = await stageInfo.func({ content: currentContent, csvData });
|
|
|
|
diagnostics.stages.push({
|
|
...stageInfo,
|
|
success: true,
|
|
duration: Date.now() - stageStart,
|
|
stats: stageResult.stats
|
|
});
|
|
|
|
currentContent = stageResult.content;
|
|
|
|
} catch (error) {
|
|
diagnostics.errors.push({ stage: stageInfo.stage, error: error.message });
|
|
diagnostics.stages.push({ ...stageInfo, success: false });
|
|
}
|
|
}
|
|
|
|
diagnostics.content = currentContent;
|
|
diagnostics.performance.totalDuration = diagnostics.stages.reduce((sum, stage) => sum + (stage.duration || 0), 0);
|
|
|
|
logSh(`🔬 DIAGNOSTIC TERMINÉ: ${diagnostics.stages.filter(s => s.success).length}/4 étapes réussies`, 'INFO');
|
|
|
|
return diagnostics;
|
|
}
|
|
|
|
module.exports = {
|
|
generateWithContext, // ← MAIN ENTRY POINT (compatible ancien code)
|
|
generateSimple, // ← Génération rapide
|
|
generateAdvanced, // ← Contrôle granulaire
|
|
generateWithPatternBreaking, // ← NOUVEAU: Niveau 2 shortcut
|
|
diagnosticPipeline // ← Tests et debug
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/ContentAssembly.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: ContentAssembly.js
|
|
// Description: Assemblage et nettoyage du contenu XML
|
|
// ========================================
|
|
|
|
const { logSh } = require('./ErrorReporting'); // Using unified logSh from ErrorReporting
|
|
|
|
/**
|
|
* Nettoie les balises <strong> du template XML
|
|
* @param {string} xmlString - Le contenu XML à nettoyer
|
|
* @returns {string} - XML nettoyé
|
|
*/
|
|
function cleanStrongTags(xmlString) {
|
|
logSh('Nettoyage balises <strong> du template...', 'DEBUG');
|
|
|
|
// Enlever toutes les balises <strong> et </strong>
|
|
let cleaned = xmlString.replace(/<\/?strong>/g, '');
|
|
|
|
// Log du nettoyage
|
|
const strongCount = (xmlString.match(/<\/?strong>/g) || []).length;
|
|
if (strongCount > 0) {
|
|
logSh(`${strongCount} balises <strong> supprimées`, 'INFO');
|
|
}
|
|
|
|
return cleaned;
|
|
}
|
|
|
|
/**
|
|
* Remplace toutes les variables CSV dans le XML
|
|
* @param {string} xmlString - Le contenu XML
|
|
* @param {object} csvData - Les données CSV
|
|
* @returns {string} - XML avec variables remplacées
|
|
*/
|
|
function replaceAllCSVVariables(xmlString, csvData) {
|
|
logSh('Remplacement variables CSV...', 'DEBUG');
|
|
|
|
let result = xmlString;
|
|
|
|
// Variables simples
|
|
result = result.replace(/\{\{T0\}\}/g, csvData.t0 || '');
|
|
result = result.replace(/\{\{MC0\}\}/g, csvData.mc0 || '');
|
|
result = result.replace(/\{\{T-1\}\}/g, csvData.tMinus1 || '');
|
|
result = result.replace(/\{\{L-1\}\}/g, csvData.lMinus1 || '');
|
|
|
|
logSh(`Variables simples remplacées: T0="${csvData.t0}", MC0="${csvData.mc0}"`, 'DEBUG');
|
|
|
|
// Variables multiples
|
|
const mcPlus1 = (csvData.mcPlus1 || '').split(',').map(s => s.trim());
|
|
const tPlus1 = (csvData.tPlus1 || '').split(',').map(s => s.trim());
|
|
const lPlus1 = (csvData.lPlus1 || '').split(',').map(s => s.trim());
|
|
|
|
logSh(`Variables multiples: MC+1[${mcPlus1.length}], T+1[${tPlus1.length}], L+1[${lPlus1.length}]`, 'DEBUG');
|
|
|
|
// Remplacer MC+1_1, MC+1_2, etc.
|
|
for (let i = 1; i <= 6; i++) {
|
|
const mcValue = mcPlus1[i-1] || `[MC+1_${i} non défini]`;
|
|
const tValue = tPlus1[i-1] || `[T+1_${i} non défini]`;
|
|
const lValue = lPlus1[i-1] || `[L+1_${i} non défini]`;
|
|
|
|
result = result.replace(new RegExp(`\\{\\{MC\\+1_${i}\\}\\}`, 'g'), mcValue);
|
|
result = result.replace(new RegExp(`\\{\\{T\\+1_${i}\\}\\}`, 'g'), tValue);
|
|
result = result.replace(new RegExp(`\\{\\{L\\+1_${i}\\}\\}`, 'g'), lValue);
|
|
|
|
if (mcPlus1[i-1]) {
|
|
logSh(`MC+1_${i} = "${mcValue}"`, 'DEBUG');
|
|
}
|
|
}
|
|
|
|
// Vérifier qu'il ne reste pas de variables non remplacées
|
|
const remainingVars = (result.match(/\{\{[^}]+\}\}/g) || []);
|
|
if (remainingVars.length > 0) {
|
|
logSh(`ATTENTION: Variables non remplacées: ${remainingVars.join(', ')}`, 'WARNING');
|
|
}
|
|
|
|
logSh('Toutes les variables CSV remplacées', 'INFO');
|
|
return result;
|
|
}
|
|
|
|
/**
|
|
* Injecte le contenu généré dans le XML final
|
|
* @param {string} cleanXML - XML nettoyé
|
|
* @param {object} generatedContent - Contenu généré par tag
|
|
* @param {array} elements - Éléments extraits
|
|
* @returns {string} - XML final avec contenu injecté
|
|
*/
|
|
function injectGeneratedContent(cleanXML, generatedContent, elements) {
|
|
logSh('🔍 === DEBUG INJECTION MAPPING ===', 'DEBUG');
|
|
logSh(`XML reçu: ${cleanXML.length} caractères`, 'DEBUG');
|
|
logSh(`Contenu généré: ${Object.keys(generatedContent).length} éléments`, 'DEBUG');
|
|
logSh(`Éléments fournis: ${elements.length} éléments`, 'DEBUG');
|
|
|
|
// Debug: montrer le XML
|
|
logSh(`🔍 XML début: ${cleanXML}`, 'DEBUG');
|
|
|
|
// Debug: montrer le contenu généré
|
|
Object.keys(generatedContent).forEach(key => {
|
|
logSh(`🔍 Généré [${key}]: "${generatedContent[key]}"`, 'DEBUG');
|
|
});
|
|
|
|
// Debug: montrer les éléments
|
|
elements.forEach((element, i) => {
|
|
logSh(`🔍 Element ${i+1}: originalTag="${element.originalTag}", originalFullMatch="${element.originalFullMatch}"`, 'DEBUG');
|
|
});
|
|
|
|
let finalXML = cleanXML;
|
|
|
|
// Créer un mapping tag pur → tag original complet
|
|
const tagMapping = {};
|
|
elements.forEach(element => {
|
|
tagMapping[element.originalTag] = element.originalFullMatch || element.originalTag;
|
|
});
|
|
|
|
logSh(`🔍 TagMapping créé: ${JSON.stringify(tagMapping, null, 2)}`, 'DEBUG');
|
|
|
|
// Remplacer en utilisant les tags originaux complets
|
|
Object.keys(generatedContent).forEach(pureTag => {
|
|
const content = generatedContent[pureTag];
|
|
|
|
logSh(`🔍 === TRAITEMENT TAG: ${pureTag} ===`, 'DEBUG');
|
|
logSh(`🔍 Contenu à injecter: "${content}"`, 'DEBUG');
|
|
|
|
// Trouver le tag original complet dans le XML
|
|
const originalTag = findOriginalTagInXML(finalXML, pureTag);
|
|
|
|
logSh(`🔍 Tag original trouvé: ${originalTag ? originalTag : 'AUCUN'}`, 'DEBUG');
|
|
|
|
if (originalTag) {
|
|
const beforeLength = finalXML.length;
|
|
finalXML = finalXML.replace(originalTag, content);
|
|
const afterLength = finalXML.length;
|
|
|
|
if (beforeLength !== afterLength) {
|
|
logSh(`✅ SUCCÈS: Remplacé ${originalTag} par contenu (${afterLength - beforeLength + originalTag.length} chars)`, 'DEBUG');
|
|
} else {
|
|
logSh(`❌ ÉCHEC: Replace n'a pas fonctionné pour ${originalTag}`, 'DEBUG');
|
|
}
|
|
} else {
|
|
// Fallback : essayer avec le tag pur
|
|
const beforeLength = finalXML.length;
|
|
finalXML = finalXML.replace(pureTag, content);
|
|
const afterLength = finalXML.length;
|
|
|
|
logSh(`⚠ FALLBACK ${pureTag}: remplacement ${beforeLength !== afterLength ? 'RÉUSSI' : 'ÉCHOUÉ'}`, 'DEBUG');
|
|
logSh(`⚠ Contenu fallback: "${content}"`, 'DEBUG');
|
|
}
|
|
});
|
|
|
|
// Vérifier les tags restants
|
|
const remainingTags = (finalXML.match(/\|[^|]*\|/g) || []);
|
|
if (remainingTags.length > 0) {
|
|
logSh(`ATTENTION: ${remainingTags.length} tags non remplacés: ${remainingTags.slice(0, 3).join(', ')}...`, 'WARNING');
|
|
}
|
|
|
|
logSh('Injection terminée', 'INFO');
|
|
return finalXML;
|
|
}
|
|
|
|
/**
|
|
* Helper pour trouver le tag original complet dans le XML
|
|
* @param {string} xmlString - Contenu XML
|
|
* @param {string} pureTag - Tag pur à rechercher
|
|
* @returns {string|null} - Tag original trouvé ou null
|
|
*/
|
|
function findOriginalTagInXML(xmlString, pureTag) {
|
|
logSh(`🔍 === RECHERCHE TAG DANS XML ===`, 'DEBUG');
|
|
logSh(`🔍 Tag pur recherché: "${pureTag}"`, 'DEBUG');
|
|
|
|
// Extraire le nom du tag pur : |Titre_H1_1| → Titre_H1_1
|
|
const tagName = pureTag.replace(/\|/g, '');
|
|
logSh(`🔍 Nom tag extrait: "${tagName}"`, 'DEBUG');
|
|
|
|
// Chercher tous les tags qui commencent par ce nom (avec espaces optionnels)
|
|
const regex = new RegExp(`\\|\\s*${tagName}[^|]*\\|`, 'g');
|
|
logSh(`🔍 Regex utilisée: ${regex}`, 'DEBUG');
|
|
|
|
// Debug: montrer tous les tags présents dans le XML
|
|
const allTags = xmlString.match(/\|[^|]*\|/g) || [];
|
|
logSh(`🔍 Tags présents dans XML: ${allTags.length}`, 'DEBUG');
|
|
allTags.forEach((tag, i) => {
|
|
logSh(`🔍 ${i+1}. "${tag}"`, 'DEBUG');
|
|
});
|
|
|
|
const matches = xmlString.match(regex);
|
|
logSh(`🔍 Matches trouvés: ${matches ? matches.length : 0}`, 'DEBUG');
|
|
|
|
if (matches && matches.length > 0) {
|
|
logSh(`🔍 Premier match: "${matches[0]}"`, 'DEBUG');
|
|
logSh(`✅ Tag original trouvé pour ${pureTag}: ${matches[0]}`, 'DEBUG');
|
|
return matches[0];
|
|
}
|
|
|
|
logSh(`❌ Aucun tag original trouvé pour ${pureTag}`, 'DEBUG');
|
|
return null;
|
|
}
|
|
|
|
// ============= EXPORTS =============
|
|
module.exports = {
|
|
cleanStrongTags,
|
|
replaceAllCSVVariables,
|
|
injectGeneratedContent,
|
|
findOriginalTagInXML
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/ArticleStorage.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: ArticleStorage.js
|
|
// Description: Système de sauvegarde articles avec texte compilé uniquement
|
|
// ========================================
|
|
|
|
require('dotenv').config();
|
|
const { google } = require('googleapis');
|
|
const { logSh } = require('./ErrorReporting');
|
|
|
|
// Configuration Google Sheets
|
|
const SHEET_CONFIG = {
|
|
sheetId: '1iA2GvWeUxX-vpnAMfVm3ZMG9LhaC070SdGssEcXAh2c'
|
|
};
|
|
|
|
/**
|
|
* NOUVELLE FONCTION : Compiler le contenu de manière organique
|
|
* Respecte la hiérarchie et les associations naturelles
|
|
*/
|
|
async function compileGeneratedTextsOrganic(generatedTexts, elements) {
|
|
if (!generatedTexts || Object.keys(generatedTexts).length === 0) {
|
|
return '';
|
|
}
|
|
|
|
logSh(`🌱 Compilation ORGANIQUE de ${Object.keys(generatedTexts).length} éléments...`, 'DEBUG');
|
|
|
|
let compiledParts = [];
|
|
|
|
// 1. DÉTECTER et GROUPER les sections organiques
|
|
const organicSections = buildOrganicSections(generatedTexts, elements);
|
|
|
|
// 2. COMPILER dans l'ordre naturel
|
|
organicSections.forEach(section => {
|
|
if (section.type === 'header_with_content') {
|
|
// H1, H2, H3 avec leur contenu associé
|
|
if (section.title) {
|
|
compiledParts.push(cleanIndividualContent(section.title));
|
|
}
|
|
if (section.content) {
|
|
compiledParts.push(cleanIndividualContent(section.content));
|
|
}
|
|
}
|
|
else if (section.type === 'standalone_content') {
|
|
// Contenu sans titre associé
|
|
compiledParts.push(cleanIndividualContent(section.content));
|
|
}
|
|
else if (section.type === 'faq_pair') {
|
|
// Paire question + réponse
|
|
if (section.question && section.answer) {
|
|
compiledParts.push(cleanIndividualContent(section.question));
|
|
compiledParts.push(cleanIndividualContent(section.answer));
|
|
}
|
|
}
|
|
});
|
|
|
|
// 3. Joindre avec espacement naturel
|
|
const finalText = compiledParts.join('\n\n');
|
|
|
|
logSh(`✅ Compilation organique terminée: ${finalText.length} caractères`, 'INFO');
|
|
return finalText.trim();
|
|
}
|
|
|
|
/**
|
|
* Construire les sections organiques en analysant les associations
|
|
*/
|
|
function buildOrganicSections(generatedTexts, elements) {
|
|
const sections = [];
|
|
const usedTags = new Set();
|
|
|
|
// 1. ANALYSER l'ordre original des éléments
|
|
const originalOrder = elements ? elements.map(el => el.originalTag) : Object.keys(generatedTexts);
|
|
|
|
logSh(`📋 Analyse de ${originalOrder.length} éléments dans l'ordre original...`, 'DEBUG');
|
|
|
|
// 2. DÉTECTER les associations naturelles
|
|
for (let i = 0; i < originalOrder.length; i++) {
|
|
const currentTag = originalOrder[i];
|
|
const currentContent = generatedTexts[currentTag];
|
|
|
|
if (!currentContent || usedTags.has(currentTag)) continue;
|
|
|
|
const currentType = identifyElementType(currentTag);
|
|
|
|
if (currentType === 'titre_h1' || currentType === 'titre_h2' || currentType === 'titre_h3') {
|
|
// CHERCHER le contenu associé qui suit
|
|
const associatedContent = findAssociatedContent(originalOrder, i, generatedTexts, usedTags);
|
|
|
|
sections.push({
|
|
type: 'header_with_content',
|
|
title: currentContent,
|
|
content: associatedContent.content,
|
|
titleTag: currentTag,
|
|
contentTag: associatedContent.tag
|
|
});
|
|
|
|
usedTags.add(currentTag);
|
|
if (associatedContent.tag) {
|
|
usedTags.add(associatedContent.tag);
|
|
}
|
|
|
|
logSh(` ✓ Section: ${currentType} + contenu associé`, 'DEBUG');
|
|
}
|
|
else if (currentType === 'faq_question') {
|
|
// CHERCHER la réponse correspondante
|
|
const matchingAnswer = findMatchingFAQAnswer(currentTag, generatedTexts);
|
|
|
|
if (matchingAnswer) {
|
|
sections.push({
|
|
type: 'faq_pair',
|
|
question: currentContent,
|
|
answer: matchingAnswer.content,
|
|
questionTag: currentTag,
|
|
answerTag: matchingAnswer.tag
|
|
});
|
|
|
|
usedTags.add(currentTag);
|
|
usedTags.add(matchingAnswer.tag);
|
|
|
|
logSh(` ✓ Paire FAQ: ${currentTag} + ${matchingAnswer.tag}`, 'DEBUG');
|
|
}
|
|
}
|
|
else if (currentType !== 'faq_reponse') {
|
|
// CONTENU STANDALONE (pas une réponse FAQ déjà traitée)
|
|
sections.push({
|
|
type: 'standalone_content',
|
|
content: currentContent,
|
|
contentTag: currentTag
|
|
});
|
|
|
|
usedTags.add(currentTag);
|
|
logSh(` ✓ Contenu standalone: ${currentType}`, 'DEBUG');
|
|
}
|
|
}
|
|
|
|
logSh(`🏗️ ${sections.length} sections organiques construites`, 'INFO');
|
|
return sections;
|
|
}
|
|
|
|
/**
|
|
* Trouver le contenu associé à un titre (paragraphe qui suit)
|
|
*/
|
|
function findAssociatedContent(originalOrder, titleIndex, generatedTexts, usedTags) {
|
|
// Chercher dans les éléments suivants
|
|
for (let j = titleIndex + 1; j < originalOrder.length; j++) {
|
|
const nextTag = originalOrder[j];
|
|
const nextContent = generatedTexts[nextTag];
|
|
|
|
if (!nextContent || usedTags.has(nextTag)) continue;
|
|
|
|
const nextType = identifyElementType(nextTag);
|
|
|
|
// Si on trouve un autre titre, on s'arrête
|
|
if (nextType === 'titre_h1' || nextType === 'titre_h2' || nextType === 'titre_h3') {
|
|
break;
|
|
}
|
|
|
|
// Si on trouve du contenu (texte, intro), c'est probablement associé
|
|
if (nextType === 'texte' || nextType === 'intro') {
|
|
return {
|
|
content: nextContent,
|
|
tag: nextTag
|
|
};
|
|
}
|
|
}
|
|
|
|
return { content: null, tag: null };
|
|
}
|
|
|
|
/**
|
|
* Extraire le numéro d'une FAQ : |Faq_q_1| ou |Faq_a_2| → "1" ou "2"
|
|
*/
|
|
function extractFAQNumber(tag) {
|
|
const match = tag.match(/(\d+)/);
|
|
return match ? match[1] : null;
|
|
}
|
|
|
|
/**
|
|
* Trouver la réponse FAQ correspondant à une question
|
|
*/
|
|
function findMatchingFAQAnswer(questionTag, generatedTexts) {
|
|
// Extraire le numéro : |Faq_q_1| → 1
|
|
const questionNumber = extractFAQNumber(questionTag);
|
|
|
|
if (!questionNumber) return null;
|
|
|
|
// Chercher la réponse correspondante
|
|
for (const tag in generatedTexts) {
|
|
const tagType = identifyElementType(tag);
|
|
|
|
if (tagType === 'faq_reponse') {
|
|
const answerNumber = extractFAQNumber(tag);
|
|
|
|
if (answerNumber === questionNumber) {
|
|
return {
|
|
content: generatedTexts[tag],
|
|
tag: tag
|
|
};
|
|
}
|
|
}
|
|
}
|
|
|
|
return null;
|
|
}
|
|
|
|
/**
|
|
* Nouvelle fonction de sauvegarde avec compilation organique
|
|
*/
|
|
async function saveGeneratedArticleOrganic(articleData, csvData, config = {}) {
|
|
try {
|
|
logSh('💾 Sauvegarde article avec compilation organique...', 'INFO');
|
|
|
|
const sheets = await getSheetsClient();
|
|
|
|
// Vérifier si la sheet existe, sinon la créer
|
|
let articlesSheet = await getOrCreateSheet(sheets, 'Generated_Articles');
|
|
|
|
// ===== COMPILATION ORGANIQUE =====
|
|
const compiledText = await compileGeneratedTextsOrganic(
|
|
articleData.generatedTexts,
|
|
articleData.originalElements // Passer les éléments originaux si disponibles
|
|
);
|
|
|
|
logSh(`📝 Texte compilé organiquement: ${compiledText.length} caractères`, 'INFO');
|
|
|
|
// Métadonnées avec format français
|
|
const now = new Date();
|
|
const frenchTimestamp = formatDateToFrench(now);
|
|
|
|
// UTILISER le slug du CSV (colonne A du Google Sheet source)
|
|
// Le slug doit venir de csvData.slug (récupéré via getBrainConfig)
|
|
const slug = csvData.slug || generateSlugFromContent(csvData.mc0, csvData.t0);
|
|
|
|
const metadata = {
|
|
timestamp: frenchTimestamp,
|
|
slug: slug,
|
|
mc0: csvData.mc0,
|
|
t0: csvData.t0,
|
|
personality: csvData.personality?.nom || 'Unknown',
|
|
antiDetectionLevel: config.antiDetectionLevel || 'MVP',
|
|
elementsCount: Object.keys(articleData.generatedTexts || {}).length,
|
|
textLength: compiledText.length,
|
|
wordCount: countWords(compiledText),
|
|
llmUsed: config.llmUsed || 'openai',
|
|
validationStatus: articleData.validationReport?.status || 'unknown'
|
|
};
|
|
|
|
// Préparer la ligne de données
|
|
const row = [
|
|
metadata.timestamp,
|
|
metadata.slug,
|
|
metadata.mc0,
|
|
metadata.t0,
|
|
metadata.personality,
|
|
metadata.antiDetectionLevel,
|
|
compiledText, // ← TEXTE ORGANIQUE
|
|
metadata.textLength,
|
|
metadata.wordCount,
|
|
metadata.elementsCount,
|
|
metadata.llmUsed,
|
|
metadata.validationStatus,
|
|
'', '', '', '',
|
|
JSON.stringify({
|
|
csvData: csvData,
|
|
config: config,
|
|
stats: metadata
|
|
})
|
|
];
|
|
|
|
// DEBUG: Vérifier le slug généré
|
|
logSh(`💾 Sauvegarde avec slug: "${metadata.slug}" (colonne B)`, 'DEBUG');
|
|
|
|
// Ajouter la ligne aux données
|
|
await sheets.spreadsheets.values.append({
|
|
spreadsheetId: SHEET_CONFIG.sheetId,
|
|
range: 'Generated_Articles!A:Q',
|
|
valueInputOption: 'USER_ENTERED',
|
|
resource: {
|
|
values: [row]
|
|
}
|
|
});
|
|
|
|
// Récupérer le numéro de ligne pour l'ID article
|
|
const response = await sheets.spreadsheets.values.get({
|
|
spreadsheetId: SHEET_CONFIG.sheetId,
|
|
range: 'Generated_Articles!A:A'
|
|
});
|
|
|
|
const articleId = response.data.values ? response.data.values.length - 1 : 1;
|
|
|
|
logSh(`✅ Article organique sauvé: ID ${articleId}, ${metadata.wordCount} mots`, 'INFO');
|
|
|
|
return {
|
|
articleId: articleId,
|
|
textLength: metadata.textLength,
|
|
wordCount: metadata.wordCount,
|
|
sheetRow: response.data.values ? response.data.values.length : 2
|
|
};
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur sauvegarde organique: ${error.toString()}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Générer un slug à partir du contenu MC0 et T0
|
|
*/
|
|
function generateSlugFromContent(mc0, t0) {
|
|
if (!mc0 && !t0) return 'article-generated';
|
|
|
|
const source = mc0 || t0;
|
|
return source
|
|
.toString()
|
|
.toLowerCase()
|
|
.replace(/[àáâäã]/g, 'a')
|
|
.replace(/[èéêë]/g, 'e')
|
|
.replace(/[ìíîï]/g, 'i')
|
|
.replace(/[òóôöõ]/g, 'o')
|
|
.replace(/[ùúûü]/g, 'u')
|
|
.replace(/[ç]/g, 'c')
|
|
.replace(/[ñ]/g, 'n')
|
|
.replace(/[^a-z0-9\s-]/g, '') // Enlever caractères spéciaux
|
|
.replace(/\s+/g, '-') // Espaces -> tirets
|
|
.replace(/-+/g, '-') // Éviter doubles tirets
|
|
.replace(/^-+|-+$/g, '') // Enlever tirets début/fin
|
|
.substring(0, 50); // Limiter longueur
|
|
}
|
|
|
|
/**
|
|
* Identifier le type d'élément par son tag
|
|
*/
|
|
function identifyElementType(tag) {
|
|
const cleanTag = tag.toLowerCase().replace(/[|{}]/g, '');
|
|
|
|
if (cleanTag.includes('titre_h1') || cleanTag.includes('h1')) return 'titre_h1';
|
|
if (cleanTag.includes('titre_h2') || cleanTag.includes('h2')) return 'titre_h2';
|
|
if (cleanTag.includes('titre_h3') || cleanTag.includes('h3')) return 'titre_h3';
|
|
if (cleanTag.includes('intro')) return 'intro';
|
|
if (cleanTag.includes('faq_q') || cleanTag.includes('faq_question')) return 'faq_question';
|
|
if (cleanTag.includes('faq_a') || cleanTag.includes('faq_reponse')) return 'faq_reponse';
|
|
|
|
return 'texte'; // Par défaut
|
|
}
|
|
|
|
/**
|
|
* Nettoyer un contenu individuel
|
|
*/
|
|
function cleanIndividualContent(content) {
|
|
if (!content) return '';
|
|
|
|
let cleaned = content.toString();
|
|
|
|
// 1. Supprimer les balises HTML
|
|
cleaned = cleaned.replace(/<[^>]*>/g, '');
|
|
|
|
// 2. Décoder les entités HTML
|
|
cleaned = cleaned.replace(/</g, '<');
|
|
cleaned = cleaned.replace(/>/g, '>');
|
|
cleaned = cleaned.replace(/&/g, '&');
|
|
cleaned = cleaned.replace(/"/g, '"');
|
|
cleaned = cleaned.replace(/'/g, "'");
|
|
cleaned = cleaned.replace(/ /g, ' ');
|
|
|
|
// 3. Nettoyer les espaces
|
|
cleaned = cleaned.replace(/\s+/g, ' ');
|
|
cleaned = cleaned.replace(/\n\s+/g, '\n');
|
|
|
|
// 4. Supprimer les caractères de contrôle étranges
|
|
cleaned = cleaned.replace(/[\x00-\x1F\x7F-\x9F]/g, '');
|
|
|
|
return cleaned.trim();
|
|
}
|
|
|
|
/**
|
|
* Créer la sheet de stockage avec headers appropriés
|
|
*/
|
|
async function createArticlesStorageSheet(sheets) {
|
|
logSh('🗄️ Création sheet Generated_Articles...', 'INFO');
|
|
|
|
try {
|
|
// Créer la nouvelle sheet
|
|
await sheets.spreadsheets.batchUpdate({
|
|
spreadsheetId: SHEET_CONFIG.sheetId,
|
|
resource: {
|
|
requests: [{
|
|
addSheet: {
|
|
properties: {
|
|
title: 'Generated_Articles'
|
|
}
|
|
}
|
|
}]
|
|
}
|
|
});
|
|
|
|
// Headers
|
|
const headers = [
|
|
'Timestamp',
|
|
'Slug',
|
|
'MC0',
|
|
'T0',
|
|
'Personality',
|
|
'AntiDetection_Level',
|
|
'Compiled_Text', // ← COLONNE PRINCIPALE
|
|
'Text_Length',
|
|
'Word_Count',
|
|
'Elements_Count',
|
|
'LLM_Used',
|
|
'Validation_Status',
|
|
'GPTZero_Score', // Scores détecteurs (à remplir)
|
|
'Originality_Score',
|
|
'CopyLeaks_Score',
|
|
'Human_Quality_Score',
|
|
'Full_Metadata_JSON' // Backup complet
|
|
];
|
|
|
|
// Ajouter les headers
|
|
await sheets.spreadsheets.values.update({
|
|
spreadsheetId: SHEET_CONFIG.sheetId,
|
|
range: 'Generated_Articles!A1:Q1',
|
|
valueInputOption: 'USER_ENTERED',
|
|
resource: {
|
|
values: [headers]
|
|
}
|
|
});
|
|
|
|
// Formatter les headers
|
|
await sheets.spreadsheets.batchUpdate({
|
|
spreadsheetId: SHEET_CONFIG.sheetId,
|
|
resource: {
|
|
requests: [{
|
|
repeatCell: {
|
|
range: {
|
|
sheetId: await getSheetIdByName(sheets, 'Generated_Articles'),
|
|
startRowIndex: 0,
|
|
endRowIndex: 1,
|
|
startColumnIndex: 0,
|
|
endColumnIndex: headers.length
|
|
},
|
|
cell: {
|
|
userEnteredFormat: {
|
|
textFormat: {
|
|
bold: true
|
|
},
|
|
backgroundColor: {
|
|
red: 0.878,
|
|
green: 0.878,
|
|
blue: 0.878
|
|
},
|
|
horizontalAlignment: 'CENTER'
|
|
}
|
|
},
|
|
fields: 'userEnteredFormat(textFormat,backgroundColor,horizontalAlignment)'
|
|
}
|
|
}]
|
|
}
|
|
});
|
|
|
|
logSh('✅ Sheet Generated_Articles créée avec succès', 'INFO');
|
|
return true;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur création sheet: ${error.toString()}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Formater date au format français DD/MM/YYYY HH:mm:ss
|
|
*/
|
|
function formatDateToFrench(date) {
|
|
// Utiliser toLocaleString avec le format français
|
|
return date.toLocaleString('fr-FR', {
|
|
day: '2-digit',
|
|
month: '2-digit',
|
|
year: 'numeric',
|
|
hour: '2-digit',
|
|
minute: '2-digit',
|
|
second: '2-digit',
|
|
hour12: false,
|
|
timeZone: 'Europe/Paris'
|
|
}).replace(',', '');
|
|
}
|
|
|
|
/**
|
|
* Compter les mots dans un texte
|
|
*/
|
|
function countWords(text) {
|
|
if (!text || text.trim() === '') return 0;
|
|
return text.trim().split(/\s+/).length;
|
|
}
|
|
|
|
/**
|
|
* Récupérer un article sauvé par ID
|
|
*/
|
|
async function getStoredArticle(articleId) {
|
|
try {
|
|
const sheets = await getSheetsClient();
|
|
|
|
const rowNumber = articleId + 2; // +2 car header + 0-indexing
|
|
const response = await sheets.spreadsheets.values.get({
|
|
spreadsheetId: SHEET_CONFIG.sheetId,
|
|
range: `Generated_Articles!A${rowNumber}:Q${rowNumber}`
|
|
});
|
|
|
|
if (!response.data.values || response.data.values.length === 0) {
|
|
throw new Error(`Article ${articleId} non trouvé`);
|
|
}
|
|
|
|
const data = response.data.values[0];
|
|
|
|
return {
|
|
articleId: articleId,
|
|
timestamp: data[0],
|
|
slug: data[1],
|
|
mc0: data[2],
|
|
t0: data[3],
|
|
personality: data[4],
|
|
antiDetectionLevel: data[5],
|
|
compiledText: data[6], // ← TEXTE PUR
|
|
textLength: data[7],
|
|
wordCount: data[8],
|
|
elementsCount: data[9],
|
|
llmUsed: data[10],
|
|
validationStatus: data[11],
|
|
gptZeroScore: data[12],
|
|
originalityScore: data[13],
|
|
copyLeaksScore: data[14],
|
|
humanScore: data[15],
|
|
fullMetadata: data[16] ? JSON.parse(data[16]) : null
|
|
};
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur récupération article ${articleId}: ${error.toString()}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Lister les derniers articles générés
|
|
*/
|
|
async function getRecentArticles(limit = 10) {
|
|
try {
|
|
const sheets = await getSheetsClient();
|
|
|
|
const response = await sheets.spreadsheets.values.get({
|
|
spreadsheetId: SHEET_CONFIG.sheetId,
|
|
range: 'Generated_Articles!A:L'
|
|
});
|
|
|
|
if (!response.data.values || response.data.values.length <= 1) {
|
|
return []; // Pas de données ou seulement headers
|
|
}
|
|
|
|
const data = response.data.values.slice(1); // Exclure headers
|
|
const startIndex = Math.max(0, data.length - limit);
|
|
const recentData = data.slice(startIndex);
|
|
|
|
return recentData.map((row, index) => ({
|
|
articleId: startIndex + index,
|
|
timestamp: row[0],
|
|
slug: row[1],
|
|
mc0: row[2],
|
|
personality: row[4],
|
|
antiDetectionLevel: row[5],
|
|
wordCount: row[8],
|
|
validationStatus: row[11]
|
|
})).reverse(); // Plus récents en premier
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur liste articles récents: ${error.toString()}`, 'ERROR');
|
|
return [];
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Mettre à jour les scores de détection d'un article
|
|
*/
|
|
async function updateDetectionScores(articleId, scores) {
|
|
try {
|
|
const sheets = await getSheetsClient();
|
|
const rowNumber = articleId + 2;
|
|
|
|
const updates = [];
|
|
|
|
// Colonnes des scores : M, N, O (GPTZero, Originality, CopyLeaks)
|
|
if (scores.gptzero !== undefined) {
|
|
updates.push({
|
|
range: `Generated_Articles!M${rowNumber}`,
|
|
values: [[scores.gptzero]]
|
|
});
|
|
}
|
|
if (scores.originality !== undefined) {
|
|
updates.push({
|
|
range: `Generated_Articles!N${rowNumber}`,
|
|
values: [[scores.originality]]
|
|
});
|
|
}
|
|
if (scores.copyleaks !== undefined) {
|
|
updates.push({
|
|
range: `Generated_Articles!O${rowNumber}`,
|
|
values: [[scores.copyleaks]]
|
|
});
|
|
}
|
|
|
|
if (updates.length > 0) {
|
|
await sheets.spreadsheets.values.batchUpdate({
|
|
spreadsheetId: SHEET_CONFIG.sheetId,
|
|
resource: {
|
|
valueInputOption: 'USER_ENTERED',
|
|
data: updates
|
|
}
|
|
});
|
|
}
|
|
|
|
logSh(`✅ Scores détection mis à jour pour article ${articleId}`, 'INFO');
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur maj scores article ${articleId}: ${error.toString()}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
// ============= HELPERS GOOGLE SHEETS =============
|
|
|
|
/**
|
|
* Obtenir le client Google Sheets authentifié
|
|
*/
|
|
async function getSheetsClient() {
|
|
const auth = new google.auth.GoogleAuth({
|
|
credentials: {
|
|
client_email: process.env.GOOGLE_SERVICE_ACCOUNT_EMAIL,
|
|
private_key: process.env.GOOGLE_PRIVATE_KEY?.replace(/\\n/g, '\n')
|
|
},
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets']
|
|
});
|
|
|
|
const authClient = await auth.getClient();
|
|
const sheets = google.sheets({ version: 'v4', auth: authClient });
|
|
|
|
return sheets;
|
|
}
|
|
|
|
/**
|
|
* Obtenir ou créer une sheet
|
|
*/
|
|
async function getOrCreateSheet(sheets, sheetName) {
|
|
try {
|
|
// Vérifier si la sheet existe
|
|
const response = await sheets.spreadsheets.get({
|
|
spreadsheetId: SHEET_CONFIG.sheetId
|
|
});
|
|
|
|
const existingSheet = response.data.sheets.find(
|
|
sheet => sheet.properties.title === sheetName
|
|
);
|
|
|
|
if (existingSheet) {
|
|
return existingSheet;
|
|
} else {
|
|
// Créer la sheet si elle n'existe pas
|
|
if (sheetName === 'Generated_Articles') {
|
|
await createArticlesStorageSheet(sheets);
|
|
return await getOrCreateSheet(sheets, sheetName); // Récursif pour récupérer la sheet créée
|
|
}
|
|
throw new Error(`Sheet ${sheetName} non supportée pour création automatique`);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur accès/création sheet ${sheetName}: ${error.toString()}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Obtenir l'ID d'une sheet par son nom
|
|
*/
|
|
async function getSheetIdByName(sheets, sheetName) {
|
|
const response = await sheets.spreadsheets.get({
|
|
spreadsheetId: SHEET_CONFIG.sheetId
|
|
});
|
|
|
|
const sheet = response.data.sheets.find(
|
|
s => s.properties.title === sheetName
|
|
);
|
|
|
|
return sheet ? sheet.properties.sheetId : null;
|
|
}
|
|
|
|
// ============= EXPORTS =============
|
|
|
|
module.exports = {
|
|
compileGeneratedTextsOrganic,
|
|
buildOrganicSections,
|
|
findAssociatedContent,
|
|
extractFAQNumber,
|
|
findMatchingFAQAnswer,
|
|
saveGeneratedArticleOrganic,
|
|
identifyElementType,
|
|
cleanIndividualContent,
|
|
createArticlesStorageSheet,
|
|
formatDateToFrench,
|
|
countWords,
|
|
getStoredArticle,
|
|
getRecentArticles,
|
|
updateDetectionScores,
|
|
getSheetsClient,
|
|
getOrCreateSheet,
|
|
getSheetIdByName,
|
|
generateSlugFromContent
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/DigitalOceanWorkflow.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: DigitalOceanWorkflow.js - REFACTORISÉ POUR NODE.JS
|
|
// RESPONSABILITÉ: Orchestration + Interface Digital Ocean UNIQUEMENT
|
|
// ========================================
|
|
|
|
const crypto = require('crypto');
|
|
const axios = require('axios');
|
|
const { GoogleSpreadsheet } = require('google-spreadsheet');
|
|
const { JWT } = require('google-auth-library');
|
|
|
|
// Import des autres modules du projet (à adapter selon votre structure)
|
|
const { logSh } = require('./ErrorReporting');
|
|
const { handleFullWorkflow } = require('./Main');
|
|
const { getPersonalities, selectPersonalityWithAI } = require('./BrainConfig');
|
|
|
|
// ============= CONFIGURATION DIGITAL OCEAN =============
|
|
const DO_CONFIG = {
|
|
endpoint: 'https://autocollant.fra1.digitaloceanspaces.com',
|
|
bucketName: 'autocollant',
|
|
accessKeyId: 'DO801XTYPE968NZGAQM3',
|
|
secretAccessKey: '5aCCBiS9K+J8gsAe3M3/0GlliHCNjtLntwla1itCN1s',
|
|
region: 'fra1'
|
|
};
|
|
|
|
// Configuration Google Sheets
|
|
const SHEET_CONFIG = {
|
|
sheetId: '1iA2GvWeUxX-vpnAMfVm3ZMG9LhaC070SdGssEcXAh2c',
|
|
serviceAccountEmail: process.env.GOOGLE_SERVICE_ACCOUNT_EMAIL,
|
|
privateKey: process.env.GOOGLE_PRIVATE_KEY?.replace(/\\n/g, '\n'),
|
|
// Alternative: utiliser fichier JSON directement
|
|
keyFile: './seo-generator-470715-85d4a971c1af.json'
|
|
};
|
|
|
|
async function deployArticle({ path, html, dryRun = false, ...rest }) {
|
|
if (!path || typeof html !== 'string') {
|
|
const err = new Error('deployArticle: invalid payload (requires {path, html})');
|
|
err.code = 'E_PAYLOAD';
|
|
throw err;
|
|
}
|
|
if (dryRun) {
|
|
return {
|
|
ok: true,
|
|
dryRun: true,
|
|
length: html.length,
|
|
path,
|
|
meta: rest || {}
|
|
};
|
|
}
|
|
// --- Impl réelle à toi ici (upload DO Spaces / API / SSH etc.) ---
|
|
// return await realDeploy({ path, html, ...rest });
|
|
|
|
// Placeholder pour ne pas casser l'appel si pas encore implémenté
|
|
return { ok: true, dryRun: false, path, length: html.length };
|
|
}
|
|
|
|
module.exports.deployArticle = module.exports.deployArticle || deployArticle;
|
|
|
|
|
|
// ============= TRIGGER PRINCIPAL REMPLACÉ PAR WEBHOOK/API =============
|
|
|
|
/**
|
|
* Point d'entrée pour déclencher le workflow
|
|
* Remplace le trigger onEdit d'Apps Script
|
|
* @param {number} rowNumber - Numéro de ligne à traiter
|
|
* @returns {Promise<object>} - Résultat du workflow
|
|
*/
|
|
async function triggerAutonomousWorkflow(rowNumber) {
|
|
try {
|
|
logSh('🚀 TRIGGER AUTONOME DÉCLENCHÉ (Digital Ocean)', 'INFO');
|
|
|
|
// Anti-bouncing simulé
|
|
await new Promise(resolve => setTimeout(resolve, 2000));
|
|
|
|
return await runAutonomousWorkflowFromTrigger(rowNumber);
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur trigger autonome DO: ${error.toString()}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* ORCHESTRATEUR: Prépare les données et délègue à Main.js
|
|
*/
|
|
async function runAutonomousWorkflowFromTrigger(rowNumber) {
|
|
const startTime = Date.now();
|
|
|
|
try {
|
|
logSh(`🎬 ORCHESTRATION AUTONOME - LIGNE ${rowNumber}`, 'INFO');
|
|
|
|
// 1. LIRE DONNÉES CSV + XML FILENAME
|
|
const csvData = await readCSVDataWithXMLFileName(rowNumber);
|
|
logSh(`✅ CSV: ${csvData.mc0}, XML: ${csvData.xmlFileName}`, 'INFO');
|
|
|
|
// 2. RÉCUPÉRER XML DEPUIS DIGITAL OCEAN
|
|
const xmlTemplate = await fetchXMLFromDigitalOceanSimple(csvData.xmlFileName);
|
|
logSh(`✅ XML récupéré: ${xmlTemplate.length} caractères`, 'INFO');
|
|
|
|
// 3. 🎯 DÉLÉGUER LE WORKFLOW À MAIN.JS
|
|
const workflowData = {
|
|
rowNumber: rowNumber,
|
|
xmlTemplate: Buffer.from(xmlTemplate).toString('base64'), // Encoder comme Make.com
|
|
csvData: csvData,
|
|
source: 'digital_ocean_autonomous'
|
|
};
|
|
|
|
const result = await handleFullWorkflow(workflowData);
|
|
|
|
const duration = Date.now() - startTime;
|
|
logSh(`🏆 ORCHESTRATION TERMINÉE en ${Math.round(duration/1000)}s`, 'INFO');
|
|
|
|
// 4. MARQUER LIGNE COMME TRAITÉE
|
|
await markRowAsProcessed(rowNumber, result);
|
|
|
|
return result;
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ERREUR ORCHESTRATION: ${error.toString()}`, 'ERROR');
|
|
await markRowAsError(rowNumber, error.toString());
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
// ============= INTERFACE DIGITAL OCEAN =============
|
|
|
|
async function fetchXMLFromDigitalOceanSimple(fileName) {
|
|
const filePath = `wp-content/XML/${fileName}`;
|
|
const fileUrl = `${DO_CONFIG.endpoint}/${filePath}`;
|
|
|
|
try {
|
|
const response = await axios.get(fileUrl); // Sans auth
|
|
return response.data;
|
|
} catch (error) {
|
|
throw new Error(`Fichier non accessible: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Récupérer XML depuis Digital Ocean Spaces avec authentification
|
|
*/
|
|
async function fetchXMLFromDigitalOcean(fileName) {
|
|
if (!fileName) {
|
|
throw new Error('Nom de fichier XML requis');
|
|
}
|
|
|
|
const filePath = `wp-content/XML/${fileName}`;
|
|
logSh(`🌊 Récupération XML: ${fileName} , ${filePath}`, 'DEBUG');
|
|
|
|
const fileUrl = `${DO_CONFIG.endpoint}/${filePath}`;
|
|
logSh(`🔗 URL complète: ${fileUrl}`, 'DEBUG');
|
|
|
|
const signature = generateAWSSignature(filePath);
|
|
|
|
try {
|
|
const response = await axios.get(fileUrl, {
|
|
headers: signature.headers
|
|
});
|
|
|
|
logSh(`📡 Response code: ${response.status}`, 'DEBUG');
|
|
logSh(`📄 Response: ${response.data.toString()}`, 'DEBUG');
|
|
|
|
if (response.status === 200) {
|
|
return response.data;
|
|
} else {
|
|
throw new Error(`HTTP ${response.status}: ${response.data}`);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur DO complète: ${error.toString()}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Lire données CSV avec nom fichier XML (colonne J)
|
|
*/
|
|
async function readCSVDataWithXMLFileName(rowNumber) {
|
|
try {
|
|
// Configuration Google Sheets - avec fallback sur fichier JSON
|
|
let serviceAccountAuth;
|
|
|
|
if (SHEET_CONFIG.serviceAccountEmail && SHEET_CONFIG.privateKey) {
|
|
// Utiliser variables d'environnement
|
|
serviceAccountAuth = new JWT({
|
|
email: SHEET_CONFIG.serviceAccountEmail,
|
|
key: SHEET_CONFIG.privateKey,
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets']
|
|
});
|
|
} else {
|
|
// Utiliser fichier JSON
|
|
serviceAccountAuth = new JWT({
|
|
keyFile: SHEET_CONFIG.keyFile,
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets']
|
|
});
|
|
}
|
|
|
|
const doc = new GoogleSpreadsheet(SHEET_CONFIG.sheetId, serviceAccountAuth);
|
|
await doc.loadInfo();
|
|
|
|
const sheet = doc.sheetsByTitle['instructions'];
|
|
if (!sheet) {
|
|
throw new Error('Sheet "instructions" non trouvée');
|
|
}
|
|
|
|
await sheet.loadCells(`A${rowNumber}:I${rowNumber}`);
|
|
|
|
const slug = sheet.getCell(rowNumber - 1, 0).value;
|
|
const t0 = sheet.getCell(rowNumber - 1, 1).value;
|
|
const mc0 = sheet.getCell(rowNumber - 1, 2).value;
|
|
const tMinus1 = sheet.getCell(rowNumber - 1, 3).value;
|
|
const lMinus1 = sheet.getCell(rowNumber - 1, 4).value;
|
|
const mcPlus1 = sheet.getCell(rowNumber - 1, 5).value;
|
|
const tPlus1 = sheet.getCell(rowNumber - 1, 6).value;
|
|
const lPlus1 = sheet.getCell(rowNumber - 1, 7).value;
|
|
const xmlFileName = sheet.getCell(rowNumber - 1, 8).value;
|
|
|
|
if (!xmlFileName || xmlFileName.toString().trim() === '') {
|
|
throw new Error(`Nom fichier XML manquant colonne I, ligne ${rowNumber}`);
|
|
}
|
|
|
|
let cleanFileName = xmlFileName.toString().trim();
|
|
if (!cleanFileName.endsWith('.xml')) {
|
|
cleanFileName += '.xml';
|
|
}
|
|
|
|
// Récupérer personnalité (délègue au système existant BrainConfig.js)
|
|
const personalities = await getPersonalities(); // Pas de paramètre, lit depuis JSON
|
|
const selectedPersonality = await selectPersonalityWithAI(mc0, t0, personalities);
|
|
|
|
return {
|
|
rowNumber: rowNumber,
|
|
slug: slug,
|
|
t0: t0,
|
|
mc0: mc0,
|
|
tMinus1: tMinus1,
|
|
lMinus1: lMinus1,
|
|
mcPlus1: mcPlus1,
|
|
tPlus1: tPlus1,
|
|
lPlus1: lPlus1,
|
|
xmlFileName: cleanFileName,
|
|
personality: selectedPersonality
|
|
};
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur lecture CSV: ${error.toString()}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
// ============= STATUTS ET VALIDATION =============
|
|
|
|
/**
|
|
* Vérifier si le workflow doit être déclenché
|
|
* En Node.js, cette logique sera adaptée selon votre stratégie (webhook, polling, etc.)
|
|
*/
|
|
function shouldTriggerWorkflow(rowNumber, xmlFileName) {
|
|
if (!rowNumber || rowNumber <= 1) {
|
|
return false;
|
|
}
|
|
|
|
if (!xmlFileName || xmlFileName.toString().trim() === '') {
|
|
logSh('⚠️ Pas de fichier XML (colonne J), workflow ignoré', 'WARNING');
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
async function markRowAsProcessed(rowNumber, result) {
|
|
try {
|
|
// Configuration Google Sheets - avec fallback sur fichier JSON
|
|
let serviceAccountAuth;
|
|
|
|
if (SHEET_CONFIG.serviceAccountEmail && SHEET_CONFIG.privateKey) {
|
|
serviceAccountAuth = new JWT({
|
|
email: SHEET_CONFIG.serviceAccountEmail,
|
|
key: SHEET_CONFIG.privateKey,
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets']
|
|
});
|
|
} else {
|
|
serviceAccountAuth = new JWT({
|
|
keyFile: SHEET_CONFIG.keyFile,
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets']
|
|
});
|
|
}
|
|
|
|
const doc = new GoogleSpreadsheet(SHEET_CONFIG.sheetId, serviceAccountAuth);
|
|
await doc.loadInfo();
|
|
|
|
const sheet = doc.sheetsByTitle['instructions'];
|
|
|
|
// Vérifier et ajouter headers si nécessaire
|
|
await sheet.loadCells('K1:N1');
|
|
if (!sheet.getCell(0, 10).value) {
|
|
sheet.getCell(0, 10).value = 'Status';
|
|
sheet.getCell(0, 11).value = 'Processed_At';
|
|
sheet.getCell(0, 12).value = 'Article_ID';
|
|
sheet.getCell(0, 13).value = 'Source';
|
|
await sheet.saveUpdatedCells();
|
|
}
|
|
|
|
// Marquer la ligne
|
|
await sheet.loadCells(`K${rowNumber}:N${rowNumber}`);
|
|
sheet.getCell(rowNumber - 1, 10).value = '✅ DO_SUCCESS';
|
|
sheet.getCell(rowNumber - 1, 11).value = new Date().toISOString();
|
|
sheet.getCell(rowNumber - 1, 12).value = result.articleStorage?.articleId || '';
|
|
sheet.getCell(rowNumber - 1, 13).value = 'Digital Ocean';
|
|
|
|
await sheet.saveUpdatedCells();
|
|
|
|
logSh(`✅ Ligne ${rowNumber} marquée comme traitée`, 'INFO');
|
|
|
|
} catch (error) {
|
|
logSh(`⚠️ Erreur marquage statut: ${error.toString()}`, 'WARNING');
|
|
}
|
|
}
|
|
|
|
async function markRowAsError(rowNumber, errorMessage) {
|
|
try {
|
|
// Configuration Google Sheets - avec fallback sur fichier JSON
|
|
let serviceAccountAuth;
|
|
|
|
if (SHEET_CONFIG.serviceAccountEmail && SHEET_CONFIG.privateKey) {
|
|
serviceAccountAuth = new JWT({
|
|
email: SHEET_CONFIG.serviceAccountEmail,
|
|
key: SHEET_CONFIG.privateKey,
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets']
|
|
});
|
|
} else {
|
|
serviceAccountAuth = new JWT({
|
|
keyFile: SHEET_CONFIG.keyFile,
|
|
scopes: ['https://www.googleapis.com/auth/spreadsheets']
|
|
});
|
|
}
|
|
|
|
const doc = new GoogleSpreadsheet(SHEET_CONFIG.sheetId, serviceAccountAuth);
|
|
await doc.loadInfo();
|
|
|
|
const sheet = doc.sheetsByTitle['instructions'];
|
|
|
|
await sheet.loadCells(`K${rowNumber}:N${rowNumber}`);
|
|
sheet.getCell(rowNumber - 1, 10).value = '❌ DO_ERROR';
|
|
sheet.getCell(rowNumber - 1, 11).value = new Date().toISOString();
|
|
sheet.getCell(rowNumber - 1, 12).value = errorMessage.substring(0, 100);
|
|
sheet.getCell(rowNumber - 1, 13).value = 'DO Error';
|
|
|
|
await sheet.saveUpdatedCells();
|
|
|
|
} catch (error) {
|
|
logSh(`⚠️ Erreur marquage erreur: ${error.toString()}`, 'WARNING');
|
|
}
|
|
}
|
|
|
|
// ============= SIGNATURE AWS V4 =============
|
|
|
|
function generateAWSSignature(filePath) {
|
|
const now = new Date();
|
|
const dateStamp = now.toISOString().slice(0, 10).replace(/-/g, '');
|
|
const timeStamp = now.toISOString().replace(/[-:]/g, '').slice(0, -5) + 'Z';
|
|
|
|
const headers = {
|
|
'Host': DO_CONFIG.endpoint.replace('https://', ''),
|
|
'X-Amz-Date': timeStamp,
|
|
'X-Amz-Content-Sha256': 'UNSIGNED-PAYLOAD'
|
|
};
|
|
|
|
const credentialScope = `${dateStamp}/${DO_CONFIG.region}/s3/aws4_request`;
|
|
|
|
const canonicalHeaders = Object.keys(headers)
|
|
.sort()
|
|
.map(key => `${key.toLowerCase()}:${headers[key]}`)
|
|
.join('\n');
|
|
|
|
const signedHeaders = Object.keys(headers)
|
|
.map(key => key.toLowerCase())
|
|
.sort()
|
|
.join(';');
|
|
|
|
const canonicalRequest = [
|
|
'GET',
|
|
`/${filePath}`,
|
|
'',
|
|
canonicalHeaders + '\n',
|
|
signedHeaders,
|
|
'UNSIGNED-PAYLOAD'
|
|
].join('\n');
|
|
|
|
const stringToSign = [
|
|
'AWS4-HMAC-SHA256',
|
|
timeStamp,
|
|
credentialScope,
|
|
crypto.createHash('sha256').update(canonicalRequest).digest('hex')
|
|
].join('\n');
|
|
|
|
// Calculs HMAC étape par étape
|
|
const kDate = crypto.createHmac('sha256', 'AWS4' + DO_CONFIG.secretAccessKey).update(dateStamp).digest();
|
|
const kRegion = crypto.createHmac('sha256', kDate).update(DO_CONFIG.region).digest();
|
|
const kService = crypto.createHmac('sha256', kRegion).update('s3').digest();
|
|
const kSigning = crypto.createHmac('sha256', kService).update('aws4_request').digest();
|
|
const signature = crypto.createHmac('sha256', kSigning).update(stringToSign).digest('hex');
|
|
|
|
headers['Authorization'] = `AWS4-HMAC-SHA256 Credential=${DO_CONFIG.accessKeyId}/${credentialScope}, SignedHeaders=${signedHeaders}, Signature=${signature}`;
|
|
|
|
return { headers: headers };
|
|
}
|
|
|
|
// ============= SETUP ET TEST =============
|
|
|
|
/**
|
|
* Configuration du trigger autonome - Remplacé par webhook ou polling en Node.js
|
|
*/
|
|
function setupAutonomousTrigger() {
|
|
logSh('⚙️ Configuration trigger autonome Digital Ocean...', 'INFO');
|
|
|
|
// En Node.js, vous pourriez utiliser:
|
|
// - Express.js avec webhooks
|
|
// - Cron jobs avec node-cron
|
|
// - Polling de la Google Sheet
|
|
// - WebSocket connections
|
|
|
|
logSh('✅ Configuration prête pour webhooks/polling Node.js', 'INFO');
|
|
logSh('🎯 Mode: Webhook/API → Digital Ocean → Main.js', 'INFO');
|
|
}
|
|
|
|
async function testDigitalOceanConnection() {
|
|
logSh('🧪 Test connexion Digital Ocean...', 'INFO');
|
|
|
|
try {
|
|
const testFiles = ['template1.xml', 'plaque-rue.xml', 'test.xml'];
|
|
|
|
for (const fileName of testFiles) {
|
|
try {
|
|
const content = await fetchXMLFromDigitalOceanSimple(fileName);
|
|
logSh(`✅ Fichier '${fileName}' accessible (${content.length} chars)`, 'INFO');
|
|
return true;
|
|
} catch (error) {
|
|
logSh(`⚠️ '${fileName}' non accessible: ${error.toString()}`, 'DEBUG');
|
|
}
|
|
}
|
|
|
|
logSh('❌ Aucun fichier test accessible dans DO', 'ERROR');
|
|
return false;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Test DO échoué: ${error.toString()}`, 'ERROR');
|
|
return false;
|
|
}
|
|
}
|
|
|
|
// ============= EXPORTS =============
|
|
|
|
module.exports = {
|
|
triggerAutonomousWorkflow,
|
|
runAutonomousWorkflowFromTrigger,
|
|
fetchXMLFromDigitalOcean,
|
|
fetchXMLFromDigitalOceanSimple,
|
|
readCSVDataWithXMLFileName,
|
|
markRowAsProcessed,
|
|
markRowAsError,
|
|
testDigitalOceanConnection,
|
|
setupAutonomousTrigger,
|
|
DO_CONFIG
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/Main.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: lib/main.js - CONVERTI POUR NODE.JS
|
|
// RESPONSABILITÉ: COEUR DU WORKFLOW DE GÉNÉRATION
|
|
// ========================================
|
|
|
|
// 🔧 CONFIGURATION ENVIRONNEMENT
|
|
require('dotenv').config({ path: require('path').join(__dirname, '..', '.env') });
|
|
|
|
|
|
// 🔄 IMPORTS NODE.JS (remplace les dépendances Apps Script)
|
|
const { getBrainConfig } = require('./BrainConfig');
|
|
const { extractElements, buildSmartHierarchy } = require('./ElementExtraction');
|
|
const { generateMissingKeywords } = require('./MissingKeywords');
|
|
const { generateWithContext } = require('./ContentGeneration');
|
|
const { injectGeneratedContent, cleanStrongTags } = require('./ContentAssembly');
|
|
const { validateWorkflowIntegrity, logSh } = require('./ErrorReporting');
|
|
const { saveGeneratedArticleOrganic } = require('./ArticleStorage');
|
|
const { tracer } = require('./trace.js');
|
|
const { fetchXMLFromDigitalOcean } = require('./DigitalOceanWorkflow');
|
|
const { spawn } = require('child_process');
|
|
const path = require('path');
|
|
|
|
// Variable pour éviter de relancer Edge plusieurs fois
|
|
let logViewerLaunched = false;
|
|
|
|
/**
|
|
* Lancer le log viewer dans Edge
|
|
*/
|
|
function launchLogViewer() {
|
|
if (logViewerLaunched || process.env.NODE_ENV === 'test') return;
|
|
|
|
try {
|
|
const logViewerPath = path.join(__dirname, '..', 'tools', 'logs-viewer.html');
|
|
const fileUrl = `file:///${logViewerPath.replace(/\\/g, '/')}`;
|
|
|
|
// Détecter l'environnement et adapter la commande
|
|
const isWSL = process.env.WSL_DISTRO_NAME || process.env.WSL_INTEROP;
|
|
const isWindows = process.platform === 'win32';
|
|
|
|
if (isWindows && !isWSL) {
|
|
// Windows natif
|
|
const edgeProcess = spawn('cmd', ['/c', 'start', 'msedge', fileUrl], {
|
|
detached: true,
|
|
stdio: 'ignore'
|
|
});
|
|
edgeProcess.unref();
|
|
} else if (isWSL) {
|
|
// WSL - utiliser cmd.exe via /mnt/c/Windows/System32/
|
|
const edgeProcess = spawn('/mnt/c/Windows/System32/cmd.exe', ['/c', 'start', 'msedge', fileUrl], {
|
|
detached: true,
|
|
stdio: 'ignore'
|
|
});
|
|
edgeProcess.unref();
|
|
} else {
|
|
// Linux/Mac - essayer xdg-open ou open
|
|
const command = process.platform === 'darwin' ? 'open' : 'xdg-open';
|
|
const browserProcess = spawn(command, [fileUrl], {
|
|
detached: true,
|
|
stdio: 'ignore'
|
|
});
|
|
browserProcess.unref();
|
|
}
|
|
|
|
logViewerLaunched = true;
|
|
logSh('🌐 Log viewer lancé', 'INFO');
|
|
} catch (error) {
|
|
logSh(`⚠️ Impossible d'ouvrir le log viewer: ${error.message}`, 'WARNING');
|
|
}
|
|
}
|
|
|
|
/**
|
|
* COEUR DU WORKFLOW - Compatible Make.com ET Digital Ocean ET Node.js
|
|
* @param {object} data - Données du workflow
|
|
* @param {string} data.xmlTemplate - XML template (base64 encodé)
|
|
* @param {object} data.csvData - Données CSV ou rowNumber
|
|
* @param {string} data.source - 'make_com' | 'digital_ocean_autonomous' | 'node_server'
|
|
*/
|
|
async function handleFullWorkflow(data) {
|
|
// Lancer le log viewer au début du workflow
|
|
launchLogViewer();
|
|
|
|
return await tracer.run('Main.handleFullWorkflow()', async () => {
|
|
await tracer.annotate({ source: data.source || 'node_server', mc0: data.csvData?.mc0 || data.rowNumber });
|
|
|
|
// 1. PRÉPARER LES DONNÉES CSV
|
|
const csvData = await tracer.run('Main.prepareCSVData()', async () => {
|
|
const result = await prepareCSVData(data);
|
|
await tracer.event(`CSV préparé: ${result.mc0}`, { csvKeys: Object.keys(result) });
|
|
return result;
|
|
}, { rowNumber: data.rowNumber, source: data.source });
|
|
|
|
// 2. DÉCODER LE XML TEMPLATE
|
|
const xmlString = await tracer.run('Main.decodeXMLTemplate()', async () => {
|
|
const result = decodeXMLTemplate(data.xmlTemplate);
|
|
await tracer.event(`XML décodé: ${result.length} caractères`);
|
|
return result;
|
|
}, { templateLength: data.xmlTemplate?.length });
|
|
|
|
// 3. PREPROCESSING XML
|
|
const processedXML = await tracer.run('Main.preprocessXML()', async () => {
|
|
const result = preprocessXML(xmlString);
|
|
await tracer.event('XML préprocessé');
|
|
global.currentXmlTemplate = result;
|
|
return result;
|
|
}, { originalLength: xmlString?.length });
|
|
|
|
// 4. EXTRAIRE ÉLÉMENTS
|
|
const elements = await tracer.run('ElementExtraction.extractElements()', async () => {
|
|
const result = await extractElements(processedXML, csvData);
|
|
await tracer.event(`${result.length} éléments extraits`);
|
|
return result;
|
|
}, { xmlLength: processedXML?.length, mc0: csvData.mc0 });
|
|
|
|
// 5. GÉNÉRER MOTS-CLÉS MANQUANTS
|
|
const finalElements = await tracer.run('MissingKeywords.generateMissingKeywords()', async () => {
|
|
const updatedElements = await generateMissingKeywords(elements, csvData);
|
|
const result = Object.keys(updatedElements).length > 0 ? updatedElements : elements;
|
|
await tracer.event('Mots-clés manquants traités');
|
|
return result;
|
|
}, { elementsCount: elements.length, mc0: csvData.mc0 });
|
|
|
|
// 6. CONSTRUIRE HIÉRARCHIE INTELLIGENTE
|
|
const hierarchy = await tracer.run('ElementExtraction.buildSmartHierarchy()', async () => {
|
|
const result = await buildSmartHierarchy(finalElements);
|
|
await tracer.event(`Hiérarchie construite: ${Object.keys(result).length} sections`);
|
|
return result;
|
|
}, { finalElementsCount: finalElements.length });
|
|
|
|
// 7. 🎯 GÉNÉRATION AVEC SELECTIVE ENHANCEMENT (Phase 2)
|
|
const generatedContent = await tracer.run('ContentGeneration.generateWithContext()', async () => {
|
|
const result = await generateWithContext(hierarchy, csvData);
|
|
await tracer.event(`Contenu généré: ${Object.keys(result).length} éléments`);
|
|
return result;
|
|
}, { elementsCount: Object.keys(hierarchy).length, personality: csvData.personality?.nom });
|
|
|
|
// 8. ASSEMBLER XML FINAL
|
|
const finalXML = await tracer.run('ContentAssembly.injectGeneratedContent()', async () => {
|
|
const result = injectGeneratedContent(processedXML, generatedContent, finalElements);
|
|
await tracer.event('XML final assemblé');
|
|
return result;
|
|
}, { contentPieces: Object.keys(generatedContent).length, elementsCount: finalElements.length });
|
|
|
|
// 9. VALIDATION INTÉGRITÉ
|
|
const validationReport = await tracer.run('ErrorReporting.validateWorkflowIntegrity()', async () => {
|
|
const result = validateWorkflowIntegrity(finalElements, generatedContent, finalXML, csvData);
|
|
await tracer.event(`Validation: ${result.status}`);
|
|
return result;
|
|
}, { finalXMLLength: finalXML?.length, contentKeys: Object.keys(generatedContent).length });
|
|
|
|
// 10. SAUVEGARDE ARTICLE
|
|
const articleStorage = await tracer.run('Main.saveArticle()', async () => {
|
|
const result = await saveArticle(finalXML, generatedContent, finalElements, csvData, data.source);
|
|
if (result) {
|
|
await tracer.event(`Article sauvé: ID ${result.articleId}`);
|
|
}
|
|
return result;
|
|
}, { source: data.source, mc0: csvData.mc0, elementsCount: finalElements.length });
|
|
|
|
// 11. RÉPONSE FINALE
|
|
const response = await tracer.run('Main.buildWorkflowResponse()', async () => {
|
|
const result = await buildWorkflowResponse(finalXML, generatedContent, finalElements, csvData, validationReport, articleStorage, data.source);
|
|
await tracer.event(`Response keys: ${Object.keys(result).join(', ')}`);
|
|
return result;
|
|
}, { validationStatus: validationReport?.status, articleId: articleStorage?.articleId });
|
|
|
|
return response;
|
|
}, { source: data.source || 'node_server', rowNumber: data.rowNumber, hasXMLTemplate: !!data.xmlTemplate });
|
|
}
|
|
|
|
// ============= PRÉPARATION DONNÉES =============
|
|
|
|
/**
|
|
* Préparer les données CSV selon la source - ASYNC pour Node.js
|
|
* RÉCUPÈRE: Google Sheets (données CSV) + Digital Ocean (XML template)
|
|
*/
|
|
async function prepareCSVData(data) {
|
|
if (data.csvData && data.csvData.mc0) {
|
|
// Données déjà préparées (Digital Ocean ou direct)
|
|
return data.csvData;
|
|
} else if (data.rowNumber) {
|
|
// 1. RÉCUPÉRER DONNÉES CSV depuis Google Sheet (OBLIGATOIRE)
|
|
await logSh(`🧠 Récupération données CSV ligne ${data.rowNumber}...`, 'INFO');
|
|
const config = await getBrainConfig(data.rowNumber);
|
|
if (!config.success) {
|
|
await logSh('❌ ÉCHEC: Impossible de récupérer les données Google Sheets', 'ERROR');
|
|
throw new Error('FATAL: Google Sheets inaccessible - arrêt du workflow');
|
|
}
|
|
|
|
// 2. VÉRIFIER XML FILENAME depuis Google Sheet (colonne I)
|
|
const xmlFileName = config.data.xmlFileName;
|
|
if (!xmlFileName || xmlFileName.trim() === '') {
|
|
await logSh('❌ ÉCHEC: Nom fichier XML manquant (colonne I Google Sheets)', 'ERROR');
|
|
throw new Error('FATAL: XML filename manquant - arrêt du workflow');
|
|
}
|
|
|
|
await logSh(`📋 CSV récupéré: ${config.data.mc0}`, 'INFO');
|
|
await logSh(`📄 XML filename: ${xmlFileName}`, 'INFO');
|
|
|
|
// 3. RÉCUPÉRER XML CONTENT depuis Digital Ocean avec AUTH (OBLIGATOIRE)
|
|
await logSh(`🌊 Récupération XML template depuis Digital Ocean (avec signature AWS)...`, 'INFO');
|
|
let xmlContent;
|
|
try {
|
|
xmlContent = await fetchXMLFromDigitalOcean(xmlFileName);
|
|
await logSh(`✅ XML récupéré: ${xmlContent.length} caractères`, 'INFO');
|
|
} catch (digitalOceanError) {
|
|
await logSh(`❌ ÉCHEC: Digital Ocean inaccessible - ${digitalOceanError.message}`, 'ERROR');
|
|
throw new Error(`FATAL: Digital Ocean échec - arrêt du workflow: ${digitalOceanError.message}`);
|
|
}
|
|
|
|
// 4. ENCODER XML pour le workflow (comme Make.com)
|
|
// Si on a récupéré un fichier XML, l'utiliser. Sinon utiliser le template par défaut déjà dans config.data.xmlTemplate
|
|
if (xmlContent) {
|
|
data.xmlTemplate = Buffer.from(xmlContent).toString('base64');
|
|
await logSh('🔄 XML depuis Digital Ocean encodé base64 pour le workflow', 'DEBUG');
|
|
} else if (config.data.xmlTemplate) {
|
|
data.xmlTemplate = Buffer.from(config.data.xmlTemplate).toString('base64');
|
|
await logSh('🔄 XML template par défaut encodé base64 pour le workflow', 'DEBUG');
|
|
}
|
|
|
|
return config.data;
|
|
} else {
|
|
throw new Error('FATAL: Données CSV invalides - rowNumber requis');
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Décoder le XML template - NODE.JS VERSION
|
|
*/
|
|
function decodeXMLTemplate(xmlTemplate) {
|
|
if (!xmlTemplate) {
|
|
throw new Error('Template XML manquant');
|
|
}
|
|
|
|
// Si le template commence déjà par <?xml, c'est du texte plain
|
|
if (xmlTemplate.startsWith('<?xml') || xmlTemplate.startsWith('<')) {
|
|
return xmlTemplate;
|
|
}
|
|
|
|
try {
|
|
// 🔄 NODE.JS : Tenter base64 uniquement si ce n'est pas déjà du XML
|
|
const decoded = Buffer.from(xmlTemplate, 'base64').toString('utf8');
|
|
return decoded;
|
|
} catch (error) {
|
|
// Si échec, considérer comme texte plain
|
|
logSh('🔍 XML pas encodé base64, utilisation directe', 'DEBUG'); // Using logSh instead of console.log
|
|
return xmlTemplate;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Preprocessing XML (nettoyage) - IDENTIQUE
|
|
*/
|
|
function preprocessXML(xmlString) {
|
|
let processed = xmlString;
|
|
|
|
// Nettoyer balises <strong>
|
|
processed = cleanStrongTags(processed);
|
|
|
|
// Autres nettoyages futurs...
|
|
|
|
return processed;
|
|
}
|
|
|
|
// ============= SAUVEGARDE =============
|
|
|
|
/**
|
|
* Sauvegarder l'article avec métadonnées source - ASYNC pour Node.js
|
|
*/
|
|
async function saveArticle(finalXML, generatedContent, finalElements, csvData, source) {
|
|
await logSh('💾 Sauvegarde article...', 'INFO');
|
|
|
|
const articleData = {
|
|
xmlContent: finalXML,
|
|
generatedTexts: generatedContent,
|
|
elementsGenerated: finalElements.length,
|
|
originalElements: finalElements
|
|
};
|
|
|
|
const storageConfig = {
|
|
antiDetectionLevel: 'Selective_Enhancement',
|
|
llmUsed: 'claude+openai+gemini+mistral',
|
|
workflowVersion: '2.0-NodeJS', // 🔄 Mise à jour version
|
|
source: source || 'node_server', // 🔄 Source par défaut
|
|
enhancementTechniques: [
|
|
'technical_terms_gpt4',
|
|
'transitions_gemini',
|
|
'personality_style_mistral'
|
|
]
|
|
};
|
|
|
|
try {
|
|
const articleStorage = await saveGeneratedArticleOrganic(articleData, csvData, storageConfig);
|
|
await logSh(`✅ Article sauvé: ID ${articleStorage.articleId}`, 'INFO');
|
|
return articleStorage;
|
|
} catch (storageError) {
|
|
await logSh(`⚠️ Erreur sauvegarde: ${storageError.toString()}`, 'WARNING');
|
|
return null; // Non-bloquant
|
|
}
|
|
}
|
|
|
|
// ============= RÉPONSE =============
|
|
|
|
/**
|
|
* Construire la réponse finale du workflow - ASYNC pour logSh
|
|
*/
|
|
async function buildWorkflowResponse(finalXML, generatedContent, finalElements, csvData, validationReport, articleStorage, source) {
|
|
const response = {
|
|
success: true,
|
|
source: source,
|
|
xmlContent: finalXML,
|
|
generatedTexts: generatedContent,
|
|
elementsGenerated: finalElements.length,
|
|
personality: csvData.personality?.nom || 'Unknown',
|
|
csvData: {
|
|
mc0: csvData.mc0,
|
|
t0: csvData.t0,
|
|
personality: csvData.personality?.nom
|
|
},
|
|
timestamp: new Date().toISOString(),
|
|
validationReport: validationReport,
|
|
articleStorage: articleStorage,
|
|
|
|
// NOUVELLES MÉTADONNÉES PHASE 2
|
|
antiDetectionLevel: 'Selective_Enhancement',
|
|
llmsUsed: ['claude', 'openai', 'gemini', 'mistral'],
|
|
enhancementApplied: true,
|
|
workflowVersion: '2.0-NodeJS', // 🔄 Version mise à jour
|
|
|
|
// STATS PERFORMANCE
|
|
stats: {
|
|
xmlLength: finalXML.length,
|
|
contentPieces: Object.keys(generatedContent).length,
|
|
wordCount: calculateTotalWordCount(generatedContent),
|
|
validationStatus: validationReport.status
|
|
}
|
|
};
|
|
|
|
await logSh(`🔍 Response.stats: ${JSON.stringify(response.stats)}`, 'DEBUG');
|
|
|
|
return response;
|
|
}
|
|
|
|
// ============= HELPERS =============
|
|
|
|
/**
|
|
* Calculer nombre total de mots - IDENTIQUE
|
|
*/
|
|
function calculateTotalWordCount(generatedContent) {
|
|
let totalWords = 0;
|
|
Object.values(generatedContent).forEach(content => {
|
|
if (content && typeof content === 'string') {
|
|
totalWords += content.trim().split(/\s+/).length;
|
|
}
|
|
});
|
|
return totalWords;
|
|
}
|
|
|
|
// ============= POINTS D'ENTRÉE SUPPLÉMENTAIRES =============
|
|
|
|
/**
|
|
* Test du workflow principal - ASYNC pour Node.js
|
|
*/
|
|
async function testMainWorkflow() {
|
|
try {
|
|
const testData = {
|
|
csvData: {
|
|
mc0: 'plaque test nodejs',
|
|
t0: 'Test workflow principal Node.js',
|
|
personality: { nom: 'Marc', style: 'professionnel' },
|
|
tMinus1: 'parent test',
|
|
mcPlus1: 'mot1,mot2,mot3,mot4',
|
|
tPlus1: 'Titre1,Titre2,Titre3,Titre4'
|
|
},
|
|
xmlTemplate: Buffer.from('<?xml version="1.0"?><test>|Test_Element{{T0}}|</test>').toString('base64'),
|
|
source: 'test_main_nodejs'
|
|
};
|
|
|
|
const result = await handleFullWorkflow(testData);
|
|
return result;
|
|
|
|
} catch (error) {
|
|
throw error;
|
|
} finally {
|
|
tracer.printSummary();
|
|
}
|
|
}
|
|
|
|
// 🔄 NODE.JS EXPORTS
|
|
module.exports = {
|
|
handleFullWorkflow,
|
|
testMainWorkflow,
|
|
prepareCSVData,
|
|
decodeXMLTemplate,
|
|
preprocessXML,
|
|
saveArticle,
|
|
buildWorkflowResponse,
|
|
calculateTotalWordCount,
|
|
launchLogViewer
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/test-manual.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: test-manual.js - ENTRY POINT MANUEL
|
|
// Description: Test workflow ligne 2 Google Sheets
|
|
// Usage: node test-manual.js
|
|
// ========================================
|
|
|
|
require('./polyfills/fetch.cjs');
|
|
require('dotenv').config();
|
|
|
|
const { handleFullWorkflow } = require('./Main');
|
|
const { logSh } = require('./ErrorReporting');
|
|
|
|
/**
|
|
* TEST MANUEL LIGNE 2
|
|
*/
|
|
async function testWorkflowLigne2() {
|
|
logSh('🚀 === DÉMARRAGE TEST MANUEL LIGNE 2 ===', 'INFO'); // Using logSh instead of console.log
|
|
|
|
const startTime = Date.now();
|
|
|
|
try {
|
|
// DONNÉES DE TEST POUR LIGNE 2
|
|
const testData = {
|
|
rowNumber: 2, // Ligne 2 Google Sheets
|
|
source: 'test_manual_nodejs'
|
|
};
|
|
|
|
logSh('📊 Configuration test:', 'INFO'); // Using logSh instead of console.log
|
|
logSh(` • Ligne: ${testData.rowNumber}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(` • Source: ${testData.source}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(` • Timestamp: ${new Date().toISOString()}`, 'INFO'); // Using logSh instead of console.log
|
|
|
|
// LANCER LE WORKFLOW
|
|
logSh('\n🎯 Lancement workflow principal...', 'INFO'); // Using logSh instead of console.log
|
|
const result = await handleFullWorkflow(testData);
|
|
|
|
// AFFICHER RÉSULTATS
|
|
const duration = Date.now() - startTime;
|
|
logSh('\n🏆 === WORKFLOW TERMINÉ AVEC SUCCÈS ===', 'INFO'); // Using logSh instead of console.log
|
|
logSh(`⏱️ Durée: ${Math.round(duration/1000)}s`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(`📊 Status: ${result.success ? '✅ SUCCESS' : '❌ ERROR'}`, 'INFO'); // Using logSh instead of console.log
|
|
|
|
if (result.success) {
|
|
logSh(`📝 Éléments générés: ${result.elementsGenerated}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(`👤 Personnalité: ${result.personality}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(`🎯 MC0: ${result.csvData?.mc0 || 'N/A'}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(`📄 XML length: ${result.stats?.xmlLength || 'N/A'} chars`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(`🔤 Mots total: ${result.stats?.wordCount || 'N/A'}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(`🧠 LLMs utilisés: ${result.llmsUsed?.join(', ') || 'N/A'}`, 'INFO'); // Using logSh instead of console.log
|
|
|
|
if (result.articleStorage) {
|
|
logSh(`💾 Article sauvé: ID ${result.articleStorage.articleId}`, 'INFO'); // Using logSh instead of console.log
|
|
}
|
|
}
|
|
|
|
logSh('\n📋 Résultat complet:', 'DEBUG'); // Using logSh instead of console.log
|
|
logSh(JSON.stringify(result, null, 2), 'DEBUG'); // Using logSh instead of console.log
|
|
|
|
return result;
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh('\n❌ === ERREUR WORKFLOW ===', 'ERROR'); // Using logSh instead of console.error
|
|
logSh(`❌ Message: ${error.message}`, 'ERROR'); // Using logSh instead of console.error
|
|
logSh(`❌ Durée avant échec: ${Math.round(duration/1000)}s`, 'ERROR'); // Using logSh instead of console.error
|
|
|
|
if (process.env.NODE_ENV === 'development') {
|
|
logSh(`❌ Stack: ${error.stack}`, 'ERROR'); // Using logSh instead of console.error
|
|
}
|
|
|
|
// Afficher conseils de debug
|
|
logSh('\n🔧 CONSEILS DE DEBUG:', 'INFO'); // Using logSh instead of console.log
|
|
logSh('1. Vérifiez vos variables d\'environnement (.env)', 'INFO'); // Using logSh instead of console.log
|
|
logSh('2. Vérifiez la connexion Google Sheets', 'INFO'); // Using logSh instead of console.log
|
|
logSh('3. Vérifiez les API keys LLM', 'INFO'); // Using logSh instead of console.log
|
|
logSh('4. Regardez les logs détaillés dans ./logs/', 'INFO'); // Using logSh instead of console.log
|
|
|
|
process.exit(1);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* VÉRIFICATIONS PRÉALABLES
|
|
*/
|
|
function checkEnvironment() {
|
|
logSh('🔍 Vérification environnement...', 'INFO'); // Using logSh instead of console.log
|
|
|
|
const required = [
|
|
'GOOGLE_SHEETS_ID',
|
|
'OPENAI_API_KEY'
|
|
];
|
|
|
|
const missing = required.filter(key => !process.env[key]);
|
|
|
|
if (missing.length > 0) {
|
|
logSh('❌ Variables d\'environnement manquantes:', 'ERROR'); // Using logSh instead of console.error
|
|
missing.forEach(key => logSh(` • ${key}`, 'ERROR')); // Using logSh instead of console.error
|
|
logSh('\n💡 Créez un fichier .env avec ces variables', 'ERROR'); // Using logSh instead of console.error
|
|
process.exit(1);
|
|
}
|
|
|
|
logSh('✅ Variables d\'environnement OK', 'INFO'); // Using logSh instead of console.log
|
|
|
|
// Info sur les variables configurées
|
|
logSh('📋 Configuration détectée:', 'INFO'); // Using logSh instead of console.log
|
|
logSh(` • Google Sheets ID: ${process.env.GOOGLE_SHEETS_ID}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(` • OpenAI: ${process.env.OPENAI_API_KEY ? '✅ Configuré' : '❌ Manquant'}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(` • Claude: ${process.env.CLAUDE_API_KEY ? '✅ Configuré' : '⚠️ Optionnel'}`, 'INFO'); // Using logSh instead of console.log
|
|
logSh(` • Gemini: ${process.env.GEMINI_API_KEY ? '✅ Configuré' : '⚠️ Optionnel'}`, 'INFO'); // Using logSh instead of console.log
|
|
}
|
|
|
|
/**
|
|
* POINT D'ENTRÉE PRINCIPAL
|
|
*/
|
|
async function main() {
|
|
try {
|
|
// Vérifications préalables
|
|
checkEnvironment();
|
|
|
|
// Test workflow
|
|
await testWorkflowLigne2();
|
|
|
|
logSh('\n🎉 Test manuel terminé avec succès !', 'INFO'); // Using logSh instead of console.log
|
|
process.exit(0);
|
|
|
|
} catch (error) {
|
|
logSh('\n💥 Erreur fatale: ' + error.message, 'ERROR'); // Using logSh instead of console.error
|
|
process.exit(1);
|
|
}
|
|
}
|
|
|
|
// Lancer si exécuté directement
|
|
if (require.main === module) {
|
|
main();
|
|
}
|
|
|
|
module.exports = { testWorkflowLigne2 };
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/DetectorStrategies.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// DETECTOR STRATEGIES - NIVEAU 3
|
|
// Responsabilité: Stratégies spécialisées par détecteur IA
|
|
// Anti-détection: Techniques ciblées contre chaque analyseur
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* STRATÉGIES DÉTECTEUR PAR DÉTECTEUR
|
|
* Chaque classe implémente une approche spécialisée
|
|
*/
|
|
|
|
class BaseDetectorStrategy {
|
|
constructor(name) {
|
|
this.name = name;
|
|
this.effectiveness = 0.8;
|
|
this.targetMetrics = [];
|
|
}
|
|
|
|
/**
|
|
* Générer instructions spécifiques pour ce détecteur
|
|
*/
|
|
generateInstructions(elementType, personality, csvData) {
|
|
throw new Error('generateInstructions must be implemented by subclass');
|
|
}
|
|
|
|
/**
|
|
* Obtenir instructions anti-détection (NOUVEAU pour modularité)
|
|
*/
|
|
getInstructions(intensity = 1.0) {
|
|
throw new Error('getInstructions must be implemented by subclass');
|
|
}
|
|
|
|
/**
|
|
* Obtenir conseils d'amélioration (NOUVEAU pour modularité)
|
|
*/
|
|
getEnhancementTips(intensity = 1.0) {
|
|
throw new Error('getEnhancementTips must be implemented by subclass');
|
|
}
|
|
|
|
/**
|
|
* Analyser efficacité contre ce détecteur
|
|
*/
|
|
analyzeEffectiveness(content) {
|
|
return {
|
|
detector: this.name,
|
|
effectiveness: this.effectiveness,
|
|
metrics: this.analyzeContent(content)
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Analyser contenu selon métriques de ce détecteur
|
|
*/
|
|
analyzeContent(content) {
|
|
return {
|
|
wordCount: content.split(/\s+/).length,
|
|
sentenceCount: content.split(/[.!?]+/).length
|
|
};
|
|
}
|
|
}
|
|
|
|
/**
|
|
* STRATÉGIE ANTI-GPTZERO
|
|
* Focus: Imprévisibilité et variation syntaxique
|
|
*/
|
|
class GPTZeroStrategy extends BaseDetectorStrategy {
|
|
constructor() {
|
|
super('GPTZero');
|
|
this.effectiveness = 0.9;
|
|
this.targetMetrics = ['perplexity', 'burstiness', 'unpredictability'];
|
|
|
|
this.techniques = {
|
|
syntaxVariation: {
|
|
name: 'Variation syntaxique drastique',
|
|
rules: [
|
|
'Alterne phrases simples (5-8 mots) et complexes (25-35 mots)',
|
|
'Utilise structures atypiques : inversion, anacoluthe, ellipse',
|
|
'Intercale incises, parenthèses, tirets pour briser linéarité',
|
|
'Varie types : déclarative → interrogative → exclamative'
|
|
]
|
|
},
|
|
lexicalUnpredictability: {
|
|
name: 'Imprévisibilité lexicale',
|
|
rules: [
|
|
'Synonymes rares ou détournés : "performant" → "qui tient la route"',
|
|
'Néologismes justifiés : "sur-mesurable" pour "très mesurable"',
|
|
'Registres mélangés : soutenu puis familier dans même paragraphe',
|
|
'Évite collocations attendues : "haute qualité" → "qualité qui dépote"'
|
|
]
|
|
},
|
|
narrativeDisruption: {
|
|
name: 'Disruption narrative',
|
|
rules: [
|
|
'Apartés personnels : "(j\'ai testé, c\'est bluffant)"',
|
|
'Questions rhétoriques inattendues au milieu des faits',
|
|
'Changements de perspective soudains : technique → humain → commercial',
|
|
'Digressions courtes puis retour au sujet'
|
|
]
|
|
}
|
|
};
|
|
}
|
|
|
|
generateInstructions(elementType, personality, csvData) {
|
|
const instructions = [];
|
|
|
|
instructions.push(`ANTI-GPTZERO - MAXIMUM IMPRÉVISIBILITÉ:`);
|
|
|
|
// Techniques syntaxiques
|
|
instructions.push(`\nSYNTAXE VARIABLE:`);
|
|
this.techniques.syntaxVariation.rules.forEach(rule => {
|
|
instructions.push(`• ${rule}`);
|
|
});
|
|
|
|
// Imprévisibilité lexicale
|
|
instructions.push(`\nLEXIQUE IMPRÉVISIBLE:`);
|
|
this.techniques.lexicalUnpredictability.rules.forEach(rule => {
|
|
instructions.push(`• ${rule}`);
|
|
});
|
|
|
|
// Adaptation selon type élément
|
|
if (elementType === 'titre_h1' || elementType === 'titre_h2') {
|
|
instructions.push(`\nTITRES ANTI-GPTZERO:`);
|
|
instructions.push(`• Évite formules attendues : "Guide complet" → "Le vrai topo sur"`);
|
|
instructions.push(`• Structure atypique : question, exclamation, fragment`);
|
|
} else if (elementType === 'texte') {
|
|
instructions.push(`\nTEXTES ANTI-GPTZERO:`);
|
|
this.techniques.narrativeDisruption.rules.forEach(rule => {
|
|
instructions.push(`• ${rule}`);
|
|
});
|
|
}
|
|
|
|
// Adaptation personnalité
|
|
if (personality) {
|
|
instructions.push(`\nAVEC STYLE ${personality.nom.toUpperCase()}:`);
|
|
instructions.push(`• Pousse ${personality.style} vers extrêmes : plus marqué, moins lisse`);
|
|
instructions.push(`• Accentue tics de langage : répétitions, expressions favorites`);
|
|
}
|
|
|
|
return instructions.join('\n');
|
|
}
|
|
|
|
/**
|
|
* Instructions modulaires pour AdversarialCore
|
|
*/
|
|
getInstructions(intensity = 1.0) {
|
|
const baseRules = [
|
|
'Surprends avec tournures inattendues et constructions atypiques',
|
|
'Évite phrases trop prévisibles ou patterns répétitifs',
|
|
'Utilise synonymes moins évidents et vocabulaire moins commun',
|
|
'Varie drastiquement la complexité syntaxique entre phrases'
|
|
];
|
|
|
|
const intensiveRules = [
|
|
'Intercale observations personnelles ou détours narratifs',
|
|
'Alterne registres de langue (soutenu → familier → technique)',
|
|
'Brise la logique linéaire avec des parenthèses, incises, apartés',
|
|
'Créé ruptures narratives et changements de perspective soudains'
|
|
];
|
|
|
|
return intensity >= 1.0 ? [...baseRules, ...intensiveRules] : baseRules;
|
|
}
|
|
|
|
/**
|
|
* Conseils d'amélioration pour enhancement method
|
|
*/
|
|
getEnhancementTips(intensity = 1.0) {
|
|
return [
|
|
'Remplace connecteurs prévisibles par variations inattendues',
|
|
'Ajoute incises courtes pour briser la linéarité',
|
|
'Varie longueurs phrases dans même paragraphe',
|
|
'Utilise synonymes moins courants mais naturels',
|
|
...(intensity > 0.8 ? [
|
|
'Insère questions rhétoriques ponctuelles',
|
|
'Ajoute nuances et hésitations authentiques'
|
|
] : [])
|
|
];
|
|
}
|
|
|
|
analyzeContent(content) {
|
|
const baseMetrics = super.analyzeContent(content);
|
|
|
|
// Analyse perplexité approximative
|
|
const sentences = content.split(/[.!?]+/).filter(s => s.trim().length > 5);
|
|
const sentenceLengths = sentences.map(s => s.split(/\s+/).length);
|
|
|
|
// Variance longueur (proxy pour burstiness)
|
|
const avgLength = sentenceLengths.reduce((a, b) => a + b, 0) / sentenceLengths.length;
|
|
const variance = sentenceLengths.reduce((acc, len) => acc + Math.pow(len - avgLength, 2), 0) / sentenceLengths.length;
|
|
const burstiness = Math.sqrt(variance) / avgLength;
|
|
|
|
// Diversité lexicale (proxy pour imprévisibilité)
|
|
const words = content.toLowerCase().split(/\s+/).filter(w => w.length > 2);
|
|
const uniqueWords = [...new Set(words)];
|
|
const lexicalDiversity = uniqueWords.length / words.length;
|
|
|
|
return {
|
|
...baseMetrics,
|
|
burstiness: Math.round(burstiness * 100) / 100,
|
|
lexicalDiversity: Math.round(lexicalDiversity * 100) / 100,
|
|
avgSentenceLength: Math.round(avgLength),
|
|
gptZeroRiskLevel: this.calculateGPTZeroRisk(burstiness, lexicalDiversity)
|
|
};
|
|
}
|
|
|
|
calculateGPTZeroRisk(burstiness, lexicalDiversity) {
|
|
// Heuristique : GPTZero détecte uniformité faible + diversité faible
|
|
const uniformityScore = Math.min(burstiness, 1) * 100;
|
|
const diversityScore = lexicalDiversity * 100;
|
|
const combinedScore = (uniformityScore + diversityScore) / 2;
|
|
|
|
if (combinedScore > 70) return 'low';
|
|
if (combinedScore > 40) return 'medium';
|
|
return 'high';
|
|
}
|
|
}
|
|
|
|
/**
|
|
* STRATÉGIE ANTI-ORIGINALITY
|
|
* Focus: Diversité sémantique et originalité
|
|
*/
|
|
class OriginalityStrategy extends BaseDetectorStrategy {
|
|
constructor() {
|
|
super('Originality');
|
|
this.effectiveness = 0.85;
|
|
this.targetMetrics = ['semantic_diversity', 'originality_score', 'vocabulary_range'];
|
|
|
|
this.techniques = {
|
|
semanticCreativity: {
|
|
name: 'Créativité sémantique',
|
|
rules: [
|
|
'Métaphores inattendues : "cette plaque, c\'est le passeport de votre façade"',
|
|
'Comparaisons originales : évite clichés, invente analogies',
|
|
'Reformulations créatives : "résistant aux intempéries" → "qui brave les saisons"',
|
|
'Néologismes justifiés et expressifs'
|
|
]
|
|
},
|
|
perspectiveShifting: {
|
|
name: 'Changements de perspective',
|
|
rules: [
|
|
'Angles multiples sur même info : technique → esthétique → pratique',
|
|
'Points de vue variés : fabricant, utilisateur, installateur, voisin',
|
|
'Temporalités mélangées : présent, futur proche, retour d\'expérience',
|
|
'Niveaux d\'abstraction : détail précis puis vue d\'ensemble'
|
|
]
|
|
},
|
|
linguisticInventiveness: {
|
|
name: 'Inventivité linguistique',
|
|
rules: [
|
|
'Jeux de mots subtils et expressions détournées',
|
|
'Régionalismes et références culturelles précises',
|
|
'Vocabulaire technique humanisé avec créativité',
|
|
'Rythmes et sonorités travaillés : allitérations, assonances'
|
|
]
|
|
}
|
|
};
|
|
}
|
|
|
|
generateInstructions(elementType, personality, csvData) {
|
|
const instructions = [];
|
|
|
|
instructions.push(`ANTI-ORIGINALITY - MAXIMUM CRÉATIVITÉ SÉMANTIQUE:`);
|
|
|
|
// Créativité sémantique
|
|
instructions.push(`\nCRÉATIVITÉ SÉMANTIQUE:`);
|
|
this.techniques.semanticCreativity.rules.forEach(rule => {
|
|
instructions.push(`• ${rule}`);
|
|
});
|
|
|
|
// Changements de perspective
|
|
instructions.push(`\nPERSPECTIVES MULTIPLES:`);
|
|
this.techniques.perspectiveShifting.rules.forEach(rule => {
|
|
instructions.push(`• ${rule}`);
|
|
});
|
|
|
|
// Spécialisation par élément
|
|
if (elementType === 'intro') {
|
|
instructions.push(`\nINTROS ANTI-ORIGINALITY:`);
|
|
instructions.push(`• Commence par angle totalement inattendu pour le sujet`);
|
|
instructions.push(`• Évite intro-types, réinvente présentation du sujet`);
|
|
instructions.push(`• Crée surprise puis retour naturel au cœur du sujet`);
|
|
} else if (elementType.includes('faq')) {
|
|
instructions.push(`\nFAQ ANTI-ORIGINALITY:`);
|
|
instructions.push(`• Questions vraiment originales, pas standard secteur`);
|
|
instructions.push(`• Réponses avec angles créatifs et exemples inédits`);
|
|
}
|
|
|
|
// Contexte métier créatif
|
|
if (csvData && csvData.mc0) {
|
|
instructions.push(`\nCRÉATIVITÉ CONTEXTUELLE ${csvData.mc0.toUpperCase()}:`);
|
|
instructions.push(`• Réinvente façon de parler de ${csvData.mc0}`);
|
|
instructions.push(`• Évite vocabulaire convenu du secteur, invente expressions`);
|
|
instructions.push(`• Trouve analogies originales spécifiques à ${csvData.mc0}`);
|
|
}
|
|
|
|
// Inventivité linguistique
|
|
instructions.push(`\nINVENTIVITÉ LINGUISTIQUE:`);
|
|
this.techniques.linguisticInventiveness.rules.forEach(rule => {
|
|
instructions.push(`• ${rule}`);
|
|
});
|
|
|
|
return instructions.join('\n');
|
|
}
|
|
|
|
/**
|
|
* Instructions modulaires pour AdversarialCore
|
|
*/
|
|
getInstructions(intensity = 1.0) {
|
|
const baseRules = [
|
|
'Vocabulaire TRÈS varié : évite répétitions même de synonymes',
|
|
'Structures phrases délibérément irrégulières et asymétriques',
|
|
'Changements angles fréquents : technique → personnel → général',
|
|
'Créativité sémantique : métaphores, comparaisons inattendues'
|
|
];
|
|
|
|
const intensiveRules = [
|
|
'Évite formulations académiques ou trop structurées',
|
|
'Intègre références culturelles, expressions régionales',
|
|
'Subvertis les attentes : commence par la fin, questionne l\'évidence',
|
|
'Réinvente façon de présenter informations basiques'
|
|
];
|
|
|
|
return intensity >= 1.0 ? [...baseRules, ...intensiveRules] : baseRules;
|
|
}
|
|
|
|
/**
|
|
* Conseils d'amélioration pour enhancement method
|
|
*/
|
|
getEnhancementTips(intensity = 1.0) {
|
|
return [
|
|
'Trouve synonymes créatifs et expressions détournées',
|
|
'Ajoute métaphores subtiles et comparaisons originales',
|
|
'Varie angles d\'approche dans même contenu',
|
|
'Utilise vocabulaire technique humanisé',
|
|
...(intensity > 0.8 ? [
|
|
'Insère références culturelles ou régionalismes',
|
|
'Crée néologismes justifiés et expressifs'
|
|
] : [])
|
|
];
|
|
}
|
|
|
|
analyzeContent(content) {
|
|
const baseMetrics = super.analyzeContent(content);
|
|
|
|
// Analyse diversité sémantique
|
|
const words = content.toLowerCase().split(/\s+/).filter(w => w.length > 3);
|
|
const uniqueWords = [...new Set(words)];
|
|
const semanticDiversity = uniqueWords.length / words.length;
|
|
|
|
// Détection créativité (heuristique)
|
|
const creativityIndicators = [
|
|
'comme', 'tel', 'sorte de', 'façon de', 'manière de', // métaphores
|
|
'(', ')', '"', // originalité structure
|
|
'?', '!', // variation tonale
|
|
];
|
|
|
|
const creativityCount = creativityIndicators.reduce((count, indicator) => {
|
|
return count + (content.match(new RegExp(indicator.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'), 'g')) || []).length;
|
|
}, 0);
|
|
|
|
const creativityScore = Math.min(100, (creativityCount / words.length) * 1000);
|
|
|
|
return {
|
|
...baseMetrics,
|
|
semanticDiversity: Math.round(semanticDiversity * 100) / 100,
|
|
creativityScore: Math.round(creativityScore),
|
|
uniqueWordRatio: Math.round((uniqueWords.length / words.length) * 100),
|
|
originalityRiskLevel: this.calculateOriginalityRisk(semanticDiversity, creativityScore)
|
|
};
|
|
}
|
|
|
|
calculateOriginalityRisk(semanticDiversity, creativityScore) {
|
|
const diversityScore = semanticDiversity * 100;
|
|
const combinedScore = (diversityScore + creativityScore) / 2;
|
|
|
|
if (combinedScore > 60) return 'low';
|
|
if (combinedScore > 35) return 'medium';
|
|
return 'high';
|
|
}
|
|
}
|
|
|
|
/**
|
|
* STRATÉGIE GÉNÉRALE
|
|
* Équilibre entre toutes les techniques
|
|
*/
|
|
class GeneralStrategy extends BaseDetectorStrategy {
|
|
constructor() {
|
|
super('General');
|
|
this.effectiveness = 0.75;
|
|
this.targetMetrics = ['overall_naturalness', 'human_likeness'];
|
|
}
|
|
|
|
generateInstructions(elementType, personality, csvData) {
|
|
const instructions = [];
|
|
|
|
instructions.push(`ANTI-DÉTECTION GÉNÉRALE - ÉQUILIBRE OPTIMAL:`);
|
|
|
|
// Instructions générales équilibrées
|
|
instructions.push(`\nNATUREL ET VARIÉ:`);
|
|
instructions.push(`• Évite mots typiques IA : "optimal", "comprehensive", "seamless", "robust"`);
|
|
instructions.push(`• Mélange phrases courtes (5-10 mots) et longues (15-25 mots) naturellement`);
|
|
instructions.push(`• Utilise expressions idiomatiques françaises authentiques`);
|
|
instructions.push(`• Intègre imperfections humaines : répétitions légères, hésitations`);
|
|
instructions.push(`• Change angle ou ton 1-2 fois selon longueur du contenu`);
|
|
|
|
// Adaptation personnalité équilibrée
|
|
if (personality) {
|
|
instructions.push(`\nSTYLE ${personality.nom} NATUREL:`);
|
|
instructions.push(`• Applique style ${personality.style} de façon authentique, pas forcée`);
|
|
instructions.push(`• Respecte vocabulaire naturel : ${personality.vocabulairePref || 'varié'}`);
|
|
instructions.push(`• Utilise connecteurs variés : ${personality.connecteursPref || 'naturels'}`);
|
|
}
|
|
|
|
return instructions.join('\n');
|
|
}
|
|
|
|
/**
|
|
* Instructions modulaires pour AdversarialCore
|
|
*/
|
|
getInstructions(intensity = 1.0) {
|
|
const baseRules = [
|
|
'Évite ces mots typiques IA : "optimal", "comprehensive", "seamless", "robust", "leverage"',
|
|
'Varie longueurs phrases : mélange phrases courtes (5-10 mots) ET longues (20-30 mots)',
|
|
'Utilise expressions françaises familières et tournures idiomatiques',
|
|
'Ajoute imperfections naturelles : répétitions légères, hésitations, reformulations'
|
|
];
|
|
|
|
const intensiveRules = [
|
|
'Change angle ou perspective 1-2 fois dans le texte',
|
|
'Préfère synonymes moins évidents : "bon" plutôt qu\'excellent", "solide" plutôt qu\'optimisé"',
|
|
'Intègre connecteurs variés et naturels selon le contexte',
|
|
'Simule variation naturelle d\'humeur et d\'énergie rédactionnelle'
|
|
];
|
|
|
|
return intensity >= 0.8 ? [...baseRules, ...intensiveRules] : baseRules;
|
|
}
|
|
|
|
/**
|
|
* Conseils d'amélioration pour enhancement method
|
|
*/
|
|
getEnhancementTips(intensity = 1.0) {
|
|
return [
|
|
'Remplace mots typiques IA par synonymes plus naturels',
|
|
'Ajoute nuances et hésitations : "peut-être", "généralement", "souvent"',
|
|
'Varie connecteurs pour éviter répétitions mécaniques',
|
|
'Personnalise avec observations subjectives légères',
|
|
...(intensity > 0.7 ? [
|
|
'Intègre "erreurs" humaines : corrections, précisions',
|
|
'Simule changement léger de ton ou d\'énergie'
|
|
] : [])
|
|
];
|
|
}
|
|
|
|
analyzeContent(content) {
|
|
const baseMetrics = super.analyzeContent(content);
|
|
|
|
// Métrique naturalité générale
|
|
const sentences = content.split(/[.!?]+/).filter(s => s.trim().length > 5);
|
|
const avgWordsPerSentence = baseMetrics.wordCount / baseMetrics.sentenceCount;
|
|
|
|
// Détection mots typiques IA
|
|
const aiWords = ['optimal', 'comprehensive', 'seamless', 'robust', 'leverage'];
|
|
const aiWordCount = aiWords.reduce((count, word) => {
|
|
return count + (content.toLowerCase().match(new RegExp(`\\b${word}\\b`, 'g')) || []).length;
|
|
}, 0);
|
|
|
|
const aiWordDensity = aiWordCount / baseMetrics.wordCount * 100;
|
|
const naturalness = Math.max(0, 100 - (aiWordDensity * 10) - Math.abs(avgWordsPerSentence - 15));
|
|
|
|
return {
|
|
...baseMetrics,
|
|
avgWordsPerSentence: Math.round(avgWordsPerSentence),
|
|
aiWordCount,
|
|
aiWordDensity: Math.round(aiWordDensity * 100) / 100,
|
|
naturalnessScore: Math.round(naturalness),
|
|
generalRiskLevel: naturalness > 70 ? 'low' : naturalness > 40 ? 'medium' : 'high'
|
|
};
|
|
}
|
|
}
|
|
|
|
/**
|
|
* FACTORY POUR CRÉER STRATÉGIES
|
|
*/
|
|
class DetectorStrategyFactory {
|
|
static strategies = {
|
|
'general': GeneralStrategy,
|
|
'gptZero': GPTZeroStrategy,
|
|
'originality': OriginalityStrategy
|
|
};
|
|
|
|
static createStrategy(detectorName) {
|
|
const StrategyClass = this.strategies[detectorName];
|
|
if (!StrategyClass) {
|
|
logSh(`⚠️ Stratégie inconnue: ${detectorName}, fallback vers général`, 'WARNING');
|
|
return new GeneralStrategy();
|
|
}
|
|
return new StrategyClass();
|
|
}
|
|
|
|
static getSupportedDetectors() {
|
|
return Object.keys(this.strategies).map(name => {
|
|
const strategy = this.createStrategy(name);
|
|
return {
|
|
name,
|
|
displayName: strategy.name,
|
|
effectiveness: strategy.effectiveness,
|
|
targetMetrics: strategy.targetMetrics
|
|
};
|
|
});
|
|
}
|
|
|
|
static analyzeContentAgainstAllDetectors(content) {
|
|
const results = {};
|
|
|
|
Object.keys(this.strategies).forEach(detectorName => {
|
|
const strategy = this.createStrategy(detectorName);
|
|
results[detectorName] = strategy.analyzeEffectiveness(content);
|
|
});
|
|
|
|
return results;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* FONCTION UTILITAIRE - SÉLECTION STRATÉGIE OPTIMALE
|
|
*/
|
|
function selectOptimalStrategy(elementType, personality, previousResults = {}) {
|
|
// Logique de sélection intelligente
|
|
|
|
// Si résultats précédents disponibles, adapter
|
|
if (previousResults.gptZero && previousResults.gptZero.effectiveness < 0.6) {
|
|
return 'gptZero'; // Renforcer anti-GPTZero
|
|
}
|
|
|
|
if (previousResults.originality && previousResults.originality.effectiveness < 0.6) {
|
|
return 'originality'; // Renforcer anti-Originality
|
|
}
|
|
|
|
// Sélection par type d'élément
|
|
if (elementType === 'titre_h1' || elementType === 'titre_h2') {
|
|
return 'gptZero'; // Titres bénéficient imprévisibilité
|
|
}
|
|
|
|
if (elementType === 'intro' || elementType === 'texte') {
|
|
return 'originality'; // Corps bénéficie créativité sémantique
|
|
}
|
|
|
|
if (elementType.includes('faq')) {
|
|
return 'general'; // FAQ équilibre naturalité
|
|
}
|
|
|
|
// Par personnalité
|
|
if (personality) {
|
|
if (personality.style === 'créatif' || personality.style === 'original') {
|
|
return 'originality';
|
|
}
|
|
if (personality.style === 'technique' || personality.style === 'expert') {
|
|
return 'gptZero';
|
|
}
|
|
}
|
|
|
|
return 'general'; // Fallback
|
|
}
|
|
|
|
module.exports = {
|
|
DetectorStrategyFactory,
|
|
GPTZeroStrategy,
|
|
OriginalityStrategy,
|
|
GeneralStrategy,
|
|
selectOptimalStrategy,
|
|
BaseDetectorStrategy
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/AdversarialCore.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ADVERSARIAL CORE - MOTEUR MODULAIRE
|
|
// Responsabilité: Moteur adversarial réutilisable sur tout contenu
|
|
// Architecture: Couches applicables à la demande
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { callLLM } = require('../LLMManager');
|
|
|
|
// Import stratégies et utilitaires
|
|
const { DetectorStrategyManager } = require('./DetectorStrategies');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - APPLICATION COUCHE ADVERSARIALE
|
|
* Input: contenu existant + configuration adversariale
|
|
* Output: contenu avec couche adversariale appliquée
|
|
*/
|
|
async function applyAdversarialLayer(existingContent, config = {}) {
|
|
return await tracer.run('AdversarialCore.applyAdversarialLayer()', async () => {
|
|
const {
|
|
detectorTarget = 'general',
|
|
intensity = 1.0,
|
|
method = 'regeneration', // 'regeneration' | 'enhancement' | 'hybrid'
|
|
preserveStructure = true,
|
|
csvData = null,
|
|
context = {}
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
adversarialLayer: true,
|
|
detectorTarget,
|
|
intensity,
|
|
method,
|
|
elementsCount: Object.keys(existingContent).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎯 APPLICATION COUCHE ADVERSARIALE: ${detectorTarget} (${method})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(existingContent).length} éléments | Intensité: ${intensity}`, 'INFO');
|
|
|
|
try {
|
|
// Initialiser stratégie détecteur
|
|
const detectorManager = new DetectorStrategyManager(detectorTarget);
|
|
const strategy = detectorManager.getStrategy();
|
|
|
|
// Appliquer méthode adversariale choisie
|
|
let adversarialContent = {};
|
|
|
|
switch (method) {
|
|
case 'regeneration':
|
|
adversarialContent = await applyRegenerationMethod(existingContent, config, strategy);
|
|
break;
|
|
case 'enhancement':
|
|
adversarialContent = await applyEnhancementMethod(existingContent, config, strategy);
|
|
break;
|
|
case 'hybrid':
|
|
adversarialContent = await applyHybridMethod(existingContent, config, strategy);
|
|
break;
|
|
default:
|
|
throw new Error(`Méthode adversariale inconnue: ${method}`);
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
elementsProcessed: Object.keys(existingContent).length,
|
|
elementsModified: countModifiedElements(existingContent, adversarialContent),
|
|
detectorTarget,
|
|
intensity,
|
|
method,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ COUCHE ADVERSARIALE APPLIQUÉE: ${stats.elementsModified}/${stats.elementsProcessed} modifiés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Couche adversariale appliquée', stats);
|
|
|
|
return {
|
|
content: adversarialContent,
|
|
stats,
|
|
original: existingContent,
|
|
config
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ COUCHE ADVERSARIALE ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: retourner contenu original
|
|
logSh(`🔄 Fallback: contenu original conservé`, 'WARNING');
|
|
return {
|
|
content: existingContent,
|
|
stats: { fallback: true, duration },
|
|
original: existingContent,
|
|
config,
|
|
error: error.message
|
|
};
|
|
}
|
|
}, { existingContent: Object.keys(existingContent), config });
|
|
}
|
|
|
|
/**
|
|
* MÉTHODE RÉGÉNÉRATION - Réécrire complètement avec prompts adversariaux
|
|
*/
|
|
async function applyRegenerationMethod(existingContent, config, strategy) {
|
|
logSh(`🔄 Méthode régénération adversariale`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const contentEntries = Object.entries(existingContent);
|
|
|
|
// Traiter en chunks pour éviter timeouts
|
|
const chunks = chunkArray(contentEntries, 4);
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
logSh(` 📦 Régénération chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
try {
|
|
const regenerationPrompt = createRegenerationPrompt(chunk, config, strategy);
|
|
|
|
const response = await callLLM('claude', regenerationPrompt, {
|
|
temperature: 0.7 + (config.intensity * 0.2), // Température variable selon intensité
|
|
maxTokens: 2000 * chunk.length
|
|
}, config.csvData?.personality);
|
|
|
|
const chunkResults = parseRegenerationResponse(response, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk ${chunkIndex + 1}: ${Object.keys(chunkResults).length} éléments régénérés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: garder contenu original pour ce chunk
|
|
chunk.forEach(([tag, content]) => {
|
|
results[tag] = content;
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* MÉTHODE ENHANCEMENT - Améliorer sans réécrire complètement
|
|
*/
|
|
async function applyEnhancementMethod(existingContent, config, strategy) {
|
|
logSh(`🔧 Méthode enhancement adversarial`, 'DEBUG');
|
|
|
|
const results = { ...existingContent }; // Base: contenu original
|
|
const elementsToEnhance = selectElementsForEnhancement(existingContent, config);
|
|
|
|
if (elementsToEnhance.length === 0) {
|
|
logSh(` ⏭️ Aucun élément nécessite enhancement`, 'DEBUG');
|
|
return results;
|
|
}
|
|
|
|
logSh(` 📋 ${elementsToEnhance.length} éléments sélectionnés pour enhancement`, 'DEBUG');
|
|
|
|
const enhancementPrompt = createEnhancementPrompt(elementsToEnhance, config, strategy);
|
|
|
|
try {
|
|
const response = await callLLM('gpt4', enhancementPrompt, {
|
|
temperature: 0.5 + (config.intensity * 0.3),
|
|
maxTokens: 3000
|
|
}, config.csvData?.personality);
|
|
|
|
const enhancedResults = parseEnhancementResponse(response, elementsToEnhance);
|
|
|
|
// Appliquer améliorations
|
|
Object.keys(enhancedResults).forEach(tag => {
|
|
if (enhancedResults[tag] !== existingContent[tag]) {
|
|
results[tag] = enhancedResults[tag];
|
|
}
|
|
});
|
|
|
|
return results;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Enhancement échoué: ${error.message}`, 'ERROR');
|
|
return results; // Fallback: contenu original
|
|
}
|
|
}
|
|
|
|
/**
|
|
* MÉTHODE HYBRIDE - Combinaison régénération + enhancement
|
|
*/
|
|
async function applyHybridMethod(existingContent, config, strategy) {
|
|
logSh(`⚡ Méthode hybride adversariale`, 'DEBUG');
|
|
|
|
// 1. Enhancement léger sur tout le contenu
|
|
const enhancedContent = await applyEnhancementMethod(existingContent, {
|
|
...config,
|
|
intensity: config.intensity * 0.6 // Intensité réduite pour enhancement
|
|
}, strategy);
|
|
|
|
// 2. Régénération ciblée sur éléments clés
|
|
const keyElements = selectKeyElementsForRegeneration(enhancedContent, config);
|
|
|
|
if (keyElements.length === 0) {
|
|
return enhancedContent;
|
|
}
|
|
|
|
const keyElementsContent = {};
|
|
keyElements.forEach(tag => {
|
|
keyElementsContent[tag] = enhancedContent[tag];
|
|
});
|
|
|
|
const regeneratedElements = await applyRegenerationMethod(keyElementsContent, {
|
|
...config,
|
|
intensity: config.intensity * 1.2 // Intensité augmentée pour régénération
|
|
}, strategy);
|
|
|
|
// 3. Merger résultats
|
|
const hybridContent = { ...enhancedContent };
|
|
Object.keys(regeneratedElements).forEach(tag => {
|
|
hybridContent[tag] = regeneratedElements[tag];
|
|
});
|
|
|
|
return hybridContent;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
/**
|
|
* Créer prompt de régénération adversariale
|
|
*/
|
|
function createRegenerationPrompt(chunk, config, strategy) {
|
|
const { detectorTarget, intensity, csvData } = config;
|
|
|
|
let prompt = `MISSION: Réécris ces contenus pour éviter détection par ${detectorTarget}.
|
|
|
|
TECHNIQUE ANTI-${detectorTarget.toUpperCase()}:
|
|
${strategy.getInstructions(intensity).join('\n')}
|
|
|
|
CONTENUS À RÉÉCRIRE:
|
|
|
|
${chunk.map(([tag, content], i) => `[${i + 1}] TAG: ${tag}
|
|
ORIGINAL: "${content}"`).join('\n\n')}
|
|
|
|
CONSIGNES:
|
|
- GARDE exactement le même message et informations factuelles
|
|
- CHANGE structure, vocabulaire, style pour éviter détection ${detectorTarget}
|
|
- Intensité adversariale: ${intensity.toFixed(2)}
|
|
${csvData?.personality ? `- Style: ${csvData.personality.nom} (${csvData.personality.style})` : ''}
|
|
|
|
IMPORTANT: Réponse DIRECTE par les contenus réécrits, pas d'explication.
|
|
|
|
FORMAT:
|
|
[1] Contenu réécrit anti-${detectorTarget}
|
|
[2] Contenu réécrit anti-${detectorTarget}
|
|
etc...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt d'enhancement adversarial
|
|
*/
|
|
function createEnhancementPrompt(elementsToEnhance, config, strategy) {
|
|
const { detectorTarget, intensity } = config;
|
|
|
|
let prompt = `MISSION: Améliore subtilement ces contenus pour réduire détection ${detectorTarget}.
|
|
|
|
AMÉLIORATIONS CIBLÉES:
|
|
${strategy.getEnhancementTips(intensity).join('\n')}
|
|
|
|
ÉLÉMENTS À AMÉLIORER:
|
|
|
|
${elementsToEnhance.map((element, i) => `[${i + 1}] TAG: ${element.tag}
|
|
CONTENU: "${element.content}"
|
|
PROBLÈME: ${element.detectionRisk}`).join('\n\n')}
|
|
|
|
CONSIGNES:
|
|
- Modifications LÉGÈRES et naturelles
|
|
- GARDE le fond du message intact
|
|
- Focus sur réduction détection ${detectorTarget}
|
|
- Intensité: ${intensity.toFixed(2)}
|
|
|
|
FORMAT:
|
|
[1] Contenu légèrement amélioré
|
|
[2] Contenu légèrement amélioré
|
|
etc...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse régénération
|
|
*/
|
|
function parseRegenerationResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const index = parseInt(match[1]) - 1;
|
|
const content = cleanAdversarialContent(match[2].trim());
|
|
if (index >= 0 && index < chunk.length) {
|
|
parsedItems[index] = content;
|
|
}
|
|
}
|
|
|
|
// Mapper aux vrais tags
|
|
chunk.forEach(([tag, originalContent], index) => {
|
|
if (parsedItems[index] && parsedItems[index].length > 10) {
|
|
results[tag] = parsedItems[index];
|
|
} else {
|
|
results[tag] = originalContent; // Fallback
|
|
logSh(`⚠️ Fallback régénération pour [${tag}]`, 'WARNING');
|
|
}
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse enhancement
|
|
*/
|
|
function parseEnhancementResponse(response, elementsToEnhance) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < elementsToEnhance.length) {
|
|
let enhancedContent = cleanAdversarialContent(match[2].trim());
|
|
const element = elementsToEnhance[index];
|
|
|
|
if (enhancedContent && enhancedContent.length > 10) {
|
|
results[element.tag] = enhancedContent;
|
|
} else {
|
|
results[element.tag] = element.content; // Fallback
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Sélectionner éléments pour enhancement
|
|
*/
|
|
function selectElementsForEnhancement(existingContent, config) {
|
|
const elements = [];
|
|
|
|
Object.entries(existingContent).forEach(([tag, content]) => {
|
|
const detectionRisk = assessDetectionRisk(content, config.detectorTarget);
|
|
|
|
if (detectionRisk.score > 0.6) { // Risque élevé
|
|
elements.push({
|
|
tag,
|
|
content,
|
|
detectionRisk: detectionRisk.reasons.join(', '),
|
|
priority: detectionRisk.score
|
|
});
|
|
}
|
|
});
|
|
|
|
// Trier par priorité (risque élevé en premier)
|
|
elements.sort((a, b) => b.priority - a.priority);
|
|
|
|
return elements;
|
|
}
|
|
|
|
/**
|
|
* Sélectionner éléments clés pour régénération (hybride)
|
|
*/
|
|
function selectKeyElementsForRegeneration(content, config) {
|
|
const keyTags = [];
|
|
|
|
Object.keys(content).forEach(tag => {
|
|
// Éléments clés: titres, intro, premiers paragraphes
|
|
if (tag.includes('Titre') || tag.includes('H1') || tag.includes('intro') ||
|
|
tag.includes('Introduction') || tag.includes('1')) {
|
|
keyTags.push(tag);
|
|
}
|
|
});
|
|
|
|
return keyTags.slice(0, 3); // Maximum 3 éléments clés
|
|
}
|
|
|
|
/**
|
|
* Évaluer risque de détection
|
|
*/
|
|
function assessDetectionRisk(content, detectorTarget) {
|
|
let score = 0;
|
|
const reasons = [];
|
|
|
|
// Indicateurs génériques de contenu IA
|
|
const aiWords = ['optimal', 'comprehensive', 'seamless', 'robust', 'leverage', 'cutting-edge'];
|
|
const aiCount = aiWords.reduce((count, word) => {
|
|
return count + (content.toLowerCase().includes(word) ? 1 : 0);
|
|
}, 0);
|
|
|
|
if (aiCount > 2) {
|
|
score += 0.4;
|
|
reasons.push('mots_typiques_ia');
|
|
}
|
|
|
|
// Structure trop parfaite
|
|
const sentences = content.split(/[.!?]+/).filter(s => s.trim().length > 10);
|
|
if (sentences.length > 2) {
|
|
const avgLength = sentences.reduce((sum, s) => sum + s.length, 0) / sentences.length;
|
|
const variance = sentences.reduce((sum, s) => sum + Math.pow(s.length - avgLength, 2), 0) / sentences.length;
|
|
const uniformity = 1 - (Math.sqrt(variance) / avgLength);
|
|
|
|
if (uniformity > 0.8) {
|
|
score += 0.3;
|
|
reasons.push('structure_uniforme');
|
|
}
|
|
}
|
|
|
|
// Spécifique selon détecteur
|
|
if (detectorTarget === 'gptZero') {
|
|
// GPTZero détecte la prévisibilité
|
|
if (content.includes('par ailleurs') && content.includes('en effet')) {
|
|
score += 0.3;
|
|
reasons.push('connecteurs_prévisibles');
|
|
}
|
|
}
|
|
|
|
return { score: Math.min(1, score), reasons };
|
|
}
|
|
|
|
/**
|
|
* Nettoyer contenu adversarial généré
|
|
*/
|
|
function cleanAdversarialContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(voici\s+)?le\s+contenu\s+(réécrit|amélioré)[:\s]*/gi, '');
|
|
content = content.replace(/^(bon,?\s*)?(alors,?\s*)?/gi, '');
|
|
content = content.replace(/\*\*[^*]+\*\*/g, '');
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
/**
|
|
* Compter éléments modifiés
|
|
*/
|
|
function countModifiedElements(original, modified) {
|
|
let count = 0;
|
|
|
|
Object.keys(original).forEach(tag => {
|
|
if (modified[tag] && modified[tag] !== original[tag]) {
|
|
count++;
|
|
}
|
|
});
|
|
|
|
return count;
|
|
}
|
|
|
|
/**
|
|
* Chunk array utility
|
|
*/
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
/**
|
|
* Sleep utility
|
|
*/
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
module.exports = {
|
|
applyAdversarialLayer, // ← MAIN ENTRY POINT MODULAIRE
|
|
applyRegenerationMethod,
|
|
applyEnhancementMethod,
|
|
applyHybridMethod,
|
|
assessDetectionRisk,
|
|
selectElementsForEnhancement
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/AdversarialPromptEngine.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ADVERSARIAL PROMPT ENGINE - NIVEAU 3
|
|
// Responsabilité: Générer instructions anti-détection pour prompts
|
|
// Anti-détection: Prévention par modification des prompts LLM
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* INSTRUCTIONS ADVERSARIALES PAR DÉTECTEUR
|
|
* Stratégies spécialisées pour contrer chaque détecteur IA
|
|
*/
|
|
const ADVERSARIAL_INSTRUCTIONS = {
|
|
general: {
|
|
title: 'CONTRAINTES ANTI-DÉTECTION GÉNÉRALES',
|
|
rules: [
|
|
'Évite ces mots typiques IA : "optimal", "comprehensive", "seamless", "robust", "leverage"',
|
|
'Varie longueurs phrases : mélange phrases courtes (5-10 mots) ET longues (20-30 mots)',
|
|
'Utilise expressions françaises familières et tournures idiomatiques',
|
|
'Ajoute imperfections naturelles : répétitions légères, hésitations, reformulations',
|
|
'Change angle ou perspective 1-2 fois dans le texte',
|
|
'Préfère synonymes moins évidents : "bon" plutôt qu\'"excellent", "solide" plutôt qu\'optimisé"',
|
|
'Intègre connecteurs variés et naturels selon le contexte'
|
|
],
|
|
intensity: 1.0,
|
|
weight: 0.8
|
|
},
|
|
|
|
gptZero: {
|
|
title: 'SPÉCIAL ANTI-GPTZERO',
|
|
rules: [
|
|
'Surprends avec tournures inattendues et constructions atypiques',
|
|
'Évite phrases trop prévisibles ou patterns répétitifs',
|
|
'Utilise synonymes moins évidents et vocabulaire moins commun',
|
|
'Varie drastiquement la complexité syntaxique entre phrases',
|
|
'Intercale observations personnelles ou détours narratifs',
|
|
'Alterne registres de langue (soutenu → familier → technique)',
|
|
'Brise la logique linéaire avec des parenthèses, incises, apartés'
|
|
],
|
|
intensity: 1.2,
|
|
weight: 0.9,
|
|
targetMetric: 'unpredictability'
|
|
},
|
|
|
|
originality: {
|
|
title: 'SPÉCIAL ANTI-ORIGINALITY',
|
|
rules: [
|
|
'Vocabulaire TRÈS varié : évite répétitions même de synonymes',
|
|
'Structures phrases délibérément irrégulières et asymétriques',
|
|
'Changements angles fréquents : technique → personnel → général',
|
|
'Créativité sémantique : métaphores, comparaisons inattendues',
|
|
'Évite formulations académiques ou trop structurées',
|
|
'Intègre références culturelles, expressions régionales',
|
|
'Subvertis les attentes : commence par la fin, questionne l\'évidence'
|
|
],
|
|
intensity: 1.3,
|
|
weight: 0.95,
|
|
targetMetric: 'semantic_diversity'
|
|
},
|
|
|
|
copyLeaks: {
|
|
title: 'SPÉCIAL ANTI-COPYLEAKS',
|
|
rules: [
|
|
'Reformule idées communes avec angles totalement originaux',
|
|
'Évite phrases-types et formulations standard du secteur',
|
|
'Personnalise chaque assertion avec exemples spécifiques',
|
|
'Réinvente la façon de présenter informations basiques',
|
|
'Utilise analogies et métaphores plutôt que descriptions directes',
|
|
'Fragmente informations techniques en observations pratiques',
|
|
'Transforme données factuelles en récits ou témoignages'
|
|
],
|
|
intensity: 1.1,
|
|
weight: 0.85,
|
|
targetMetric: 'originality_score'
|
|
},
|
|
|
|
winston: {
|
|
title: 'SPÉCIAL ANTI-WINSTON',
|
|
rules: [
|
|
'Évite cohérence stylistique trop parfaite entre paragraphes',
|
|
'Simule variation naturelle d\'humeur et d\'énergie rédactionnelle',
|
|
'Intègre "erreurs" humaines : répétitions, corrections, précisions',
|
|
'Varie niveau de détail : parfois précis, parfois elliptique',
|
|
'Alterne registres émotionnels : enthousiaste → neutre → critique',
|
|
'Inclus hésitations et nuances : "peut-être", "généralement", "souvent"',
|
|
'Personnalise avec opinions subjectives et préférences'
|
|
],
|
|
intensity: 1.0,
|
|
weight: 0.9,
|
|
targetMetric: 'human_variation'
|
|
}
|
|
};
|
|
|
|
/**
|
|
* INSTRUCTIONS PERSONNALISÉES PAR TYPE D'ÉLÉMENT
|
|
*/
|
|
const ELEMENT_SPECIFIC_INSTRUCTIONS = {
|
|
titre_h1: {
|
|
base: 'Crée un titre percutant mais naturel',
|
|
adversarial: 'Évite formules marketing lisses, préfère authentique et direct'
|
|
},
|
|
titre_h2: {
|
|
base: 'Génère un sous-titre informatif',
|
|
adversarial: 'Varie structure : question, affirmation, exclamation selon contexte'
|
|
},
|
|
intro: {
|
|
base: 'Rédige introduction engageante',
|
|
adversarial: 'Commence par angle inattendu : anecdote, constat, question rhétorique'
|
|
},
|
|
texte: {
|
|
base: 'Développe paragraphe informatif',
|
|
adversarial: 'Mélange informations factuelles et observations personnelles'
|
|
},
|
|
faq_question: {
|
|
base: 'Formule question client naturelle',
|
|
adversarial: 'Utilise formulations vraiment utilisées par clients, pas académiques'
|
|
},
|
|
faq_reponse: {
|
|
base: 'Réponds de façon experte et rassurante',
|
|
adversarial: 'Ajoute nuances, "ça dépend", précisions contextuelles comme humain'
|
|
}
|
|
};
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - GÉNÉRATEUR DE PROMPTS ADVERSARIAUX
|
|
* @param {string} basePrompt - Prompt de base
|
|
* @param {Object} config - Configuration adversariale
|
|
* @returns {string} - Prompt enrichi d'instructions anti-détection
|
|
*/
|
|
function createAdversarialPrompt(basePrompt, config = {}) {
|
|
return tracer.run('AdversarialPromptEngine.createAdversarialPrompt()', () => {
|
|
const {
|
|
detectorTarget = 'general',
|
|
intensity = 1.0,
|
|
elementType = 'generic',
|
|
personality = null,
|
|
contextualMode = true,
|
|
csvData = null,
|
|
debugMode = false
|
|
} = config;
|
|
|
|
tracer.annotate({
|
|
detectorTarget,
|
|
intensity,
|
|
elementType,
|
|
personalityStyle: personality?.style
|
|
});
|
|
|
|
try {
|
|
// 1. Sélectionner stratégie détecteur
|
|
const strategy = ADVERSARIAL_INSTRUCTIONS[detectorTarget] || ADVERSARIAL_INSTRUCTIONS.general;
|
|
|
|
// 2. Adapter intensité
|
|
const effectiveIntensity = intensity * (strategy.intensity || 1.0);
|
|
const shouldApplyStrategy = Math.random() < (strategy.weight || 0.8);
|
|
|
|
if (!shouldApplyStrategy && detectorTarget !== 'general') {
|
|
// Fallback sur stratégie générale
|
|
return createAdversarialPrompt(basePrompt, { ...config, detectorTarget: 'general' });
|
|
}
|
|
|
|
// 3. Construire instructions adversariales
|
|
const adversarialSection = buildAdversarialInstructions(strategy, {
|
|
elementType,
|
|
personality,
|
|
effectiveIntensity,
|
|
contextualMode,
|
|
csvData
|
|
});
|
|
|
|
// 4. Assembler prompt final
|
|
const enhancedPrompt = assembleEnhancedPrompt(basePrompt, adversarialSection, {
|
|
strategy,
|
|
elementType,
|
|
debugMode
|
|
});
|
|
|
|
if (debugMode) {
|
|
logSh(`🎯 Prompt adversarial généré: ${detectorTarget} (intensité: ${effectiveIntensity.toFixed(2)})`, 'DEBUG');
|
|
logSh(` Instructions: ${strategy.rules.length} règles appliquées`, 'DEBUG');
|
|
}
|
|
|
|
tracer.event('Prompt adversarial créé', {
|
|
detectorTarget,
|
|
rulesCount: strategy.rules.length,
|
|
promptLength: enhancedPrompt.length
|
|
});
|
|
|
|
return enhancedPrompt;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur génération prompt adversarial: ${error.message}`, 'ERROR');
|
|
// Fallback: retourner prompt original
|
|
return basePrompt;
|
|
}
|
|
}, config);
|
|
}
|
|
|
|
/**
|
|
* Construire section instructions adversariales
|
|
*/
|
|
function buildAdversarialInstructions(strategy, config) {
|
|
const { elementType, personality, effectiveIntensity, contextualMode, csvData } = config;
|
|
|
|
let instructions = `\n\n=== ${strategy.title} ===\n`;
|
|
|
|
// Règles de base de la stratégie
|
|
const activeRules = selectActiveRules(strategy.rules, effectiveIntensity);
|
|
activeRules.forEach(rule => {
|
|
instructions += `• ${rule}\n`;
|
|
});
|
|
|
|
// Instructions spécifiques au type d'élément
|
|
if (ELEMENT_SPECIFIC_INSTRUCTIONS[elementType]) {
|
|
const elementInstructions = ELEMENT_SPECIFIC_INSTRUCTIONS[elementType];
|
|
instructions += `\nSPÉCIFIQUE ${elementType.toUpperCase()}:\n`;
|
|
instructions += `• ${elementInstructions.adversarial}\n`;
|
|
}
|
|
|
|
// Adaptations personnalité
|
|
if (personality && contextualMode) {
|
|
const personalityAdaptations = generatePersonalityAdaptations(personality, strategy);
|
|
if (personalityAdaptations) {
|
|
instructions += `\nADAPTATION PERSONNALITÉ ${personality.nom.toUpperCase()}:\n`;
|
|
instructions += personalityAdaptations;
|
|
}
|
|
}
|
|
|
|
// Contexte métier si disponible
|
|
if (csvData && contextualMode) {
|
|
const contextualInstructions = generateContextualInstructions(csvData, strategy);
|
|
if (contextualInstructions) {
|
|
instructions += `\nCONTEXTE MÉTIER:\n`;
|
|
instructions += contextualInstructions;
|
|
}
|
|
}
|
|
|
|
instructions += `\nIMPORTANT: Ces contraintes doivent sembler naturelles, pas forcées.\n`;
|
|
|
|
return instructions;
|
|
}
|
|
|
|
/**
|
|
* Sélectionner règles actives selon intensité
|
|
*/
|
|
function selectActiveRules(allRules, intensity) {
|
|
if (intensity >= 1.0) {
|
|
return allRules; // Toutes les règles
|
|
}
|
|
|
|
// Sélection proportionnelle à l'intensité
|
|
const ruleCount = Math.ceil(allRules.length * intensity);
|
|
return allRules.slice(0, ruleCount);
|
|
}
|
|
|
|
/**
|
|
* Générer adaptations personnalité
|
|
*/
|
|
function generatePersonalityAdaptations(personality, strategy) {
|
|
if (!personality) return null;
|
|
|
|
const adaptations = [];
|
|
|
|
// Style de la personnalité
|
|
if (personality.style) {
|
|
adaptations.push(`• Respecte le style ${personality.style} de ${personality.nom} tout en appliquant les contraintes`);
|
|
}
|
|
|
|
// Vocabulaire préféré
|
|
if (personality.vocabulairePref) {
|
|
adaptations.push(`• Intègre vocabulaire naturel: ${personality.vocabulairePref}`);
|
|
}
|
|
|
|
// Connecteurs préférés
|
|
if (personality.connecteursPref) {
|
|
adaptations.push(`• Utilise connecteurs variés: ${personality.connecteursPref}`);
|
|
}
|
|
|
|
// Longueur phrases selon personnalité
|
|
if (personality.longueurPhrases) {
|
|
adaptations.push(`• Longueur phrases: ${personality.longueurPhrases} mais avec variation anti-détection`);
|
|
}
|
|
|
|
return adaptations.length > 0 ? adaptations.join('\n') + '\n' : null;
|
|
}
|
|
|
|
/**
|
|
* Générer instructions contextuelles métier
|
|
*/
|
|
function generateContextualInstructions(csvData, strategy) {
|
|
if (!csvData.mc0) return null;
|
|
|
|
const instructions = [];
|
|
|
|
// Contexte sujet
|
|
instructions.push(`• Sujet: ${csvData.mc0} - utilise terminologie naturelle du domaine`);
|
|
|
|
// Éviter jargon selon détecteur
|
|
if (strategy.targetMetric === 'unpredictability') {
|
|
instructions.push(`• Évite jargon technique trop prévisible, privilégie explications accessibles`);
|
|
} else if (strategy.targetMetric === 'semantic_diversity') {
|
|
instructions.push(`• Varie façons de nommer/décrire ${csvData.mc0} - synonymes créatifs`);
|
|
}
|
|
|
|
return instructions.join('\n') + '\n';
|
|
}
|
|
|
|
/**
|
|
* Assembler prompt final
|
|
*/
|
|
function assembleEnhancedPrompt(basePrompt, adversarialSection, config) {
|
|
const { strategy, elementType, debugMode } = config;
|
|
|
|
// Structure du prompt amélioré
|
|
let enhancedPrompt = basePrompt;
|
|
|
|
// Injecter instructions adversariales
|
|
enhancedPrompt += adversarialSection;
|
|
|
|
// Rappel final selon stratégie
|
|
if (strategy.targetMetric) {
|
|
enhancedPrompt += `\nOBJECTIF PRIORITAIRE: Maximiser ${strategy.targetMetric} tout en conservant qualité.\n`;
|
|
}
|
|
|
|
// Instructions de réponse
|
|
enhancedPrompt += `\nRÉPONDS DIRECTEMENT par le contenu demandé, en appliquant naturellement ces contraintes.`;
|
|
|
|
return enhancedPrompt;
|
|
}
|
|
|
|
/**
|
|
* Analyser efficacité d'un prompt adversarial
|
|
*/
|
|
function analyzePromptEffectiveness(originalPrompt, adversarialPrompt, generatedContent) {
|
|
const analysis = {
|
|
promptEnhancement: {
|
|
originalLength: originalPrompt.length,
|
|
adversarialLength: adversarialPrompt.length,
|
|
enhancementRatio: adversarialPrompt.length / originalPrompt.length,
|
|
instructionsAdded: (adversarialPrompt.match(/•/g) || []).length
|
|
},
|
|
contentMetrics: analyzeGeneratedContent(generatedContent),
|
|
effectiveness: 0
|
|
};
|
|
|
|
// Score d'efficacité simple
|
|
analysis.effectiveness = Math.min(100,
|
|
(analysis.promptEnhancement.enhancementRatio - 1) * 50 +
|
|
analysis.contentMetrics.diversityScore
|
|
);
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* Analyser contenu généré
|
|
*/
|
|
function analyzeGeneratedContent(content) {
|
|
if (!content || typeof content !== 'string') {
|
|
return { diversityScore: 0, wordCount: 0, sentenceVariation: 0 };
|
|
}
|
|
|
|
const words = content.split(/\s+/).filter(w => w.length > 2);
|
|
const sentences = content.split(/[.!?]+/).filter(s => s.trim().length > 5);
|
|
|
|
// Diversité vocabulaire
|
|
const uniqueWords = [...new Set(words.map(w => w.toLowerCase()))];
|
|
const diversityScore = uniqueWords.length / Math.max(1, words.length) * 100;
|
|
|
|
// Variation longueurs phrases
|
|
const sentenceLengths = sentences.map(s => s.split(/\s+/).length);
|
|
const avgLength = sentenceLengths.reduce((a, b) => a + b, 0) / Math.max(1, sentenceLengths.length);
|
|
const variance = sentenceLengths.reduce((acc, len) => acc + Math.pow(len - avgLength, 2), 0) / Math.max(1, sentenceLengths.length);
|
|
const sentenceVariation = Math.sqrt(variance) / Math.max(1, avgLength) * 100;
|
|
|
|
return {
|
|
diversityScore: Math.round(diversityScore),
|
|
wordCount: words.length,
|
|
sentenceCount: sentences.length,
|
|
sentenceVariation: Math.round(sentenceVariation),
|
|
avgSentenceLength: Math.round(avgLength)
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Obtenir liste des détecteurs supportés
|
|
*/
|
|
function getSupportedDetectors() {
|
|
return Object.keys(ADVERSARIAL_INSTRUCTIONS).map(key => ({
|
|
id: key,
|
|
name: ADVERSARIAL_INSTRUCTIONS[key].title,
|
|
intensity: ADVERSARIAL_INSTRUCTIONS[key].intensity,
|
|
weight: ADVERSARIAL_INSTRUCTIONS[key].weight,
|
|
rulesCount: ADVERSARIAL_INSTRUCTIONS[key].rules.length,
|
|
targetMetric: ADVERSARIAL_INSTRUCTIONS[key].targetMetric || 'general'
|
|
}));
|
|
}
|
|
|
|
module.exports = {
|
|
createAdversarialPrompt, // ← MAIN ENTRY POINT
|
|
buildAdversarialInstructions,
|
|
analyzePromptEffectiveness,
|
|
analyzeGeneratedContent,
|
|
getSupportedDetectors,
|
|
ADVERSARIAL_INSTRUCTIONS,
|
|
ELEMENT_SPECIFIC_INSTRUCTIONS
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/AdversarialInitialGeneration.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ÉTAPE 1: GÉNÉRATION INITIALE ADVERSARIALE
|
|
// Responsabilité: Créer le contenu de base avec Claude + anti-détection
|
|
// LLM: Claude Sonnet (température 0.7) + Prompts adversariaux
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { createAdversarialPrompt } = require('./AdversarialPromptEngine');
|
|
const { DetectorStrategyManager } = require('./DetectorStrategies');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - GÉNÉRATION INITIALE ADVERSARIALE
|
|
* Input: { content: {}, csvData: {}, context: {}, adversarialConfig: {} }
|
|
* Output: { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function generateInitialContentAdversarial(input) {
|
|
return await tracer.run('AdversarialInitialGeneration.generateInitialContentAdversarial()', async () => {
|
|
const { hierarchy, csvData, context = {}, adversarialConfig = {} } = input;
|
|
|
|
// Configuration adversariale par défaut
|
|
const config = {
|
|
detectorTarget: adversarialConfig.detectorTarget || 'general',
|
|
intensity: adversarialConfig.intensity || 1.0,
|
|
enableAdaptiveStrategy: adversarialConfig.enableAdaptiveStrategy || true,
|
|
contextualMode: adversarialConfig.contextualMode !== false,
|
|
...adversarialConfig
|
|
};
|
|
|
|
// Initialiser manager détecteur
|
|
const detectorManager = new DetectorStrategyManager(config.detectorTarget);
|
|
|
|
await tracer.annotate({
|
|
step: '1/4',
|
|
llmProvider: 'claude',
|
|
elementsCount: Object.keys(hierarchy).length,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎯 ÉTAPE 1/4 ADVERSARIAL: Génération initiale (Claude + ${config.detectorTarget})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(hierarchy).length} éléments à générer`, 'INFO');
|
|
|
|
try {
|
|
// Collecter tous les éléments dans l'ordre XML
|
|
const allElements = collectElementsInXMLOrder(hierarchy);
|
|
|
|
// Séparer FAQ pairs et autres éléments
|
|
const { faqPairs, otherElements } = separateElementTypes(allElements);
|
|
|
|
// Générer en chunks pour éviter timeouts
|
|
const results = {};
|
|
|
|
// 1. Générer éléments normaux avec prompts adversariaux
|
|
if (otherElements.length > 0) {
|
|
const normalResults = await generateNormalElementsAdversarial(otherElements, csvData, config, detectorManager);
|
|
Object.assign(results, normalResults);
|
|
}
|
|
|
|
// 2. Générer paires FAQ adversariales si présentes
|
|
if (faqPairs.length > 0) {
|
|
const faqResults = await generateFAQPairsAdversarial(faqPairs, csvData, config, detectorManager);
|
|
Object.assign(results, faqResults);
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: Object.keys(results).length,
|
|
generated: Object.keys(results).length,
|
|
faqPairs: faqPairs.length,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ ÉTAPE 1/4 TERMINÉE: ${stats.generated} éléments générés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event(`Génération initiale terminée`, stats);
|
|
|
|
return {
|
|
content: results,
|
|
stats,
|
|
debug: {
|
|
llmProvider: 'claude',
|
|
step: 1,
|
|
elementsGenerated: Object.keys(results),
|
|
adversarialConfig: config,
|
|
detectorTarget: config.detectorTarget,
|
|
intensity: config.intensity
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ÉTAPE 1/4 ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`InitialGeneration failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Générer éléments normaux avec prompts adversariaux en chunks
|
|
*/
|
|
async function generateNormalElementsAdversarial(elements, csvData, adversarialConfig, detectorManager) {
|
|
logSh(`🎯 Génération éléments normaux adversariaux: ${elements.length} éléments`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(elements, 4); // Chunks de 4 pour éviter timeouts
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
logSh(` 📦 Chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
try {
|
|
const basePrompt = createBatchPrompt(chunk, csvData);
|
|
|
|
// Générer prompt adversarial
|
|
const adversarialPrompt = createAdversarialPrompt(basePrompt, {
|
|
detectorTarget: adversarialConfig.detectorTarget,
|
|
intensity: adversarialConfig.intensity,
|
|
elementType: getElementTypeFromChunk(chunk),
|
|
personality: csvData.personality,
|
|
contextualMode: adversarialConfig.contextualMode,
|
|
csvData: csvData,
|
|
debugMode: false
|
|
});
|
|
|
|
const response = await callLLM('claude', adversarialPrompt, {
|
|
temperature: 0.7,
|
|
maxTokens: 2000 * chunk.length
|
|
}, csvData.personality);
|
|
|
|
const chunkResults = parseBatchResponse(response, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk ${chunkIndex + 1}: ${Object.keys(chunkResults).length} éléments générés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Générer paires FAQ adversariales cohérentes
|
|
*/
|
|
async function generateFAQPairsAdversarial(faqPairs, csvData, adversarialConfig, detectorManager) {
|
|
logSh(`🎯 Génération paires FAQ adversariales: ${faqPairs.length} paires`, 'DEBUG');
|
|
|
|
const basePrompt = createFAQPairsPrompt(faqPairs, csvData);
|
|
|
|
// Générer prompt adversarial spécialisé FAQ
|
|
const adversarialPrompt = createAdversarialPrompt(basePrompt, {
|
|
detectorTarget: adversarialConfig.detectorTarget,
|
|
intensity: adversarialConfig.intensity * 1.1, // Intensité légèrement plus élevée pour FAQ
|
|
elementType: 'faq_mixed',
|
|
personality: csvData.personality,
|
|
contextualMode: adversarialConfig.contextualMode,
|
|
csvData: csvData,
|
|
debugMode: false
|
|
});
|
|
|
|
const response = await callLLM('claude', adversarialPrompt, {
|
|
temperature: 0.8,
|
|
maxTokens: 3000
|
|
}, csvData.personality);
|
|
|
|
return parseFAQResponse(response, faqPairs);
|
|
}
|
|
|
|
/**
|
|
* Créer prompt batch pour éléments normaux
|
|
*/
|
|
function createBatchPrompt(elements, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `=== GÉNÉRATION CONTENU INITIAL ===
|
|
Entreprise: Autocollant.fr - signalétique personnalisée
|
|
Sujet: ${csvData.mc0}
|
|
Rédacteur: ${personality.nom} (${personality.style})
|
|
|
|
ÉLÉMENTS À GÉNÉRER:
|
|
|
|
`;
|
|
|
|
elements.forEach((elementInfo, index) => {
|
|
const cleanTag = elementInfo.tag.replace(/\|/g, '');
|
|
prompt += `${index + 1}. [${cleanTag}] - ${getElementDescription(elementInfo)}\n`;
|
|
});
|
|
|
|
prompt += `
|
|
STYLE ${personality.nom.toUpperCase()}:
|
|
- Vocabulaire: ${personality.vocabulairePref}
|
|
- Phrases: ${personality.longueurPhrases}
|
|
- Niveau: ${personality.niveauTechnique}
|
|
|
|
CONSIGNES:
|
|
- Contenu SEO optimisé pour ${csvData.mc0}
|
|
- Style ${personality.style} naturel
|
|
- Pas de références techniques dans contenu
|
|
- RÉPONSE DIRECTE par le contenu
|
|
|
|
FORMAT:
|
|
[${elements[0].tag.replace(/\|/g, '')}]
|
|
Contenu généré...
|
|
|
|
[${elements[1] ? elements[1].tag.replace(/\|/g, '') : 'element2'}]
|
|
Contenu généré...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse batch
|
|
*/
|
|
function parseBatchResponse(response, elements) {
|
|
const results = {};
|
|
const regex = /\[([^\]]+)\]\s*([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
const content = cleanGeneratedContent(match[2].trim());
|
|
parsedItems[tag] = content;
|
|
}
|
|
|
|
// Mapper aux vrais tags
|
|
elements.forEach(element => {
|
|
const cleanTag = element.tag.replace(/\|/g, '');
|
|
if (parsedItems[cleanTag] && parsedItems[cleanTag].length > 10) {
|
|
results[element.tag] = parsedItems[cleanTag];
|
|
} else {
|
|
results[element.tag] = `Contenu professionnel pour ${element.element.name || cleanTag}`;
|
|
logSh(`⚠️ Fallback pour [${cleanTag}]`, 'WARNING');
|
|
}
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt pour paires FAQ
|
|
*/
|
|
function createFAQPairsPrompt(faqPairs, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `=== GÉNÉRATION PAIRES FAQ ===
|
|
Sujet: ${csvData.mc0}
|
|
Rédacteur: ${personality.nom} (${personality.style})
|
|
|
|
PAIRES À GÉNÉRER:
|
|
`;
|
|
|
|
faqPairs.forEach((pair, index) => {
|
|
const qTag = pair.question.tag.replace(/\|/g, '');
|
|
const aTag = pair.answer.tag.replace(/\|/g, '');
|
|
prompt += `${index + 1}. [${qTag}] + [${aTag}]\n`;
|
|
});
|
|
|
|
prompt += `
|
|
CONSIGNES:
|
|
- Questions naturelles de clients
|
|
- Réponses expertes ${personality.style}
|
|
- Couvrir: prix, livraison, personnalisation
|
|
|
|
FORMAT:
|
|
[${faqPairs[0].question.tag.replace(/\|/g, '')}]
|
|
Question client naturelle ?
|
|
|
|
[${faqPairs[0].answer.tag.replace(/\|/g, '')}]
|
|
Réponse utile et rassurante.`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse FAQ
|
|
*/
|
|
function parseFAQResponse(response, faqPairs) {
|
|
const results = {};
|
|
const regex = /\[([^\]]+)\]\s*([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
const content = cleanGeneratedContent(match[2].trim());
|
|
parsedItems[tag] = content;
|
|
}
|
|
|
|
// Mapper aux paires FAQ
|
|
faqPairs.forEach(pair => {
|
|
const qCleanTag = pair.question.tag.replace(/\|/g, '');
|
|
const aCleanTag = pair.answer.tag.replace(/\|/g, '');
|
|
|
|
if (parsedItems[qCleanTag]) results[pair.question.tag] = parsedItems[qCleanTag];
|
|
if (parsedItems[aCleanTag]) results[pair.answer.tag] = parsedItems[aCleanTag];
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
function collectElementsInXMLOrder(hierarchy) {
|
|
const allElements = [];
|
|
|
|
Object.keys(hierarchy).forEach(path => {
|
|
const section = hierarchy[path];
|
|
|
|
if (section.title) {
|
|
allElements.push({
|
|
tag: section.title.originalElement.originalTag,
|
|
element: section.title.originalElement,
|
|
type: section.title.originalElement.type
|
|
});
|
|
}
|
|
|
|
if (section.text) {
|
|
allElements.push({
|
|
tag: section.text.originalElement.originalTag,
|
|
element: section.text.originalElement,
|
|
type: section.text.originalElement.type
|
|
});
|
|
}
|
|
|
|
section.questions.forEach(q => {
|
|
allElements.push({
|
|
tag: q.originalElement.originalTag,
|
|
element: q.originalElement,
|
|
type: q.originalElement.type
|
|
});
|
|
});
|
|
});
|
|
|
|
return allElements;
|
|
}
|
|
|
|
function separateElementTypes(allElements) {
|
|
const faqPairs = [];
|
|
const otherElements = [];
|
|
const faqQuestions = {};
|
|
const faqAnswers = {};
|
|
|
|
// Collecter FAQ questions et answers
|
|
allElements.forEach(element => {
|
|
if (element.type === 'faq_question') {
|
|
const numberMatch = element.tag.match(/(\d+)/);
|
|
const faqNumber = numberMatch ? numberMatch[1] : '1';
|
|
faqQuestions[faqNumber] = element;
|
|
} else if (element.type === 'faq_reponse') {
|
|
const numberMatch = element.tag.match(/(\d+)/);
|
|
const faqNumber = numberMatch ? numberMatch[1] : '1';
|
|
faqAnswers[faqNumber] = element;
|
|
} else {
|
|
otherElements.push(element);
|
|
}
|
|
});
|
|
|
|
// Créer paires FAQ
|
|
Object.keys(faqQuestions).forEach(number => {
|
|
const question = faqQuestions[number];
|
|
const answer = faqAnswers[number];
|
|
|
|
if (question && answer) {
|
|
faqPairs.push({ number, question, answer });
|
|
} else if (question) {
|
|
otherElements.push(question);
|
|
} else if (answer) {
|
|
otherElements.push(answer);
|
|
}
|
|
});
|
|
|
|
return { faqPairs, otherElements };
|
|
}
|
|
|
|
function getElementDescription(elementInfo) {
|
|
switch (elementInfo.type) {
|
|
case 'titre_h1': return 'Titre principal accrocheur';
|
|
case 'titre_h2': return 'Titre de section';
|
|
case 'titre_h3': return 'Sous-titre';
|
|
case 'intro': return 'Introduction engageante';
|
|
case 'texte': return 'Paragraphe informatif';
|
|
default: return 'Contenu pertinent';
|
|
}
|
|
}
|
|
|
|
function cleanGeneratedContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?Titre_[HU]\d+_\d+[.,\s]*/gi, '');
|
|
content = content.replace(/\*\*[^*]+\*\*/g, '');
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
/**
|
|
* Helper: Déterminer type d'élément dominant dans un chunk
|
|
*/
|
|
function getElementTypeFromChunk(chunk) {
|
|
if (!chunk || chunk.length === 0) return 'generic';
|
|
|
|
// Compter les types dans le chunk
|
|
const typeCounts = {};
|
|
chunk.forEach(element => {
|
|
const type = element.type || 'generic';
|
|
typeCounts[type] = (typeCounts[type] || 0) + 1;
|
|
});
|
|
|
|
// Retourner type le plus fréquent
|
|
return Object.keys(typeCounts).reduce((a, b) =>
|
|
typeCounts[a] > typeCounts[b] ? a : b
|
|
);
|
|
}
|
|
|
|
module.exports = {
|
|
generateInitialContentAdversarial, // ← MAIN ENTRY POINT ADVERSARIAL
|
|
generateNormalElementsAdversarial,
|
|
generateFAQPairsAdversarial,
|
|
createBatchPrompt,
|
|
parseBatchResponse,
|
|
collectElementsInXMLOrder,
|
|
separateElementTypes,
|
|
getElementTypeFromChunk
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/AdversarialLayers.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ADVERSARIAL LAYERS - COUCHES MODULAIRES
|
|
// Responsabilité: Couches adversariales composables et réutilisables
|
|
// Architecture: Fonction pipeline |> layer1 |> layer2 |> layer3
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { applyAdversarialLayer } = require('./AdversarialCore');
|
|
|
|
/**
|
|
* COUCHE ANTI-GPTZEERO - Spécialisée contre GPTZero
|
|
*/
|
|
async function applyAntiGPTZeroLayer(content, options = {}) {
|
|
return await applyAdversarialLayer(content, {
|
|
detectorTarget: 'gptZero',
|
|
intensity: options.intensity || 1.0,
|
|
method: options.method || 'regeneration',
|
|
...options
|
|
});
|
|
}
|
|
|
|
/**
|
|
* COUCHE ANTI-ORIGINALITY - Spécialisée contre Originality.ai
|
|
*/
|
|
async function applyAntiOriginalityLayer(content, options = {}) {
|
|
return await applyAdversarialLayer(content, {
|
|
detectorTarget: 'originality',
|
|
intensity: options.intensity || 1.1,
|
|
method: options.method || 'hybrid',
|
|
...options
|
|
});
|
|
}
|
|
|
|
/**
|
|
* COUCHE ANTI-WINSTON - Spécialisée contre Winston AI
|
|
*/
|
|
async function applyAntiWinstonLayer(content, options = {}) {
|
|
return await applyAdversarialLayer(content, {
|
|
detectorTarget: 'winston',
|
|
intensity: options.intensity || 0.9,
|
|
method: options.method || 'enhancement',
|
|
...options
|
|
});
|
|
}
|
|
|
|
/**
|
|
* COUCHE GÉNÉRALE - Protection généraliste multi-détecteurs
|
|
*/
|
|
async function applyGeneralAdversarialLayer(content, options = {}) {
|
|
return await applyAdversarialLayer(content, {
|
|
detectorTarget: 'general',
|
|
intensity: options.intensity || 0.8,
|
|
method: options.method || 'hybrid',
|
|
...options
|
|
});
|
|
}
|
|
|
|
/**
|
|
* COUCHE LÉGÈRE - Modifications subtiles pour préserver qualité
|
|
*/
|
|
async function applyLightAdversarialLayer(content, options = {}) {
|
|
return await applyAdversarialLayer(content, {
|
|
detectorTarget: options.detectorTarget || 'general',
|
|
intensity: 0.5,
|
|
method: 'enhancement',
|
|
preserveStructure: true,
|
|
...options
|
|
});
|
|
}
|
|
|
|
/**
|
|
* COUCHE INTENSIVE - Maximum anti-détection
|
|
*/
|
|
async function applyIntensiveAdversarialLayer(content, options = {}) {
|
|
return await applyAdversarialLayer(content, {
|
|
detectorTarget: options.detectorTarget || 'gptZero',
|
|
intensity: 1.5,
|
|
method: 'regeneration',
|
|
preserveStructure: false,
|
|
...options
|
|
});
|
|
}
|
|
|
|
/**
|
|
* PIPELINE COMPOSABLE - Application séquentielle de couches
|
|
*/
|
|
async function applyLayerPipeline(content, layers = [], globalOptions = {}) {
|
|
return await tracer.run('AdversarialLayers.applyLayerPipeline()', async () => {
|
|
await tracer.annotate({
|
|
layersPipeline: true,
|
|
layersCount: layers.length,
|
|
elementsCount: Object.keys(content).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🔄 PIPELINE COUCHES ADVERSARIALES: ${layers.length} couches`, 'INFO');
|
|
|
|
let currentContent = content;
|
|
const pipelineStats = {
|
|
layers: [],
|
|
totalDuration: 0,
|
|
totalModifications: 0,
|
|
success: true
|
|
};
|
|
|
|
try {
|
|
for (let i = 0; i < layers.length; i++) {
|
|
const layer = layers[i];
|
|
const layerStartTime = Date.now();
|
|
|
|
logSh(` 🎯 Couche ${i + 1}/${layers.length}: ${layer.name || layer.type || 'anonyme'}`, 'DEBUG');
|
|
|
|
try {
|
|
const layerResult = await applyLayerByConfig(currentContent, layer, globalOptions);
|
|
|
|
currentContent = layerResult.content;
|
|
|
|
const layerStats = {
|
|
name: layer.name || `layer_${i + 1}`,
|
|
type: layer.type,
|
|
duration: Date.now() - layerStartTime,
|
|
modificationsCount: layerResult.stats?.elementsModified || 0,
|
|
success: true
|
|
};
|
|
|
|
pipelineStats.layers.push(layerStats);
|
|
pipelineStats.totalModifications += layerStats.modificationsCount;
|
|
|
|
logSh(` ✅ ${layerStats.name}: ${layerStats.modificationsCount} modifs (${layerStats.duration}ms)`, 'DEBUG');
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Couche ${i + 1} échouée: ${error.message}`, 'ERROR');
|
|
|
|
pipelineStats.layers.push({
|
|
name: layer.name || `layer_${i + 1}`,
|
|
type: layer.type,
|
|
duration: Date.now() - layerStartTime,
|
|
success: false,
|
|
error: error.message
|
|
});
|
|
|
|
// Continuer avec le contenu précédent si une couche échoue
|
|
if (!globalOptions.stopOnError) {
|
|
continue;
|
|
} else {
|
|
throw error;
|
|
}
|
|
}
|
|
}
|
|
|
|
pipelineStats.totalDuration = Date.now() - startTime;
|
|
pipelineStats.success = pipelineStats.layers.every(layer => layer.success);
|
|
|
|
logSh(`🔄 PIPELINE TERMINÉ: ${pipelineStats.totalModifications} modifs totales (${pipelineStats.totalDuration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Pipeline couches terminé', pipelineStats);
|
|
|
|
return {
|
|
content: currentContent,
|
|
stats: pipelineStats,
|
|
original: content
|
|
};
|
|
|
|
} catch (error) {
|
|
pipelineStats.totalDuration = Date.now() - startTime;
|
|
pipelineStats.success = false;
|
|
|
|
logSh(`❌ PIPELINE COUCHES ÉCHOUÉ après ${pipelineStats.totalDuration}ms: ${error.message}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}, { layers: layers.map(l => l.name || l.type), content: Object.keys(content) });
|
|
}
|
|
|
|
/**
|
|
* COUCHES PRÉDÉFINIES - Configurations courantes
|
|
*/
|
|
const PREDEFINED_LAYERS = {
|
|
// Stack défensif léger
|
|
lightDefense: [
|
|
{ type: 'general', name: 'General Light', intensity: 0.6, method: 'enhancement' },
|
|
{ type: 'anti-gptZero', name: 'GPTZero Light', intensity: 0.5, method: 'enhancement' }
|
|
],
|
|
|
|
// Stack défensif standard
|
|
standardDefense: [
|
|
{ type: 'general', name: 'General Standard', intensity: 0.8, method: 'hybrid' },
|
|
{ type: 'anti-gptZero', name: 'GPTZero Standard', intensity: 0.9, method: 'enhancement' },
|
|
{ type: 'anti-originality', name: 'Originality Standard', intensity: 0.8, method: 'enhancement' }
|
|
],
|
|
|
|
// Stack défensif intensif
|
|
heavyDefense: [
|
|
{ type: 'general', name: 'General Heavy', intensity: 1.0, method: 'regeneration' },
|
|
{ type: 'anti-gptZero', name: 'GPTZero Heavy', intensity: 1.2, method: 'regeneration' },
|
|
{ type: 'anti-originality', name: 'Originality Heavy', intensity: 1.1, method: 'hybrid' },
|
|
{ type: 'anti-winston', name: 'Winston Heavy', intensity: 1.0, method: 'enhancement' }
|
|
],
|
|
|
|
// Stack ciblé GPTZero
|
|
gptZeroFocused: [
|
|
{ type: 'anti-gptZero', name: 'GPTZero Primary', intensity: 1.3, method: 'regeneration' },
|
|
{ type: 'general', name: 'General Support', intensity: 0.7, method: 'enhancement' }
|
|
],
|
|
|
|
// Stack ciblé Originality
|
|
originalityFocused: [
|
|
{ type: 'anti-originality', name: 'Originality Primary', intensity: 1.4, method: 'hybrid' },
|
|
{ type: 'general', name: 'General Support', intensity: 0.8, method: 'enhancement' }
|
|
]
|
|
};
|
|
|
|
/**
|
|
* APPLIQUER STACK PRÉDÉFINI
|
|
*/
|
|
async function applyPredefinedStack(content, stackName, options = {}) {
|
|
const stack = PREDEFINED_LAYERS[stackName];
|
|
|
|
if (!stack) {
|
|
throw new Error(`Stack prédéfini inconnu: ${stackName}. Disponibles: ${Object.keys(PREDEFINED_LAYERS).join(', ')}`);
|
|
}
|
|
|
|
logSh(`📦 APPLICATION STACK PRÉDÉFINI: ${stackName}`, 'INFO');
|
|
|
|
return await applyLayerPipeline(content, stack, options);
|
|
}
|
|
|
|
/**
|
|
* COUCHES ADAPTATIVES - S'adaptent selon le contenu
|
|
*/
|
|
async function applyAdaptiveLayers(content, options = {}) {
|
|
const {
|
|
targetDetectors = ['gptZero', 'originality'],
|
|
maxIntensity = 1.0,
|
|
analysisMode = true
|
|
} = options;
|
|
|
|
logSh(`🧠 COUCHES ADAPTATIVES: Analyse + adaptation auto`, 'INFO');
|
|
|
|
// 1. Analyser le contenu pour détecter les risques
|
|
const contentAnalysis = analyzeContentRisks(content);
|
|
|
|
// 2. Construire pipeline adaptatif selon l'analyse
|
|
const adaptiveLayers = [];
|
|
|
|
// Niveau de base selon risque global
|
|
const baseIntensity = Math.min(maxIntensity, contentAnalysis.globalRisk * 1.2);
|
|
|
|
if (baseIntensity > 0.3) {
|
|
adaptiveLayers.push({
|
|
type: 'general',
|
|
name: 'Adaptive Base',
|
|
intensity: baseIntensity,
|
|
method: baseIntensity > 0.7 ? 'hybrid' : 'enhancement'
|
|
});
|
|
}
|
|
|
|
// Couches spécifiques selon détecteurs ciblés
|
|
targetDetectors.forEach(detector => {
|
|
const detectorRisk = contentAnalysis.detectorRisks[detector] || 0;
|
|
|
|
if (detectorRisk > 0.4) {
|
|
const intensity = Math.min(maxIntensity * 1.1, detectorRisk * 1.5);
|
|
adaptiveLayers.push({
|
|
type: `anti-${detector}`,
|
|
name: `Adaptive ${detector}`,
|
|
intensity,
|
|
method: intensity > 0.8 ? 'regeneration' : 'enhancement'
|
|
});
|
|
}
|
|
});
|
|
|
|
logSh(` 🎯 ${adaptiveLayers.length} couches adaptatives générées`, 'DEBUG');
|
|
|
|
if (adaptiveLayers.length === 0) {
|
|
logSh(` ✅ Contenu déjà optimal, aucune couche nécessaire`, 'INFO');
|
|
return { content, stats: { adaptive: true, layersApplied: 0 }, original: content };
|
|
}
|
|
|
|
return await applyLayerPipeline(content, adaptiveLayers, options);
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
/**
|
|
* Appliquer couche selon configuration
|
|
*/
|
|
async function applyLayerByConfig(content, layerConfig, globalOptions = {}) {
|
|
const { type, intensity, method, ...layerOptions } = layerConfig;
|
|
const options = { ...globalOptions, ...layerOptions, intensity, method };
|
|
|
|
switch (type) {
|
|
case 'general':
|
|
return await applyGeneralAdversarialLayer(content, options);
|
|
case 'anti-gptZero':
|
|
return await applyAntiGPTZeroLayer(content, options);
|
|
case 'anti-originality':
|
|
return await applyAntiOriginalityLayer(content, options);
|
|
case 'anti-winston':
|
|
return await applyAntiWinstonLayer(content, options);
|
|
case 'light':
|
|
return await applyLightAdversarialLayer(content, options);
|
|
case 'intensive':
|
|
return await applyIntensiveAdversarialLayer(content, options);
|
|
default:
|
|
throw new Error(`Type de couche inconnu: ${type}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Analyser risques du contenu pour adaptation
|
|
*/
|
|
function analyzeContentRisks(content) {
|
|
const analysis = {
|
|
globalRisk: 0,
|
|
detectorRisks: {},
|
|
riskFactors: []
|
|
};
|
|
|
|
const allContent = Object.values(content).join(' ');
|
|
|
|
// Risques génériques
|
|
let riskScore = 0;
|
|
|
|
// 1. Mots typiques IA
|
|
const aiWords = ['optimal', 'comprehensive', 'seamless', 'robust', 'leverage', 'cutting-edge', 'furthermore', 'moreover'];
|
|
const aiWordCount = aiWords.filter(word => allContent.toLowerCase().includes(word)).length;
|
|
|
|
if (aiWordCount > 2) {
|
|
riskScore += 0.3;
|
|
analysis.riskFactors.push(`mots_ia: ${aiWordCount}`);
|
|
}
|
|
|
|
// 2. Structure uniforme
|
|
const contentLengths = Object.values(content).map(c => c.length);
|
|
const avgLength = contentLengths.reduce((a, b) => a + b, 0) / contentLengths.length;
|
|
const variance = contentLengths.reduce((sum, len) => sum + Math.pow(len - avgLength, 2), 0) / contentLengths.length;
|
|
const uniformity = 1 - (Math.sqrt(variance) / Math.max(avgLength, 1));
|
|
|
|
if (uniformity > 0.8) {
|
|
riskScore += 0.2;
|
|
analysis.riskFactors.push(`uniformité: ${uniformity.toFixed(2)}`);
|
|
}
|
|
|
|
// 3. Connecteurs répétitifs
|
|
const repetitiveConnectors = ['par ailleurs', 'en effet', 'de plus', 'cependant'];
|
|
const connectorCount = repetitiveConnectors.filter(conn =>
|
|
(allContent.match(new RegExp(conn, 'gi')) || []).length > 1
|
|
).length;
|
|
|
|
if (connectorCount > 2) {
|
|
riskScore += 0.2;
|
|
analysis.riskFactors.push(`connecteurs_répétitifs: ${connectorCount}`);
|
|
}
|
|
|
|
analysis.globalRisk = Math.min(1, riskScore);
|
|
|
|
// Risques spécifiques par détecteur
|
|
analysis.detectorRisks = {
|
|
gptZero: analysis.globalRisk + (uniformity > 0.7 ? 0.3 : 0),
|
|
originality: analysis.globalRisk + (aiWordCount > 3 ? 0.4 : 0),
|
|
winston: analysis.globalRisk + (connectorCount > 2 ? 0.2 : 0)
|
|
};
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* Obtenir informations sur les stacks disponibles
|
|
*/
|
|
function getAvailableStacks() {
|
|
return Object.keys(PREDEFINED_LAYERS).map(stackName => ({
|
|
name: stackName,
|
|
layersCount: PREDEFINED_LAYERS[stackName].length,
|
|
description: getStackDescription(stackName),
|
|
layers: PREDEFINED_LAYERS[stackName]
|
|
}));
|
|
}
|
|
|
|
/**
|
|
* Description des stacks prédéfinis
|
|
*/
|
|
function getStackDescription(stackName) {
|
|
const descriptions = {
|
|
lightDefense: 'Protection légère préservant la qualité',
|
|
standardDefense: 'Protection équilibrée multi-détecteurs',
|
|
heavyDefense: 'Protection maximale tous détecteurs',
|
|
gptZeroFocused: 'Optimisation spécifique anti-GPTZero',
|
|
originalityFocused: 'Optimisation spécifique anti-Originality.ai'
|
|
};
|
|
|
|
return descriptions[stackName] || 'Stack personnalisé';
|
|
}
|
|
|
|
module.exports = {
|
|
// Couches individuelles
|
|
applyAntiGPTZeroLayer,
|
|
applyAntiOriginalityLayer,
|
|
applyAntiWinstonLayer,
|
|
applyGeneralAdversarialLayer,
|
|
applyLightAdversarialLayer,
|
|
applyIntensiveAdversarialLayer,
|
|
|
|
// Pipeline et stacks
|
|
applyLayerPipeline, // ← MAIN ENTRY POINT PIPELINE
|
|
applyPredefinedStack, // ← MAIN ENTRY POINT STACKS
|
|
applyAdaptiveLayers, // ← MAIN ENTRY POINT ADAPTATIF
|
|
|
|
// Utilitaires
|
|
getAvailableStacks,
|
|
analyzeContentRisks,
|
|
PREDEFINED_LAYERS
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/AdversarialStyleEnhancement.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ÉTAPE 4: ENHANCEMENT STYLE PERSONNALITÉ ADVERSARIAL
|
|
// Responsabilité: Appliquer le style personnalité avec Mistral + anti-détection
|
|
// LLM: Mistral (température 0.8) + Prompts adversariaux
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { createAdversarialPrompt } = require('./AdversarialPromptEngine');
|
|
const { DetectorStrategyManager } = require('./DetectorStrategies');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - ENHANCEMENT STYLE
|
|
* Input: { content: {}, csvData: {}, context: {} }
|
|
* Output: { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function applyPersonalityStyleAdversarial(input) {
|
|
return await tracer.run('AdversarialStyleEnhancement.applyPersonalityStyleAdversarial()', async () => {
|
|
const { content, csvData, context = {}, adversarialConfig = {} } = input;
|
|
|
|
// Configuration adversariale par défaut
|
|
const config = {
|
|
detectorTarget: adversarialConfig.detectorTarget || 'general',
|
|
intensity: adversarialConfig.intensity || 1.0,
|
|
enableAdaptiveStrategy: adversarialConfig.enableAdaptiveStrategy || true,
|
|
contextualMode: adversarialConfig.contextualMode !== false,
|
|
...adversarialConfig
|
|
};
|
|
|
|
// Initialiser manager détecteur
|
|
const detectorManager = new DetectorStrategyManager(config.detectorTarget);
|
|
|
|
await tracer.annotate({
|
|
step: '4/4',
|
|
llmProvider: 'mistral',
|
|
elementsCount: Object.keys(content).length,
|
|
personality: csvData.personality?.nom,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎯 ÉTAPE 4/4 ADVERSARIAL: Enhancement style ${csvData.personality?.nom} (Mistral + ${config.detectorTarget})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à styliser`, 'INFO');
|
|
|
|
try {
|
|
const personality = csvData.personality;
|
|
|
|
if (!personality) {
|
|
logSh(`⚠️ ÉTAPE 4/4: Aucune personnalité définie, style standard`, 'WARNING');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration: Date.now() - startTime },
|
|
debug: { llmProvider: 'mistral', step: 4, personalityApplied: 'none' }
|
|
};
|
|
}
|
|
|
|
// 1. Préparer éléments pour stylisation
|
|
const styleElements = prepareElementsForStyling(content);
|
|
|
|
// 2. Appliquer style en chunks avec prompts adversariaux
|
|
const styledResults = await applyStyleInChunksAdversarial(styleElements, csvData, config, detectorManager);
|
|
|
|
// 3. Merger résultats
|
|
const finalContent = { ...content };
|
|
let actuallyStyled = 0;
|
|
|
|
Object.keys(styledResults).forEach(tag => {
|
|
if (styledResults[tag] !== content[tag]) {
|
|
finalContent[tag] = styledResults[tag];
|
|
actuallyStyled++;
|
|
}
|
|
});
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: Object.keys(content).length,
|
|
enhanced: actuallyStyled,
|
|
personality: personality.nom,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ ÉTAPE 4/4 TERMINÉE: ${stats.enhanced} éléments stylisés ${personality.nom} (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event(`Enhancement style terminé`, stats);
|
|
|
|
return {
|
|
content: finalContent,
|
|
stats,
|
|
debug: {
|
|
llmProvider: 'mistral',
|
|
step: 4,
|
|
personalityApplied: personality.nom,
|
|
styleCharacteristics: {
|
|
vocabulaire: personality.vocabulairePref,
|
|
connecteurs: personality.connecteursPref,
|
|
style: personality.style
|
|
},
|
|
adversarialConfig: config,
|
|
detectorTarget: config.detectorTarget,
|
|
intensity: config.intensity
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ÉTAPE 4/4 ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: retourner contenu original si Mistral indisponible
|
|
logSh(`🔄 Fallback: contenu original conservé`, 'WARNING');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration },
|
|
debug: { llmProvider: 'mistral', step: 4, error: error.message, fallback: true }
|
|
};
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Préparer éléments pour stylisation
|
|
*/
|
|
function prepareElementsForStyling(content) {
|
|
const styleElements = [];
|
|
|
|
Object.keys(content).forEach(tag => {
|
|
const text = content[tag];
|
|
|
|
// Tous les éléments peuvent bénéficier d'adaptation personnalité
|
|
// Même les courts (titres) peuvent être adaptés au style
|
|
styleElements.push({
|
|
tag,
|
|
content: text,
|
|
priority: calculateStylePriority(text, tag)
|
|
});
|
|
});
|
|
|
|
// Trier par priorité (titres d'abord, puis textes longs)
|
|
styleElements.sort((a, b) => b.priority - a.priority);
|
|
|
|
return styleElements;
|
|
}
|
|
|
|
/**
|
|
* Calculer priorité de stylisation
|
|
*/
|
|
function calculateStylePriority(text, tag) {
|
|
let priority = 1.0;
|
|
|
|
// Titres = haute priorité (plus visible)
|
|
if (tag.includes('Titre') || tag.includes('H1') || tag.includes('H2')) {
|
|
priority += 0.5;
|
|
}
|
|
|
|
// Textes longs = priorité selon longueur
|
|
if (text.length > 200) {
|
|
priority += 0.3;
|
|
} else if (text.length > 100) {
|
|
priority += 0.2;
|
|
}
|
|
|
|
// Introduction = haute priorité
|
|
if (tag.includes('intro') || tag.includes('Introduction')) {
|
|
priority += 0.4;
|
|
}
|
|
|
|
return priority;
|
|
}
|
|
|
|
/**
|
|
* Appliquer style en chunks avec prompts adversariaux
|
|
*/
|
|
async function applyStyleInChunksAdversarial(styleElements, csvData, adversarialConfig, detectorManager) {
|
|
logSh(`🎯 Stylisation adversarial: ${styleElements.length} éléments selon ${csvData.personality.nom}`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(styleElements, 8); // Chunks de 8 pour Mistral
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(` 📦 Chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
const basePrompt = createStylePrompt(chunk, csvData);
|
|
|
|
// Générer prompt adversarial pour stylisation
|
|
const adversarialPrompt = createAdversarialPrompt(basePrompt, {
|
|
detectorTarget: adversarialConfig.detectorTarget,
|
|
intensity: adversarialConfig.intensity * 1.1, // Intensité plus élevée pour style (plus visible)
|
|
elementType: 'style_enhancement',
|
|
personality: csvData.personality,
|
|
contextualMode: adversarialConfig.contextualMode,
|
|
csvData: csvData,
|
|
debugMode: false
|
|
});
|
|
|
|
const styledResponse = await callLLM('mistral', adversarialPrompt, {
|
|
temperature: 0.8,
|
|
maxTokens: 3000
|
|
}, csvData.personality);
|
|
|
|
const chunkResults = parseStyleResponse(styledResponse, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk ${chunkIndex + 1}: ${Object.keys(chunkResults).length} stylisés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: garder contenu original
|
|
chunk.forEach(element => {
|
|
results[element.tag] = element.content;
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt de stylisation
|
|
*/
|
|
function createStylePrompt(chunk, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `MISSION: Adapte UNIQUEMENT le style de ces contenus selon ${personality.nom}.
|
|
|
|
CONTEXTE: Article SEO e-commerce ${csvData.mc0}
|
|
PERSONNALITÉ: ${personality.nom}
|
|
DESCRIPTION: ${personality.description}
|
|
STYLE: ${personality.style} adapté web professionnel
|
|
VOCABULAIRE: ${personality.vocabulairePref}
|
|
CONNECTEURS: ${personality.connecteursPref}
|
|
NIVEAU TECHNIQUE: ${personality.niveauTechnique}
|
|
LONGUEUR PHRASES: ${personality.longueurPhrases}
|
|
|
|
CONTENUS À STYLISER:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag} (Priorité: ${item.priority.toFixed(1)})
|
|
CONTENU: "${item.content}"`).join('\n\n')}
|
|
|
|
OBJECTIFS STYLISATION ${personality.nom.toUpperCase()}:
|
|
- Adapte le TON selon ${personality.style}
|
|
- Vocabulaire: ${personality.vocabulairePref}
|
|
- Connecteurs variés: ${personality.connecteursPref}
|
|
- Phrases: ${personality.longueurPhrases}
|
|
- Niveau: ${personality.niveauTechnique}
|
|
|
|
CONSIGNES STRICTES:
|
|
- GARDE le même contenu informatif et technique
|
|
- Adapte SEULEMENT ton, expressions, vocabulaire selon ${personality.nom}
|
|
- RESPECTE longueur approximative (±20%)
|
|
- ÉVITE répétitions excessives
|
|
- Style ${personality.nom} reconnaissable mais NATUREL web
|
|
- PAS de messages d'excuse
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu stylisé selon ${personality.nom}
|
|
[2] Contenu stylisé selon ${personality.nom}
|
|
etc...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse stylisation
|
|
*/
|
|
function parseStyleResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let styledContent = match[2].trim();
|
|
const element = chunk[index];
|
|
|
|
// Nettoyer le contenu stylisé
|
|
styledContent = cleanStyledContent(styledContent);
|
|
|
|
if (styledContent && styledContent.length > 10) {
|
|
results[element.tag] = styledContent;
|
|
logSh(`✅ Styled [${element.tag}]: "${styledContent.substring(0, 100)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content;
|
|
logSh(`⚠️ Fallback [${element.tag}]: stylisation invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < chunk.length) {
|
|
const element = chunk[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Nettoyer contenu stylisé
|
|
*/
|
|
function cleanStyledContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?voici\s+/gi, '');
|
|
content = content.replace(/^pour\s+ce\s+contenu[,\s]*/gi, '');
|
|
content = content.replace(/\*\*[^*]+\*\*/g, '');
|
|
|
|
// Réduire répétitions excessives mais garder le style personnalité
|
|
content = content.replace(/(du coup[,\s]+){4,}/gi, 'du coup ');
|
|
content = content.replace(/(bon[,\s]+){4,}/gi, 'bon ');
|
|
content = content.replace(/(franchement[,\s]+){3,}/gi, 'franchement ');
|
|
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
/**
|
|
* Obtenir instructions de style dynamiques
|
|
*/
|
|
function getPersonalityStyleInstructions(personality) {
|
|
if (!personality) return "Style professionnel standard";
|
|
|
|
return `STYLE ${personality.nom.toUpperCase()} (${personality.style}):
|
|
- Description: ${personality.description}
|
|
- Vocabulaire: ${personality.vocabulairePref || 'professionnel'}
|
|
- Connecteurs: ${personality.connecteursPref || 'par ailleurs, en effet'}
|
|
- Mots-clés: ${personality.motsClesSecteurs || 'technique, qualité'}
|
|
- Phrases: ${personality.longueurPhrases || 'Moyennes'}
|
|
- Niveau: ${personality.niveauTechnique || 'Accessible'}
|
|
- CTA: ${personality.ctaStyle || 'Professionnel'}`;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
module.exports = {
|
|
applyPersonalityStyleAdversarial, // ← MAIN ENTRY POINT ADVERSARIAL
|
|
prepareElementsForStyling,
|
|
calculateStylePriority,
|
|
applyStyleInChunksAdversarial,
|
|
createStylePrompt,
|
|
parseStyleResponse,
|
|
getPersonalityStyleInstructions
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/AdversarialTechnicalEnhancem… │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ÉTAPE 2: ENHANCEMENT TECHNIQUE ADVERSARIAL
|
|
// Responsabilité: Améliorer la précision technique avec GPT-4 + anti-détection
|
|
// LLM: GPT-4o-mini (température 0.4) + Prompts adversariaux
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { createAdversarialPrompt } = require('./AdversarialPromptEngine');
|
|
const { DetectorStrategyManager } = require('./DetectorStrategies');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - ENHANCEMENT TECHNIQUE ADVERSARIAL
|
|
* Input: { content: {}, csvData: {}, context: {}, adversarialConfig: {} }
|
|
* Output: { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function enhanceTechnicalTermsAdversarial(input) {
|
|
return await tracer.run('AdversarialTechnicalEnhancement.enhanceTechnicalTermsAdversarial()', async () => {
|
|
const { content, csvData, context = {}, adversarialConfig = {} } = input;
|
|
|
|
// Configuration adversariale par défaut
|
|
const config = {
|
|
detectorTarget: adversarialConfig.detectorTarget || 'general',
|
|
intensity: adversarialConfig.intensity || 1.0,
|
|
enableAdaptiveStrategy: adversarialConfig.enableAdaptiveStrategy || true,
|
|
contextualMode: adversarialConfig.contextualMode !== false,
|
|
...adversarialConfig
|
|
};
|
|
|
|
// Initialiser manager détecteur
|
|
const detectorManager = new DetectorStrategyManager(config.detectorTarget);
|
|
|
|
await tracer.annotate({
|
|
step: '2/4',
|
|
llmProvider: 'gpt4',
|
|
elementsCount: Object.keys(content).length,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎯 ÉTAPE 2/4 ADVERSARIAL: Enhancement technique (GPT-4 + ${config.detectorTarget})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à analyser`, 'INFO');
|
|
|
|
try {
|
|
// 1. Analyser tous les éléments pour détecter termes techniques (adversarial)
|
|
const technicalAnalysis = await analyzeTechnicalTermsAdversarial(content, csvData, config, detectorManager);
|
|
|
|
// 2. Filter les éléments qui ont besoin d'enhancement
|
|
const elementsNeedingEnhancement = technicalAnalysis.filter(item => item.needsEnhancement);
|
|
|
|
logSh(` 📋 Analyse: ${elementsNeedingEnhancement.length}/${Object.keys(content).length} éléments nécessitent enhancement`, 'INFO');
|
|
|
|
if (elementsNeedingEnhancement.length === 0) {
|
|
logSh(`✅ ÉTAPE 2/4: Aucun enhancement nécessaire`, 'INFO');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration: Date.now() - startTime },
|
|
debug: { llmProvider: 'gpt4', step: 2, enhancementsApplied: [] }
|
|
};
|
|
}
|
|
|
|
// 3. Améliorer les éléments sélectionnés avec prompts adversariaux
|
|
const enhancedResults = await enhanceSelectedElementsAdversarial(elementsNeedingEnhancement, csvData, config, detectorManager);
|
|
|
|
// 4. Merger avec contenu original
|
|
const finalContent = { ...content };
|
|
let actuallyEnhanced = 0;
|
|
|
|
Object.keys(enhancedResults).forEach(tag => {
|
|
if (enhancedResults[tag] !== content[tag]) {
|
|
finalContent[tag] = enhancedResults[tag];
|
|
actuallyEnhanced++;
|
|
}
|
|
});
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: Object.keys(content).length,
|
|
enhanced: actuallyEnhanced,
|
|
candidate: elementsNeedingEnhancement.length,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ ÉTAPE 2/4 TERMINÉE: ${stats.enhanced} éléments améliorés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event(`Enhancement technique terminé`, stats);
|
|
|
|
return {
|
|
content: finalContent,
|
|
stats,
|
|
debug: {
|
|
llmProvider: 'gpt4',
|
|
step: 2,
|
|
enhancementsApplied: Object.keys(enhancedResults),
|
|
technicalTermsFound: elementsNeedingEnhancement.map(e => e.technicalTerms),
|
|
adversarialConfig: config,
|
|
detectorTarget: config.detectorTarget,
|
|
intensity: config.intensity
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ÉTAPE 2/4 ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`TechnicalEnhancement failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Analyser tous les éléments pour détecter termes techniques (adversarial)
|
|
*/
|
|
async function analyzeTechnicalTermsAdversarial(content, csvData, adversarialConfig, detectorManager) {
|
|
logSh(`🎯 Analyse termes techniques adversarial batch`, 'DEBUG');
|
|
|
|
const contentEntries = Object.keys(content);
|
|
|
|
const analysisPrompt = `MISSION: Analyser ces ${contentEntries.length} contenus et identifier leurs termes techniques.
|
|
|
|
CONTEXTE: ${csvData.mc0} - Secteur: signalétique/impression
|
|
|
|
CONTENUS À ANALYSER:
|
|
|
|
${contentEntries.map((tag, i) => `[${i + 1}] TAG: ${tag}
|
|
CONTENU: "${content[tag]}"`).join('\n\n')}
|
|
|
|
CONSIGNES:
|
|
- Identifie UNIQUEMENT les vrais termes techniques métier/industrie
|
|
- Évite mots génériques (qualité, service, pratique, personnalisé)
|
|
- Focus: matériaux, procédés, normes, dimensions, technologies
|
|
- Si aucun terme technique → "AUCUN"
|
|
|
|
EXEMPLES VALIDES: dibond, impression UV, fraisage CNC, épaisseur 3mm
|
|
EXEMPLES INVALIDES: durable, pratique, personnalisé, moderne
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] dibond, impression UV OU AUCUN
|
|
[2] AUCUN
|
|
[3] aluminium, fraisage CNC OU AUCUN
|
|
etc...`;
|
|
|
|
try {
|
|
// Générer prompt adversarial pour analyse
|
|
const adversarialAnalysisPrompt = createAdversarialPrompt(analysisPrompt, {
|
|
detectorTarget: adversarialConfig.detectorTarget,
|
|
intensity: adversarialConfig.intensity * 0.8, // Intensité modérée pour analyse
|
|
elementType: 'technical_analysis',
|
|
personality: csvData.personality,
|
|
contextualMode: adversarialConfig.contextualMode,
|
|
csvData: csvData,
|
|
debugMode: false
|
|
});
|
|
|
|
const analysisResponse = await callLLM('gpt4', adversarialAnalysisPrompt, {
|
|
temperature: 0.3,
|
|
maxTokens: 2000
|
|
}, csvData.personality);
|
|
|
|
return parseAnalysisResponse(analysisResponse, content, contentEntries);
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Analyse termes techniques échouée: ${error.message}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Améliorer les éléments sélectionnés avec prompts adversariaux
|
|
*/
|
|
async function enhanceSelectedElementsAdversarial(elementsNeedingEnhancement, csvData, adversarialConfig, detectorManager) {
|
|
logSh(`🎯 Enhancement adversarial ${elementsNeedingEnhancement.length} éléments`, 'DEBUG');
|
|
|
|
const enhancementPrompt = `MISSION: Améliore UNIQUEMENT la précision technique de ces contenus.
|
|
|
|
CONTEXTE: ${csvData.mc0} - Secteur signalétique/impression
|
|
PERSONNALITÉ: ${csvData.personality?.nom} (${csvData.personality?.style})
|
|
|
|
CONTENUS À AMÉLIORER:
|
|
|
|
${elementsNeedingEnhancement.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
CONTENU: "${item.content}"
|
|
TERMES TECHNIQUES: ${item.technicalTerms.join(', ')}`).join('\n\n')}
|
|
|
|
CONSIGNES:
|
|
- GARDE même longueur, structure et ton ${csvData.personality?.style}
|
|
- Intègre naturellement les termes techniques listés
|
|
- NE CHANGE PAS le fond du message
|
|
- Vocabulaire expert mais accessible
|
|
- Termes secteur: dibond, aluminium, impression UV, fraisage, PMMA
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu avec amélioration technique
|
|
[2] Contenu avec amélioration technique
|
|
etc...`;
|
|
|
|
try {
|
|
// Générer prompt adversarial pour enhancement
|
|
const adversarialEnhancementPrompt = createAdversarialPrompt(enhancementPrompt, {
|
|
detectorTarget: adversarialConfig.detectorTarget,
|
|
intensity: adversarialConfig.intensity,
|
|
elementType: 'technical_enhancement',
|
|
personality: csvData.personality,
|
|
contextualMode: adversarialConfig.contextualMode,
|
|
csvData: csvData,
|
|
debugMode: false
|
|
});
|
|
|
|
const enhancedResponse = await callLLM('gpt4', adversarialEnhancementPrompt, {
|
|
temperature: 0.4,
|
|
maxTokens: 5000
|
|
}, csvData.personality);
|
|
|
|
return parseEnhancementResponse(enhancedResponse, elementsNeedingEnhancement);
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Enhancement éléments échoué: ${error.message}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Parser réponse analyse
|
|
*/
|
|
function parseAnalysisResponse(response, content, contentEntries) {
|
|
const results = [];
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\[\d+\]|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const index = parseInt(match[1]) - 1;
|
|
const termsText = match[2].trim();
|
|
parsedItems[index] = termsText;
|
|
}
|
|
|
|
contentEntries.forEach((tag, index) => {
|
|
const termsText = parsedItems[index] || 'AUCUN';
|
|
const hasTerms = !termsText.toUpperCase().includes('AUCUN');
|
|
|
|
const technicalTerms = hasTerms ?
|
|
termsText.split(',').map(t => t.trim()).filter(t => t.length > 0) :
|
|
[];
|
|
|
|
results.push({
|
|
tag,
|
|
content: content[tag],
|
|
technicalTerms,
|
|
needsEnhancement: hasTerms && technicalTerms.length > 0
|
|
});
|
|
|
|
logSh(`🔍 [${tag}]: ${hasTerms ? technicalTerms.join(', ') : 'aucun terme technique'}`, 'DEBUG');
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse enhancement
|
|
*/
|
|
function parseEnhancementResponse(response, elementsNeedingEnhancement) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < elementsNeedingEnhancement.length) {
|
|
let enhancedContent = match[2].trim();
|
|
const element = elementsNeedingEnhancement[index];
|
|
|
|
// Nettoyer le contenu généré
|
|
enhancedContent = cleanEnhancedContent(enhancedContent);
|
|
|
|
if (enhancedContent && enhancedContent.length > 10) {
|
|
results[element.tag] = enhancedContent;
|
|
logSh(`✅ Enhanced [${element.tag}]: "${enhancedContent.substring(0, 100)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content;
|
|
logSh(`⚠️ Fallback [${element.tag}]: contenu invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < elementsNeedingEnhancement.length) {
|
|
const element = elementsNeedingEnhancement[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Nettoyer contenu amélioré
|
|
*/
|
|
function cleanEnhancedContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?pour\s+/gi, '');
|
|
content = content.replace(/\*\*[^*]+\*\*/g, '');
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
module.exports = {
|
|
enhanceTechnicalTermsAdversarial, // ← MAIN ENTRY POINT ADVERSARIAL
|
|
analyzeTechnicalTermsAdversarial,
|
|
enhanceSelectedElementsAdversarial,
|
|
parseAnalysisResponse,
|
|
parseEnhancementResponse
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/AdversarialTransitionEnhance… │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ÉTAPE 3: ENHANCEMENT TRANSITIONS ADVERSARIAL
|
|
// Responsabilité: Améliorer la fluidité avec Gemini + anti-détection
|
|
// LLM: Gemini (température 0.6) + Prompts adversariaux
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { createAdversarialPrompt } = require('./AdversarialPromptEngine');
|
|
const { DetectorStrategyManager } = require('./DetectorStrategies');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - ENHANCEMENT TRANSITIONS ADVERSARIAL
|
|
* Input: { content: {}, csvData: {}, context: {}, adversarialConfig: {} }
|
|
* Output: { content: {}, stats: {}, debug: {} }
|
|
*/
|
|
async function enhanceTransitionsAdversarial(input) {
|
|
return await tracer.run('AdversarialTransitionEnhancement.enhanceTransitionsAdversarial()', async () => {
|
|
const { content, csvData, context = {}, adversarialConfig = {} } = input;
|
|
|
|
// Configuration adversariale par défaut
|
|
const config = {
|
|
detectorTarget: adversarialConfig.detectorTarget || 'general',
|
|
intensity: adversarialConfig.intensity || 1.0,
|
|
enableAdaptiveStrategy: adversarialConfig.enableAdaptiveStrategy || true,
|
|
contextualMode: adversarialConfig.contextualMode !== false,
|
|
...adversarialConfig
|
|
};
|
|
|
|
// Initialiser manager détecteur
|
|
const detectorManager = new DetectorStrategyManager(config.detectorTarget);
|
|
|
|
await tracer.annotate({
|
|
step: '3/4',
|
|
llmProvider: 'gemini',
|
|
elementsCount: Object.keys(content).length,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎯 ÉTAPE 3/4 ADVERSARIAL: Enhancement transitions (Gemini + ${config.detectorTarget})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments à analyser`, 'INFO');
|
|
|
|
try {
|
|
// 1. Analyser quels éléments ont besoin d'amélioration transitions
|
|
const elementsNeedingTransitions = analyzeTransitionNeeds(content);
|
|
|
|
logSh(` 📋 Analyse: ${elementsNeedingTransitions.length}/${Object.keys(content).length} éléments nécessitent fluidité`, 'INFO');
|
|
|
|
if (elementsNeedingTransitions.length === 0) {
|
|
logSh(`✅ ÉTAPE 3/4: Transitions déjà optimales`, 'INFO');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration: Date.now() - startTime },
|
|
debug: { llmProvider: 'gemini', step: 3, enhancementsApplied: [] }
|
|
};
|
|
}
|
|
|
|
// 2. Améliorer en chunks avec prompts adversariaux pour Gemini
|
|
const improvedResults = await improveTransitionsInChunksAdversarial(elementsNeedingTransitions, csvData, config, detectorManager);
|
|
|
|
// 3. Merger avec contenu original
|
|
const finalContent = { ...content };
|
|
let actuallyImproved = 0;
|
|
|
|
Object.keys(improvedResults).forEach(tag => {
|
|
if (improvedResults[tag] !== content[tag]) {
|
|
finalContent[tag] = improvedResults[tag];
|
|
actuallyImproved++;
|
|
}
|
|
});
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: Object.keys(content).length,
|
|
enhanced: actuallyImproved,
|
|
candidate: elementsNeedingTransitions.length,
|
|
duration
|
|
};
|
|
|
|
logSh(`✅ ÉTAPE 3/4 TERMINÉE: ${stats.enhanced} éléments fluidifiés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event(`Enhancement transitions terminé`, stats);
|
|
|
|
return {
|
|
content: finalContent,
|
|
stats,
|
|
debug: {
|
|
llmProvider: 'gemini',
|
|
step: 3,
|
|
enhancementsApplied: Object.keys(improvedResults),
|
|
transitionIssues: elementsNeedingTransitions.map(e => e.issues),
|
|
adversarialConfig: config,
|
|
detectorTarget: config.detectorTarget,
|
|
intensity: config.intensity
|
|
}
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ ÉTAPE 3/4 ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: retourner contenu original si Gemini indisponible
|
|
logSh(`🔄 Fallback: contenu original conservé`, 'WARNING');
|
|
return {
|
|
content,
|
|
stats: { processed: Object.keys(content).length, enhanced: 0, duration },
|
|
debug: { llmProvider: 'gemini', step: 3, error: error.message, fallback: true }
|
|
};
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* Analyser besoin d'amélioration transitions
|
|
*/
|
|
function analyzeTransitionNeeds(content) {
|
|
const elementsNeedingTransitions = [];
|
|
|
|
Object.keys(content).forEach(tag => {
|
|
const text = content[tag];
|
|
|
|
// Filtrer les éléments longs (>150 chars) qui peuvent bénéficier d'améliorations
|
|
if (text.length > 150) {
|
|
const needsTransitions = evaluateTransitionQuality(text);
|
|
|
|
if (needsTransitions.needsImprovement) {
|
|
elementsNeedingTransitions.push({
|
|
tag,
|
|
content: text,
|
|
issues: needsTransitions.issues,
|
|
score: needsTransitions.score
|
|
});
|
|
|
|
logSh(` 🔍 [${tag}]: Score=${needsTransitions.score.toFixed(2)}, Issues: ${needsTransitions.issues.join(', ')}`, 'DEBUG');
|
|
}
|
|
} else {
|
|
logSh(` ⏭️ [${tag}]: Trop court (${text.length}c), ignoré`, 'DEBUG');
|
|
}
|
|
});
|
|
|
|
// Trier par score (plus problématique en premier)
|
|
elementsNeedingTransitions.sort((a, b) => a.score - b.score);
|
|
|
|
return elementsNeedingTransitions;
|
|
}
|
|
|
|
/**
|
|
* Évaluer qualité transitions d'un texte
|
|
*/
|
|
function evaluateTransitionQuality(text) {
|
|
const sentences = text.split(/[.!?]+/).filter(s => s.trim().length > 10);
|
|
|
|
if (sentences.length < 2) {
|
|
return { needsImprovement: false, score: 1.0, issues: [] };
|
|
}
|
|
|
|
const issues = [];
|
|
let score = 1.0; // Score parfait = 1.0, problématique = 0.0
|
|
|
|
// Analyse 1: Connecteurs répétitifs
|
|
const repetitiveConnectors = analyzeRepetitiveConnectors(text);
|
|
if (repetitiveConnectors > 0.3) {
|
|
issues.push('connecteurs_répétitifs');
|
|
score -= 0.3;
|
|
}
|
|
|
|
// Analyse 2: Transitions abruptes
|
|
const abruptTransitions = analyzeAbruptTransitions(sentences);
|
|
if (abruptTransitions > 0.4) {
|
|
issues.push('transitions_abruptes');
|
|
score -= 0.4;
|
|
}
|
|
|
|
// Analyse 3: Manque de variété dans longueurs
|
|
const sentenceVariety = analyzeSentenceVariety(sentences);
|
|
if (sentenceVariety < 0.3) {
|
|
issues.push('phrases_uniformes');
|
|
score -= 0.2;
|
|
}
|
|
|
|
// Analyse 4: Trop formel ou trop familier
|
|
const formalityIssues = analyzeFormalityBalance(text);
|
|
if (formalityIssues > 0.5) {
|
|
issues.push('formalité_déséquilibrée');
|
|
score -= 0.1;
|
|
}
|
|
|
|
return {
|
|
needsImprovement: score < 0.6,
|
|
score: Math.max(0, score),
|
|
issues
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Améliorer transitions en chunks avec prompts adversariaux
|
|
*/
|
|
async function improveTransitionsInChunksAdversarial(elementsNeedingTransitions, csvData, adversarialConfig, detectorManager) {
|
|
logSh(`🎯 Amélioration transitions adversarial: ${elementsNeedingTransitions.length} éléments`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(elementsNeedingTransitions, 6); // Chunks plus petits pour Gemini
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(` 📦 Chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
const basePrompt = createTransitionImprovementPrompt(chunk, csvData);
|
|
|
|
// Générer prompt adversarial pour amélioration transitions
|
|
const adversarialPrompt = createAdversarialPrompt(basePrompt, {
|
|
detectorTarget: adversarialConfig.detectorTarget,
|
|
intensity: adversarialConfig.intensity * 0.9, // Intensité légèrement réduite pour transitions
|
|
elementType: 'transition_enhancement',
|
|
personality: csvData.personality,
|
|
contextualMode: adversarialConfig.contextualMode,
|
|
csvData: csvData,
|
|
debugMode: false
|
|
});
|
|
|
|
const improvedResponse = await callLLM('gemini', adversarialPrompt, {
|
|
temperature: 0.6,
|
|
maxTokens: 2500
|
|
}, csvData.personality);
|
|
|
|
const chunkResults = parseTransitionResponse(improvedResponse, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk ${chunkIndex + 1}: ${Object.keys(chunkResults).length} améliorés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: garder contenu original pour ce chunk
|
|
chunk.forEach(element => {
|
|
results[element.tag] = element.content;
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt amélioration transitions
|
|
*/
|
|
function createTransitionImprovementPrompt(chunk, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `MISSION: Améliore UNIQUEMENT les transitions et fluidité de ces contenus.
|
|
|
|
CONTEXTE: Article SEO ${csvData.mc0}
|
|
PERSONNALITÉ: ${personality?.nom} (${personality?.style} web professionnel)
|
|
CONNECTEURS PRÉFÉRÉS: ${personality?.connecteursPref}
|
|
|
|
CONTENUS À FLUIDIFIER:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
PROBLÈMES: ${item.issues.join(', ')}
|
|
CONTENU: "${item.content}"`).join('\n\n')}
|
|
|
|
OBJECTIFS:
|
|
- Connecteurs plus naturels et variés: ${personality?.connecteursPref}
|
|
- Transitions fluides entre idées
|
|
- ÉVITE répétitions excessives ("du coup", "franchement", "par ailleurs")
|
|
- Style ${personality?.style} mais professionnel web
|
|
|
|
CONSIGNES STRICTES:
|
|
- NE CHANGE PAS le fond du message
|
|
- GARDE même structure et longueur
|
|
- Améliore SEULEMENT la fluidité
|
|
- RESPECTE le style ${personality?.nom}
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu avec transitions améliorées
|
|
[2] Contenu avec transitions améliorées
|
|
etc...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse amélioration transitions
|
|
*/
|
|
function parseTransitionResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let improvedContent = match[2].trim();
|
|
const element = chunk[index];
|
|
|
|
// Nettoyer le contenu amélioré
|
|
improvedContent = cleanImprovedContent(improvedContent);
|
|
|
|
if (improvedContent && improvedContent.length > 10) {
|
|
results[element.tag] = improvedContent;
|
|
logSh(`✅ Improved [${element.tag}]: "${improvedContent.substring(0, 100)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content;
|
|
logSh(`⚠️ Fallback [${element.tag}]: amélioration invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < chunk.length) {
|
|
const element = chunk[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
function analyzeRepetitiveConnectors(content) {
|
|
const connectors = ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc'];
|
|
let totalConnectors = 0;
|
|
let repetitions = 0;
|
|
|
|
connectors.forEach(connector => {
|
|
const matches = (content.match(new RegExp(`\\b${connector}\\b`, 'gi')) || []);
|
|
totalConnectors += matches.length;
|
|
if (matches.length > 1) repetitions += matches.length - 1;
|
|
});
|
|
|
|
return totalConnectors > 0 ? repetitions / totalConnectors : 0;
|
|
}
|
|
|
|
function analyzeAbruptTransitions(sentences) {
|
|
if (sentences.length < 2) return 0;
|
|
|
|
let abruptCount = 0;
|
|
|
|
for (let i = 1; i < sentences.length; i++) {
|
|
const current = sentences[i].trim();
|
|
const hasConnector = hasTransitionWord(current);
|
|
|
|
if (!hasConnector && current.length > 30) {
|
|
abruptCount++;
|
|
}
|
|
}
|
|
|
|
return abruptCount / (sentences.length - 1);
|
|
}
|
|
|
|
function analyzeSentenceVariety(sentences) {
|
|
if (sentences.length < 2) return 1;
|
|
|
|
const lengths = sentences.map(s => s.trim().length);
|
|
const avgLength = lengths.reduce((a, b) => a + b, 0) / lengths.length;
|
|
const variance = lengths.reduce((acc, len) => acc + Math.pow(len - avgLength, 2), 0) / lengths.length;
|
|
const stdDev = Math.sqrt(variance);
|
|
|
|
return Math.min(1, stdDev / avgLength);
|
|
}
|
|
|
|
function analyzeFormalityBalance(content) {
|
|
const formalIndicators = ['il convient de', 'par conséquent', 'néanmoins', 'toutefois'];
|
|
const casualIndicators = ['du coup', 'bon', 'franchement', 'nickel'];
|
|
|
|
let formalCount = 0;
|
|
let casualCount = 0;
|
|
|
|
formalIndicators.forEach(indicator => {
|
|
if (content.toLowerCase().includes(indicator)) formalCount++;
|
|
});
|
|
|
|
casualIndicators.forEach(indicator => {
|
|
if (content.toLowerCase().includes(indicator)) casualCount++;
|
|
});
|
|
|
|
const total = formalCount + casualCount;
|
|
if (total === 0) return 0;
|
|
|
|
// Déséquilibre si trop d'un côté
|
|
const balance = Math.abs(formalCount - casualCount) / total;
|
|
return balance;
|
|
}
|
|
|
|
function hasTransitionWord(sentence) {
|
|
const connectors = ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc', 'ensuite', 'puis', 'également', 'aussi'];
|
|
return connectors.some(connector => sentence.toLowerCase().includes(connector));
|
|
}
|
|
|
|
function cleanImprovedContent(content) {
|
|
if (!content) return content;
|
|
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?/, '');
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
module.exports = {
|
|
enhanceTransitionsAdversarial, // ← MAIN ENTRY POINT ADVERSARIAL
|
|
analyzeTransitionNeeds,
|
|
evaluateTransitionQuality,
|
|
improveTransitionsInChunksAdversarial,
|
|
createTransitionImprovementPrompt,
|
|
parseTransitionResponse
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/AdversarialUtils.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ADVERSARIAL UTILS - UTILITAIRES MODULAIRES
|
|
// Responsabilité: Fonctions utilitaires partagées par tous les modules adversariaux
|
|
// Architecture: Helper functions réutilisables et composables
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
|
|
/**
|
|
* ANALYSEURS DE CONTENU
|
|
*/
|
|
|
|
/**
|
|
* Analyser score de diversité lexicale
|
|
*/
|
|
function analyzeLexicalDiversity(content) {
|
|
if (!content || typeof content !== 'string') return 0;
|
|
|
|
const words = content.toLowerCase()
|
|
.split(/\s+/)
|
|
.filter(word => word.length > 2)
|
|
.map(word => word.replace(/[^\w]/g, ''));
|
|
|
|
if (words.length === 0) return 0;
|
|
|
|
const uniqueWords = [...new Set(words)];
|
|
return (uniqueWords.length / words.length) * 100;
|
|
}
|
|
|
|
/**
|
|
* Analyser variation des longueurs de phrases
|
|
*/
|
|
function analyzeSentenceVariation(content) {
|
|
if (!content || typeof content !== 'string') return 0;
|
|
|
|
const sentences = content.split(/[.!?]+/)
|
|
.map(s => s.trim())
|
|
.filter(s => s.length > 5);
|
|
|
|
if (sentences.length < 2) return 0;
|
|
|
|
const lengths = sentences.map(s => s.split(/\s+/).length);
|
|
const avgLength = lengths.reduce((a, b) => a + b, 0) / lengths.length;
|
|
const variance = lengths.reduce((acc, len) => acc + Math.pow(len - avgLength, 2), 0) / lengths.length;
|
|
const stdDev = Math.sqrt(variance);
|
|
|
|
return Math.min(100, (stdDev / avgLength) * 100);
|
|
}
|
|
|
|
/**
|
|
* Détecter mots typiques IA
|
|
*/
|
|
function detectAIFingerprints(content) {
|
|
const aiFingerprints = {
|
|
words: ['optimal', 'comprehensive', 'seamless', 'robust', 'leverage', 'cutting-edge', 'state-of-the-art', 'furthermore', 'moreover'],
|
|
phrases: ['it is important to note', 'it should be noted', 'it is worth mentioning', 'in conclusion', 'to summarize'],
|
|
connectors: ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc']
|
|
};
|
|
|
|
const results = {
|
|
words: 0,
|
|
phrases: 0,
|
|
connectors: 0,
|
|
totalScore: 0
|
|
};
|
|
|
|
const lowerContent = content.toLowerCase();
|
|
|
|
// Compter mots IA
|
|
aiFingerprints.words.forEach(word => {
|
|
const matches = (lowerContent.match(new RegExp(`\\b${word}\\b`, 'g')) || []);
|
|
results.words += matches.length;
|
|
});
|
|
|
|
// Compter phrases typiques
|
|
aiFingerprints.phrases.forEach(phrase => {
|
|
if (lowerContent.includes(phrase)) {
|
|
results.phrases += 1;
|
|
}
|
|
});
|
|
|
|
// Compter connecteurs répétitifs
|
|
aiFingerprints.connectors.forEach(connector => {
|
|
const matches = (lowerContent.match(new RegExp(`\\b${connector}\\b`, 'g')) || []);
|
|
if (matches.length > 1) {
|
|
results.connectors += matches.length - 1; // Pénalité répétition
|
|
}
|
|
});
|
|
|
|
// Score total (sur 100)
|
|
const wordCount = content.split(/\s+/).length;
|
|
results.totalScore = Math.min(100,
|
|
(results.words * 5 + results.phrases * 10 + results.connectors * 3) / Math.max(wordCount, 1) * 100
|
|
);
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Analyser uniformité structurelle
|
|
*/
|
|
function analyzeStructuralUniformity(content) {
|
|
const sentences = content.split(/[.!?]+/)
|
|
.map(s => s.trim())
|
|
.filter(s => s.length > 5);
|
|
|
|
if (sentences.length < 3) return 0;
|
|
|
|
const structures = sentences.map(sentence => {
|
|
const words = sentence.split(/\s+/);
|
|
return {
|
|
length: words.length,
|
|
startsWithConnector: /^(par ailleurs|en effet|de plus|cependant|ainsi|donc|ensuite|puis)/i.test(sentence),
|
|
hasComma: sentence.includes(','),
|
|
hasSubordinate: /qui|que|dont|où|quand|comme|parce que|puisque|bien que/i.test(sentence)
|
|
};
|
|
});
|
|
|
|
// Calculer uniformité
|
|
const avgLength = structures.reduce((sum, s) => sum + s.length, 0) / structures.length;
|
|
const lengthVariance = structures.reduce((sum, s) => sum + Math.pow(s.length - avgLength, 2), 0) / structures.length;
|
|
|
|
const connectorRatio = structures.filter(s => s.startsWithConnector).length / structures.length;
|
|
const commaRatio = structures.filter(s => s.hasComma).length / structures.length;
|
|
|
|
// Plus c'est uniforme, plus le score est élevé (mauvais pour anti-détection)
|
|
const uniformityScore = 100 - (Math.sqrt(lengthVariance) / avgLength * 100) -
|
|
(Math.abs(0.3 - connectorRatio) * 50) - (Math.abs(0.5 - commaRatio) * 30);
|
|
|
|
return Math.max(0, Math.min(100, uniformityScore));
|
|
}
|
|
|
|
/**
|
|
* COMPARATEURS DE CONTENU
|
|
*/
|
|
|
|
/**
|
|
* Comparer deux contenus et calculer taux de modification
|
|
*/
|
|
function compareContentModification(original, modified) {
|
|
if (!original || !modified) return 0;
|
|
|
|
const originalWords = original.toLowerCase().split(/\s+/).filter(w => w.length > 2);
|
|
const modifiedWords = modified.toLowerCase().split(/\s+/).filter(w => w.length > 2);
|
|
|
|
// Calcul de distance Levenshtein approximative (par mots)
|
|
let changes = 0;
|
|
const maxLength = Math.max(originalWords.length, modifiedWords.length);
|
|
|
|
for (let i = 0; i < maxLength; i++) {
|
|
if (originalWords[i] !== modifiedWords[i]) {
|
|
changes++;
|
|
}
|
|
}
|
|
|
|
return (changes / maxLength) * 100;
|
|
}
|
|
|
|
/**
|
|
* Évaluer amélioration adversariale
|
|
*/
|
|
function evaluateAdversarialImprovement(original, modified, detectorTarget = 'general') {
|
|
const originalFingerprints = detectAIFingerprints(original);
|
|
const modifiedFingerprints = detectAIFingerprints(modified);
|
|
|
|
const originalDiversity = analyzeLexicalDiversity(original);
|
|
const modifiedDiversity = analyzeLexicalDiversity(modified);
|
|
|
|
const originalVariation = analyzeSentenceVariation(original);
|
|
const modifiedVariation = analyzeSentenceVariation(modified);
|
|
|
|
const fingerprintReduction = originalFingerprints.totalScore - modifiedFingerprints.totalScore;
|
|
const diversityIncrease = modifiedDiversity - originalDiversity;
|
|
const variationIncrease = modifiedVariation - originalVariation;
|
|
|
|
const improvementScore = (
|
|
fingerprintReduction * 0.4 +
|
|
diversityIncrease * 0.3 +
|
|
variationIncrease * 0.3
|
|
);
|
|
|
|
return {
|
|
fingerprintReduction,
|
|
diversityIncrease,
|
|
variationIncrease,
|
|
improvementScore: Math.round(improvementScore * 100) / 100,
|
|
modificationRate: compareContentModification(original, modified),
|
|
recommendation: getImprovementRecommendation(improvementScore, detectorTarget)
|
|
};
|
|
}
|
|
|
|
/**
|
|
* UTILITAIRES DE CONTENU
|
|
*/
|
|
|
|
/**
|
|
* Nettoyer contenu adversarial généré
|
|
*/
|
|
function cleanAdversarialContent(content) {
|
|
if (!content || typeof content !== 'string') return content;
|
|
|
|
let cleaned = content;
|
|
|
|
// Supprimer préfixes de génération
|
|
cleaned = cleaned.replace(/^(voici\s+)?le\s+contenu\s+(réécrit|amélioré|modifié)[:\s]*/gi, '');
|
|
cleaned = cleaned.replace(/^(bon,?\s*)?(alors,?\s*)?(pour\s+)?(ce\s+contenu[,\s]*)?/gi, '');
|
|
|
|
// Nettoyer formatage
|
|
cleaned = cleaned.replace(/\*\*[^*]+\*\*/g, ''); // Gras markdown
|
|
cleaned = cleaned.replace(/\s{2,}/g, ' '); // Espaces multiples
|
|
cleaned = cleaned.replace(/([.!?])\s*([.!?])/g, '$1 '); // Double ponctuation
|
|
|
|
// Nettoyer début/fin
|
|
cleaned = cleaned.trim();
|
|
cleaned = cleaned.replace(/^[,.\s]+/, '');
|
|
cleaned = cleaned.replace(/[,\s]+$/, '');
|
|
|
|
return cleaned;
|
|
}
|
|
|
|
/**
|
|
* Valider qualité du contenu adversarial
|
|
*/
|
|
function validateAdversarialContent(content, originalContent, minLength = 10, maxModificationRate = 90) {
|
|
const validation = {
|
|
isValid: true,
|
|
issues: [],
|
|
suggestions: []
|
|
};
|
|
|
|
// Vérifier longueur minimale
|
|
if (!content || content.length < minLength) {
|
|
validation.isValid = false;
|
|
validation.issues.push('Contenu trop court');
|
|
validation.suggestions.push('Augmenter la longueur du contenu généré');
|
|
}
|
|
|
|
// Vérifier cohérence
|
|
if (originalContent) {
|
|
const modificationRate = compareContentModification(originalContent, content);
|
|
|
|
if (modificationRate > maxModificationRate) {
|
|
validation.issues.push('Modification trop importante');
|
|
validation.suggestions.push('Réduire l\'intensité adversariale pour préserver le sens');
|
|
}
|
|
|
|
if (modificationRate < 5) {
|
|
validation.issues.push('Modification insuffisante');
|
|
validation.suggestions.push('Augmenter l\'intensité adversariale');
|
|
}
|
|
}
|
|
|
|
// Vérifier empreintes IA résiduelles
|
|
const fingerprints = detectAIFingerprints(content);
|
|
if (fingerprints.totalScore > 15) {
|
|
validation.issues.push('Empreintes IA encore présentes');
|
|
validation.suggestions.push('Appliquer post-processing anti-fingerprints');
|
|
}
|
|
|
|
return validation;
|
|
}
|
|
|
|
/**
|
|
* UTILITAIRES TECHNIQUES
|
|
*/
|
|
|
|
/**
|
|
* Chunk array avec préservation des paires
|
|
*/
|
|
function chunkArraySmart(array, size, preservePairs = false) {
|
|
if (!preservePairs) {
|
|
return chunkArray(array, size);
|
|
}
|
|
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
let chunk = array.slice(i, i + size);
|
|
|
|
// Si on coupe au milieu d'une paire (nombre impair), ajuster
|
|
if (chunk.length % 2 !== 0 && i + size < array.length) {
|
|
chunk = array.slice(i, i + size - 1);
|
|
}
|
|
|
|
chunks.push(chunk);
|
|
}
|
|
|
|
return chunks;
|
|
}
|
|
|
|
/**
|
|
* Chunk array standard
|
|
*/
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
/**
|
|
* Sleep avec variation
|
|
*/
|
|
function sleep(ms, variation = 0.2) {
|
|
const actualMs = ms + (Math.random() - 0.5) * ms * variation;
|
|
return new Promise(resolve => setTimeout(resolve, Math.max(100, actualMs)));
|
|
}
|
|
|
|
/**
|
|
* RECOMMANDATIONS
|
|
*/
|
|
|
|
/**
|
|
* Obtenir recommandation d'amélioration
|
|
*/
|
|
function getImprovementRecommendation(score, detectorTarget) {
|
|
const recommendations = {
|
|
general: {
|
|
good: "Bon niveau d'amélioration générale",
|
|
medium: "Appliquer techniques de variation syntaxique",
|
|
poor: "Nécessite post-processing intensif"
|
|
},
|
|
gptZero: {
|
|
good: "Imprévisibilité suffisante contre GPTZero",
|
|
medium: "Ajouter plus de ruptures narratives",
|
|
poor: "Intensifier variation syntaxique et lexicale"
|
|
},
|
|
originality: {
|
|
good: "Créativité suffisante contre Originality",
|
|
medium: "Enrichir diversité sémantique",
|
|
poor: "Réinventer présentation des informations"
|
|
}
|
|
};
|
|
|
|
const category = score > 10 ? 'good' : score > 5 ? 'medium' : 'poor';
|
|
return recommendations[detectorTarget]?.[category] || recommendations.general[category];
|
|
}
|
|
|
|
/**
|
|
* MÉTRIQUES ET STATS
|
|
*/
|
|
|
|
/**
|
|
* Calculer score composite anti-détection
|
|
*/
|
|
function calculateAntiDetectionScore(content, detectorTarget = 'general') {
|
|
const diversity = analyzeLexicalDiversity(content);
|
|
const variation = analyzeSentenceVariation(content);
|
|
const fingerprints = detectAIFingerprints(content);
|
|
const uniformity = analyzeStructuralUniformity(content);
|
|
|
|
const baseScore = (diversity * 0.3 + variation * 0.3 + (100 - fingerprints.totalScore) * 0.2 + (100 - uniformity) * 0.2);
|
|
|
|
// Ajustements selon détecteur
|
|
let adjustedScore = baseScore;
|
|
switch (detectorTarget) {
|
|
case 'gptZero':
|
|
adjustedScore = baseScore * (variation / 100) * 1.2; // Favorise variation
|
|
break;
|
|
case 'originality':
|
|
adjustedScore = baseScore * (diversity / 100) * 1.2; // Favorise diversité
|
|
break;
|
|
}
|
|
|
|
return Math.min(100, Math.max(0, Math.round(adjustedScore)));
|
|
}
|
|
|
|
module.exports = {
|
|
// Analyseurs
|
|
analyzeLexicalDiversity,
|
|
analyzeSentenceVariation,
|
|
detectAIFingerprints,
|
|
analyzeStructuralUniformity,
|
|
|
|
// Comparateurs
|
|
compareContentModification,
|
|
evaluateAdversarialImprovement,
|
|
|
|
// Utilitaires contenu
|
|
cleanAdversarialContent,
|
|
validateAdversarialContent,
|
|
|
|
// Utilitaires techniques
|
|
chunkArray,
|
|
chunkArraySmart,
|
|
sleep,
|
|
|
|
// Métriques
|
|
calculateAntiDetectionScore,
|
|
getImprovementRecommendation
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/ContentGenerationAdversarial.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// ORCHESTRATEUR CONTENU ADVERSARIAL - NIVEAU 3
|
|
// Responsabilité: Pipeline complet de génération anti-détection
|
|
// Architecture: 4 étapes adversariales séparées et modulaires
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
// Importation des 4 étapes adversariales
|
|
const { generateInitialContentAdversarial } = require('./AdversarialInitialGeneration');
|
|
const { enhanceTechnicalTermsAdversarial } = require('./AdversarialTechnicalEnhancement');
|
|
const { enhanceTransitionsAdversarial } = require('./AdversarialTransitionEnhancement');
|
|
const { applyPersonalityStyleAdversarial } = require('./AdversarialStyleEnhancement');
|
|
|
|
// Importation du moteur adversarial
|
|
const { createAdversarialPrompt, getSupportedDetectors, analyzePromptEffectiveness } = require('./AdversarialPromptEngine');
|
|
const { DetectorStrategyManager } = require('./DetectorStrategies');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - PIPELINE ADVERSARIAL COMPLET
|
|
* Input: { hierarchy, csvData, adversarialConfig, context }
|
|
* Output: { content, stats, debug, adversarialMetrics }
|
|
*/
|
|
async function generateWithAdversarialContext(input) {
|
|
return await tracer.run('ContentGenerationAdversarial.generateWithAdversarialContext()', async () => {
|
|
const { hierarchy, csvData, adversarialConfig = {}, context = {} } = input;
|
|
|
|
// Configuration adversariale par défaut
|
|
const config = {
|
|
detectorTarget: adversarialConfig.detectorTarget || 'general',
|
|
intensity: adversarialConfig.intensity || 1.0,
|
|
enableAdaptiveStrategy: adversarialConfig.enableAdaptiveStrategy !== false,
|
|
contextualMode: adversarialConfig.contextualMode !== false,
|
|
enableAllSteps: adversarialConfig.enableAllSteps !== false,
|
|
// Configuration par étape
|
|
steps: {
|
|
initial: adversarialConfig.steps?.initial !== false,
|
|
technical: adversarialConfig.steps?.technical !== false,
|
|
transitions: adversarialConfig.steps?.transitions !== false,
|
|
style: adversarialConfig.steps?.style !== false
|
|
},
|
|
...adversarialConfig
|
|
};
|
|
|
|
await tracer.annotate({
|
|
adversarialPipeline: true,
|
|
detectorTarget: config.detectorTarget,
|
|
intensity: config.intensity,
|
|
enabledSteps: Object.keys(config.steps).filter(k => config.steps[k]),
|
|
elementsCount: Object.keys(hierarchy).length,
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎯 PIPELINE ADVERSARIAL NIVEAU 3: Anti-détection ${config.detectorTarget}`, 'INFO');
|
|
logSh(` 🎚️ Intensité: ${config.intensity.toFixed(2)} | Étapes: ${Object.keys(config.steps).filter(k => config.steps[k]).join(', ')}`, 'INFO');
|
|
|
|
// Initialiser manager détecteur global
|
|
const detectorManager = new DetectorStrategyManager(config.detectorTarget);
|
|
|
|
try {
|
|
let currentContent = {};
|
|
let pipelineStats = {
|
|
steps: {},
|
|
totalDuration: 0,
|
|
elementsProcessed: 0,
|
|
adversarialMetrics: {
|
|
promptsGenerated: 0,
|
|
detectorTarget: config.detectorTarget,
|
|
averageIntensity: config.intensity,
|
|
effectivenessScore: 0
|
|
}
|
|
};
|
|
|
|
// ========================================
|
|
// ÉTAPE 1: GÉNÉRATION INITIALE ADVERSARIALE
|
|
// ========================================
|
|
if (config.steps.initial) {
|
|
logSh(`🎯 ÉTAPE 1/4: Génération initiale adversariale`, 'INFO');
|
|
|
|
const step1Result = await generateInitialContentAdversarial({
|
|
hierarchy,
|
|
csvData,
|
|
context,
|
|
adversarialConfig: config
|
|
});
|
|
|
|
currentContent = step1Result.content;
|
|
pipelineStats.steps.initial = step1Result.stats;
|
|
pipelineStats.adversarialMetrics.promptsGenerated += Object.keys(currentContent).length;
|
|
|
|
logSh(`✅ ÉTAPE 1/4: ${step1Result.stats.generated} éléments générés (${step1Result.stats.duration}ms)`, 'INFO');
|
|
} else {
|
|
logSh(`⏭️ ÉTAPE 1/4: Ignorée (configuration)`, 'INFO');
|
|
}
|
|
|
|
// ========================================
|
|
// ÉTAPE 2: ENHANCEMENT TECHNIQUE ADVERSARIAL
|
|
// ========================================
|
|
if (config.steps.technical && Object.keys(currentContent).length > 0) {
|
|
logSh(`🎯 ÉTAPE 2/4: Enhancement technique adversarial`, 'INFO');
|
|
|
|
const step2Result = await enhanceTechnicalTermsAdversarial({
|
|
content: currentContent,
|
|
csvData,
|
|
context,
|
|
adversarialConfig: config
|
|
});
|
|
|
|
currentContent = step2Result.content;
|
|
pipelineStats.steps.technical = step2Result.stats;
|
|
pipelineStats.adversarialMetrics.promptsGenerated += step2Result.stats.enhanced;
|
|
|
|
logSh(`✅ ÉTAPE 2/4: ${step2Result.stats.enhanced} éléments améliorés (${step2Result.stats.duration}ms)`, 'INFO');
|
|
} else {
|
|
logSh(`⏭️ ÉTAPE 2/4: Ignorée (configuration ou pas de contenu)`, 'INFO');
|
|
}
|
|
|
|
// ========================================
|
|
// ÉTAPE 3: ENHANCEMENT TRANSITIONS ADVERSARIAL
|
|
// ========================================
|
|
if (config.steps.transitions && Object.keys(currentContent).length > 0) {
|
|
logSh(`🎯 ÉTAPE 3/4: Enhancement transitions adversarial`, 'INFO');
|
|
|
|
const step3Result = await enhanceTransitionsAdversarial({
|
|
content: currentContent,
|
|
csvData,
|
|
context,
|
|
adversarialConfig: config
|
|
});
|
|
|
|
currentContent = step3Result.content;
|
|
pipelineStats.steps.transitions = step3Result.stats;
|
|
pipelineStats.adversarialMetrics.promptsGenerated += step3Result.stats.enhanced;
|
|
|
|
logSh(`✅ ÉTAPE 3/4: ${step3Result.stats.enhanced} éléments fluidifiés (${step3Result.stats.duration}ms)`, 'INFO');
|
|
} else {
|
|
logSh(`⏭️ ÉTAPE 3/4: Ignorée (configuration ou pas de contenu)`, 'INFO');
|
|
}
|
|
|
|
// ========================================
|
|
// ÉTAPE 4: ENHANCEMENT STYLE ADVERSARIAL
|
|
// ========================================
|
|
if (config.steps.style && Object.keys(currentContent).length > 0 && csvData.personality) {
|
|
logSh(`🎯 ÉTAPE 4/4: Enhancement style adversarial`, 'INFO');
|
|
|
|
const step4Result = await applyPersonalityStyleAdversarial({
|
|
content: currentContent,
|
|
csvData,
|
|
context,
|
|
adversarialConfig: config
|
|
});
|
|
|
|
currentContent = step4Result.content;
|
|
pipelineStats.steps.style = step4Result.stats;
|
|
pipelineStats.adversarialMetrics.promptsGenerated += step4Result.stats.enhanced;
|
|
|
|
logSh(`✅ ÉTAPE 4/4: ${step4Result.stats.enhanced} éléments stylisés (${step4Result.stats.duration}ms)`, 'INFO');
|
|
} else {
|
|
logSh(`⏭️ ÉTAPE 4/4: Ignorée (configuration, pas de contenu ou pas de personnalité)`, 'INFO');
|
|
}
|
|
|
|
// ========================================
|
|
// FINALISATION PIPELINE
|
|
// ========================================
|
|
const totalDuration = Date.now() - startTime;
|
|
pipelineStats.totalDuration = totalDuration;
|
|
pipelineStats.elementsProcessed = Object.keys(currentContent).length;
|
|
|
|
// Calculer score d'efficacité adversarial
|
|
pipelineStats.adversarialMetrics.effectivenessScore = calculateAdversarialEffectiveness(
|
|
pipelineStats,
|
|
config,
|
|
currentContent
|
|
);
|
|
|
|
logSh(`🎯 PIPELINE ADVERSARIAL TERMINÉ: ${pipelineStats.elementsProcessed} éléments (${totalDuration}ms)`, 'INFO');
|
|
logSh(` 📊 Score efficacité: ${pipelineStats.adversarialMetrics.effectivenessScore.toFixed(2)}%`, 'INFO');
|
|
|
|
await tracer.event(`Pipeline adversarial terminé`, {
|
|
...pipelineStats,
|
|
detectorTarget: config.detectorTarget,
|
|
intensity: config.intensity
|
|
});
|
|
|
|
return {
|
|
content: currentContent,
|
|
stats: pipelineStats,
|
|
debug: {
|
|
adversarialPipeline: true,
|
|
detectorTarget: config.detectorTarget,
|
|
intensity: config.intensity,
|
|
stepsExecuted: Object.keys(config.steps).filter(k => config.steps[k]),
|
|
detectorManager: detectorManager.getStrategyInfo()
|
|
},
|
|
adversarialMetrics: pipelineStats.adversarialMetrics
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ PIPELINE ADVERSARIAL ÉCHOUÉ après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`AdversarialContentGeneration failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* MODE SIMPLE ADVERSARIAL (équivalent à generateSimple mais adversarial)
|
|
*/
|
|
async function generateSimpleAdversarial(hierarchy, csvData, adversarialConfig = {}) {
|
|
return await generateWithAdversarialContext({
|
|
hierarchy,
|
|
csvData,
|
|
adversarialConfig: {
|
|
detectorTarget: 'general',
|
|
intensity: 0.8,
|
|
enableAllSteps: false,
|
|
steps: {
|
|
initial: true,
|
|
technical: false,
|
|
transitions: false,
|
|
style: true
|
|
},
|
|
...adversarialConfig
|
|
}
|
|
});
|
|
}
|
|
|
|
/**
|
|
* MODE AVANCÉ ADVERSARIAL (configuration personnalisée)
|
|
*/
|
|
async function generateAdvancedAdversarial(hierarchy, csvData, options = {}) {
|
|
const {
|
|
detectorTarget = 'general',
|
|
intensity = 1.0,
|
|
technical = true,
|
|
transitions = true,
|
|
style = true,
|
|
...otherConfig
|
|
} = options;
|
|
|
|
return await generateWithAdversarialContext({
|
|
hierarchy,
|
|
csvData,
|
|
adversarialConfig: {
|
|
detectorTarget,
|
|
intensity,
|
|
enableAdaptiveStrategy: true,
|
|
contextualMode: true,
|
|
steps: {
|
|
initial: true,
|
|
technical,
|
|
transitions,
|
|
style
|
|
},
|
|
...otherConfig
|
|
}
|
|
});
|
|
}
|
|
|
|
/**
|
|
* DIAGNOSTIC PIPELINE ADVERSARIAL
|
|
*/
|
|
async function diagnosticAdversarialPipeline(hierarchy, csvData, detectorTargets = ['general', 'gptZero', 'originality']) {
|
|
logSh(`🔬 DIAGNOSTIC ADVERSARIAL: Testing ${detectorTargets.length} détecteurs`, 'INFO');
|
|
|
|
const results = {};
|
|
|
|
for (const target of detectorTargets) {
|
|
try {
|
|
logSh(` 🎯 Test détecteur: ${target}`, 'DEBUG');
|
|
|
|
const result = await generateWithAdversarialContext({
|
|
hierarchy,
|
|
csvData,
|
|
adversarialConfig: {
|
|
detectorTarget: target,
|
|
intensity: 1.0,
|
|
enableAllSteps: true
|
|
}
|
|
});
|
|
|
|
results[target] = {
|
|
success: true,
|
|
content: result.content,
|
|
stats: result.stats,
|
|
effectivenessScore: result.adversarialMetrics.effectivenessScore
|
|
};
|
|
|
|
logSh(` ✅ ${target}: Score ${result.adversarialMetrics.effectivenessScore.toFixed(2)}%`, 'DEBUG');
|
|
|
|
} catch (error) {
|
|
results[target] = {
|
|
success: false,
|
|
error: error.message,
|
|
effectivenessScore: 0
|
|
};
|
|
|
|
logSh(` ❌ ${target}: Échec - ${error.message}`, 'ERROR');
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
/**
|
|
* Calculer efficacité adversariale
|
|
*/
|
|
function calculateAdversarialEffectiveness(pipelineStats, config, content) {
|
|
let effectiveness = 0;
|
|
|
|
// Base score selon intensité
|
|
effectiveness += config.intensity * 30;
|
|
|
|
// Bonus selon nombre d'étapes
|
|
const stepsExecuted = Object.keys(config.steps).filter(k => config.steps[k]).length;
|
|
effectiveness += stepsExecuted * 10;
|
|
|
|
// Bonus selon prompts adversariaux générés
|
|
const promptRatio = pipelineStats.adversarialMetrics.promptsGenerated / Math.max(1, pipelineStats.elementsProcessed);
|
|
effectiveness += promptRatio * 20;
|
|
|
|
// Analyse contenu si disponible
|
|
if (Object.keys(content).length > 0) {
|
|
const contentSample = Object.values(content).join(' ').substring(0, 1000);
|
|
const diversityScore = analyzeDiversityScore(contentSample);
|
|
effectiveness += diversityScore * 0.3;
|
|
}
|
|
|
|
return Math.min(100, Math.max(0, effectiveness));
|
|
}
|
|
|
|
/**
|
|
* Analyser score de diversité
|
|
*/
|
|
function analyzeDiversityScore(content) {
|
|
if (!content || typeof content !== 'string') return 0;
|
|
|
|
const words = content.split(/\s+/).filter(w => w.length > 2);
|
|
if (words.length === 0) return 0;
|
|
|
|
const uniqueWords = [...new Set(words.map(w => w.toLowerCase()))];
|
|
const diversityRatio = uniqueWords.length / words.length;
|
|
|
|
return diversityRatio * 100;
|
|
}
|
|
|
|
/**
|
|
* Obtenir informations détecteurs supportés
|
|
*/
|
|
function getAdversarialDetectorInfo() {
|
|
return getSupportedDetectors();
|
|
}
|
|
|
|
/**
|
|
* Comparer efficacité de différents détecteurs
|
|
*/
|
|
async function compareAdversarialStrategies(hierarchy, csvData, detectorTargets = ['general', 'gptZero', 'originality', 'winston']) {
|
|
const results = await diagnosticAdversarialPipeline(hierarchy, csvData, detectorTargets);
|
|
|
|
const comparison = {
|
|
bestStrategy: null,
|
|
bestScore: 0,
|
|
strategies: [],
|
|
averageScore: 0
|
|
};
|
|
|
|
let totalScore = 0;
|
|
let successCount = 0;
|
|
|
|
detectorTargets.forEach(target => {
|
|
const result = results[target];
|
|
if (result.success) {
|
|
const strategyInfo = {
|
|
detector: target,
|
|
effectivenessScore: result.effectivenessScore,
|
|
duration: result.stats.totalDuration,
|
|
elementsProcessed: result.stats.elementsProcessed
|
|
};
|
|
|
|
comparison.strategies.push(strategyInfo);
|
|
totalScore += result.effectivenessScore;
|
|
successCount++;
|
|
|
|
if (result.effectivenessScore > comparison.bestScore) {
|
|
comparison.bestStrategy = target;
|
|
comparison.bestScore = result.effectivenessScore;
|
|
}
|
|
}
|
|
});
|
|
|
|
comparison.averageScore = successCount > 0 ? totalScore / successCount : 0;
|
|
|
|
return comparison;
|
|
}
|
|
|
|
module.exports = {
|
|
generateWithAdversarialContext, // ← MAIN ENTRY POINT
|
|
generateSimpleAdversarial,
|
|
generateAdvancedAdversarial,
|
|
diagnosticAdversarialPipeline,
|
|
compareAdversarialStrategies,
|
|
getAdversarialDetectorInfo,
|
|
calculateAdversarialEffectiveness
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/ComparisonFramework.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FRAMEWORK DE COMPARAISON ADVERSARIAL
|
|
// Responsabilité: Comparer pipelines normales vs adversariales
|
|
// Utilisation: A/B testing et validation efficacité anti-détection
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
// Pipelines à comparer
|
|
const { generateWithContext } = require('../ContentGeneration'); // Pipeline normale
|
|
const { generateWithAdversarialContext, compareAdversarialStrategies } = require('./ContentGenerationAdversarial'); // Pipeline adversariale
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - COMPARAISON A/B PIPELINE
|
|
* Compare pipeline normale vs adversariale sur même input
|
|
*/
|
|
async function compareNormalVsAdversarial(input, options = {}) {
|
|
return await tracer.run('ComparisonFramework.compareNormalVsAdversarial()', async () => {
|
|
const {
|
|
hierarchy,
|
|
csvData,
|
|
adversarialConfig = {},
|
|
runBothPipelines = true,
|
|
analyzeContent = true
|
|
} = input;
|
|
|
|
const {
|
|
detectorTarget = 'general',
|
|
intensity = 1.0,
|
|
iterations = 1
|
|
} = options;
|
|
|
|
await tracer.annotate({
|
|
comparisonType: 'normal_vs_adversarial',
|
|
detectorTarget,
|
|
intensity,
|
|
iterations,
|
|
elementsCount: Object.keys(hierarchy).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🆚 COMPARAISON A/B: Pipeline normale vs adversariale`, 'INFO');
|
|
logSh(` 🎯 Détecteur cible: ${detectorTarget} | Intensité: ${intensity} | Itérations: ${iterations}`, 'INFO');
|
|
|
|
const results = {
|
|
normal: null,
|
|
adversarial: null,
|
|
comparison: null,
|
|
iterations: []
|
|
};
|
|
|
|
try {
|
|
for (let i = 0; i < iterations; i++) {
|
|
logSh(`🔄 Itération ${i + 1}/${iterations}`, 'INFO');
|
|
|
|
const iterationResults = {
|
|
iteration: i + 1,
|
|
normal: null,
|
|
adversarial: null,
|
|
metrics: {}
|
|
};
|
|
|
|
// ========================================
|
|
// PIPELINE NORMALE
|
|
// ========================================
|
|
if (runBothPipelines) {
|
|
logSh(` 📊 Génération pipeline normale...`, 'DEBUG');
|
|
|
|
const normalStartTime = Date.now();
|
|
try {
|
|
const normalResult = await generateWithContext(hierarchy, csvData, {
|
|
technical: true,
|
|
transitions: true,
|
|
style: true
|
|
});
|
|
|
|
iterationResults.normal = {
|
|
success: true,
|
|
content: normalResult,
|
|
duration: Date.now() - normalStartTime,
|
|
elementsCount: Object.keys(normalResult).length
|
|
};
|
|
|
|
logSh(` ✅ Pipeline normale: ${iterationResults.normal.elementsCount} éléments (${iterationResults.normal.duration}ms)`, 'DEBUG');
|
|
|
|
} catch (error) {
|
|
iterationResults.normal = {
|
|
success: false,
|
|
error: error.message,
|
|
duration: Date.now() - normalStartTime
|
|
};
|
|
|
|
logSh(` ❌ Pipeline normale échouée: ${error.message}`, 'ERROR');
|
|
}
|
|
}
|
|
|
|
// ========================================
|
|
// PIPELINE ADVERSARIALE
|
|
// ========================================
|
|
logSh(` 🎯 Génération pipeline adversariale...`, 'DEBUG');
|
|
|
|
const adversarialStartTime = Date.now();
|
|
try {
|
|
const adversarialResult = await generateWithAdversarialContext({
|
|
hierarchy,
|
|
csvData,
|
|
adversarialConfig: {
|
|
detectorTarget,
|
|
intensity,
|
|
enableAllSteps: true,
|
|
...adversarialConfig
|
|
}
|
|
});
|
|
|
|
iterationResults.adversarial = {
|
|
success: true,
|
|
content: adversarialResult.content,
|
|
stats: adversarialResult.stats,
|
|
adversarialMetrics: adversarialResult.adversarialMetrics,
|
|
duration: Date.now() - adversarialStartTime,
|
|
elementsCount: Object.keys(adversarialResult.content).length
|
|
};
|
|
|
|
logSh(` ✅ Pipeline adversariale: ${iterationResults.adversarial.elementsCount} éléments (${iterationResults.adversarial.duration}ms)`, 'DEBUG');
|
|
logSh(` 📊 Score efficacité: ${adversarialResult.adversarialMetrics.effectivenessScore.toFixed(2)}%`, 'DEBUG');
|
|
|
|
} catch (error) {
|
|
iterationResults.adversarial = {
|
|
success: false,
|
|
error: error.message,
|
|
duration: Date.now() - adversarialStartTime
|
|
};
|
|
|
|
logSh(` ❌ Pipeline adversariale échouée: ${error.message}`, 'ERROR');
|
|
}
|
|
|
|
// ========================================
|
|
// ANALYSE COMPARATIVE ITÉRATION
|
|
// ========================================
|
|
if (analyzeContent && iterationResults.normal?.success && iterationResults.adversarial?.success) {
|
|
iterationResults.metrics = analyzeContentComparison(
|
|
iterationResults.normal.content,
|
|
iterationResults.adversarial.content
|
|
);
|
|
|
|
logSh(` 📈 Diversité: Normal=${iterationResults.metrics.diversity.normal.toFixed(2)}% | Adversarial=${iterationResults.metrics.diversity.adversarial.toFixed(2)}%`, 'DEBUG');
|
|
}
|
|
|
|
results.iterations.push(iterationResults);
|
|
}
|
|
|
|
// ========================================
|
|
// CONSOLIDATION RÉSULTATS
|
|
// ========================================
|
|
const totalDuration = Date.now() - startTime;
|
|
|
|
// Prendre les meilleurs résultats ou derniers si une seule itération
|
|
const lastIteration = results.iterations[results.iterations.length - 1];
|
|
results.normal = lastIteration.normal;
|
|
results.adversarial = lastIteration.adversarial;
|
|
|
|
// Analyse comparative globale
|
|
results.comparison = generateGlobalComparison(results.iterations, options);
|
|
|
|
logSh(`🆚 COMPARAISON TERMINÉE: ${iterations} itérations (${totalDuration}ms)`, 'INFO');
|
|
|
|
if (results.comparison.winner) {
|
|
logSh(`🏆 Gagnant: ${results.comparison.winner} (score: ${results.comparison.bestScore.toFixed(2)})`, 'INFO');
|
|
}
|
|
|
|
await tracer.event('Comparaison A/B terminée', {
|
|
iterations,
|
|
winner: results.comparison.winner,
|
|
totalDuration
|
|
});
|
|
|
|
return results;
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ COMPARAISON A/B ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw new Error(`ComparisonFramework failed: ${error.message}`);
|
|
}
|
|
}, input);
|
|
}
|
|
|
|
/**
|
|
* COMPARAISON MULTI-DÉTECTEURS
|
|
*/
|
|
async function compareMultiDetectors(hierarchy, csvData, detectorTargets = ['general', 'gptZero', 'originality']) {
|
|
logSh(`🎯 COMPARAISON MULTI-DÉTECTEURS: ${detectorTargets.length} stratégies`, 'INFO');
|
|
|
|
const results = {};
|
|
const startTime = Date.now();
|
|
|
|
for (const detector of detectorTargets) {
|
|
logSh(` 🔍 Test détecteur: ${detector}`, 'DEBUG');
|
|
|
|
try {
|
|
const comparison = await compareNormalVsAdversarial({
|
|
hierarchy,
|
|
csvData,
|
|
adversarialConfig: { detectorTarget: detector }
|
|
}, {
|
|
detectorTarget: detector,
|
|
intensity: 1.0,
|
|
iterations: 1
|
|
});
|
|
|
|
results[detector] = {
|
|
success: true,
|
|
comparison,
|
|
effectivenessGain: comparison.adversarial?.adversarialMetrics?.effectivenessScore || 0
|
|
};
|
|
|
|
logSh(` ✅ ${detector}: +${results[detector].effectivenessGain.toFixed(2)}% efficacité`, 'DEBUG');
|
|
|
|
} catch (error) {
|
|
results[detector] = {
|
|
success: false,
|
|
error: error.message,
|
|
effectivenessGain: 0
|
|
};
|
|
|
|
logSh(` ❌ ${detector}: Échec - ${error.message}`, 'ERROR');
|
|
}
|
|
}
|
|
|
|
// Analyse du meilleur détecteur
|
|
const bestDetector = Object.keys(results).reduce((best, current) => {
|
|
if (!results[best]?.success) return current;
|
|
if (!results[current]?.success) return best;
|
|
return results[current].effectivenessGain > results[best].effectivenessGain ? current : best;
|
|
});
|
|
|
|
const totalDuration = Date.now() - startTime;
|
|
|
|
logSh(`🎯 MULTI-DÉTECTEURS TERMINÉ: Meilleur=${bestDetector} (${totalDuration}ms)`, 'INFO');
|
|
|
|
return {
|
|
results,
|
|
bestDetector,
|
|
bestScore: results[bestDetector]?.effectivenessGain || 0,
|
|
totalDuration
|
|
};
|
|
}
|
|
|
|
/**
|
|
* BENCHMARK PERFORMANCE
|
|
*/
|
|
async function benchmarkPerformance(hierarchy, csvData, configurations = []) {
|
|
const defaultConfigs = [
|
|
{ name: 'Normal', type: 'normal' },
|
|
{ name: 'Simple Adversarial', type: 'adversarial', detectorTarget: 'general', intensity: 0.5 },
|
|
{ name: 'Intense Adversarial', type: 'adversarial', detectorTarget: 'gptZero', intensity: 1.0 },
|
|
{ name: 'Max Adversarial', type: 'adversarial', detectorTarget: 'originality', intensity: 1.5 }
|
|
];
|
|
|
|
const configs = configurations.length > 0 ? configurations : defaultConfigs;
|
|
|
|
logSh(`⚡ BENCHMARK PERFORMANCE: ${configs.length} configurations`, 'INFO');
|
|
|
|
const results = [];
|
|
|
|
for (const config of configs) {
|
|
logSh(` 🔧 Test: ${config.name}`, 'DEBUG');
|
|
|
|
const startTime = Date.now();
|
|
|
|
try {
|
|
let result;
|
|
|
|
if (config.type === 'normal') {
|
|
result = await generateWithContext(hierarchy, csvData);
|
|
} else {
|
|
const adversarialResult = await generateWithAdversarialContext({
|
|
hierarchy,
|
|
csvData,
|
|
adversarialConfig: {
|
|
detectorTarget: config.detectorTarget || 'general',
|
|
intensity: config.intensity || 1.0
|
|
}
|
|
});
|
|
result = adversarialResult.content;
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
|
|
results.push({
|
|
name: config.name,
|
|
type: config.type,
|
|
success: true,
|
|
duration,
|
|
elementsCount: Object.keys(result).length,
|
|
performance: Object.keys(result).length / (duration / 1000) // éléments par seconde
|
|
});
|
|
|
|
logSh(` ✅ ${config.name}: ${Object.keys(result).length} éléments (${duration}ms)`, 'DEBUG');
|
|
|
|
} catch (error) {
|
|
results.push({
|
|
name: config.name,
|
|
type: config.type,
|
|
success: false,
|
|
error: error.message,
|
|
duration: Date.now() - startTime
|
|
});
|
|
|
|
logSh(` ❌ ${config.name}: Échec - ${error.message}`, 'ERROR');
|
|
}
|
|
}
|
|
|
|
// Analyser les résultats
|
|
const successfulResults = results.filter(r => r.success);
|
|
const fastest = successfulResults.reduce((best, current) =>
|
|
current.duration < best.duration ? current : best, successfulResults[0]);
|
|
const mostEfficient = successfulResults.reduce((best, current) =>
|
|
current.performance > best.performance ? current : best, successfulResults[0]);
|
|
|
|
logSh(`⚡ BENCHMARK TERMINÉ: Fastest=${fastest?.name} | Most efficient=${mostEfficient?.name}`, 'INFO');
|
|
|
|
return {
|
|
results,
|
|
fastest,
|
|
mostEfficient,
|
|
summary: {
|
|
totalConfigs: configs.length,
|
|
successful: successfulResults.length,
|
|
failed: results.length - successfulResults.length
|
|
}
|
|
};
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
/**
|
|
* Analyser différences de contenu entre normal et adversarial
|
|
*/
|
|
function analyzeContentComparison(normalContent, adversarialContent) {
|
|
const metrics = {
|
|
diversity: {
|
|
normal: analyzeDiversityScore(Object.values(normalContent).join(' ')),
|
|
adversarial: analyzeDiversityScore(Object.values(adversarialContent).join(' '))
|
|
},
|
|
length: {
|
|
normal: Object.values(normalContent).join(' ').length,
|
|
adversarial: Object.values(adversarialContent).join(' ').length
|
|
},
|
|
elementsCount: {
|
|
normal: Object.keys(normalContent).length,
|
|
adversarial: Object.keys(adversarialContent).length
|
|
},
|
|
differences: compareContentElements(normalContent, adversarialContent)
|
|
};
|
|
|
|
return metrics;
|
|
}
|
|
|
|
/**
|
|
* Score de diversité lexicale
|
|
*/
|
|
function analyzeDiversityScore(content) {
|
|
if (!content || typeof content !== 'string') return 0;
|
|
|
|
const words = content.split(/\s+/).filter(w => w.length > 2);
|
|
if (words.length === 0) return 0;
|
|
|
|
const uniqueWords = [...new Set(words.map(w => w.toLowerCase()))];
|
|
return (uniqueWords.length / words.length) * 100;
|
|
}
|
|
|
|
/**
|
|
* Comparer éléments de contenu
|
|
*/
|
|
function compareContentElements(normalContent, adversarialContent) {
|
|
const differences = {
|
|
modified: 0,
|
|
identical: 0,
|
|
totalElements: Math.max(Object.keys(normalContent).length, Object.keys(adversarialContent).length)
|
|
};
|
|
|
|
const allTags = [...new Set([...Object.keys(normalContent), ...Object.keys(adversarialContent)])];
|
|
|
|
allTags.forEach(tag => {
|
|
if (normalContent[tag] && adversarialContent[tag]) {
|
|
if (normalContent[tag] === adversarialContent[tag]) {
|
|
differences.identical++;
|
|
} else {
|
|
differences.modified++;
|
|
}
|
|
}
|
|
});
|
|
|
|
differences.modificationRate = differences.totalElements > 0 ?
|
|
(differences.modified / differences.totalElements) * 100 : 0;
|
|
|
|
return differences;
|
|
}
|
|
|
|
/**
|
|
* Générer analyse comparative globale
|
|
*/
|
|
function generateGlobalComparison(iterations, options) {
|
|
const successfulIterations = iterations.filter(it =>
|
|
it.normal?.success && it.adversarial?.success);
|
|
|
|
if (successfulIterations.length === 0) {
|
|
return {
|
|
winner: null,
|
|
bestScore: 0,
|
|
summary: 'Aucune itération réussie'
|
|
};
|
|
}
|
|
|
|
// Moyenner les métriques
|
|
const avgMetrics = {
|
|
diversity: {
|
|
normal: 0,
|
|
adversarial: 0
|
|
},
|
|
performance: {
|
|
normal: 0,
|
|
adversarial: 0
|
|
}
|
|
};
|
|
|
|
successfulIterations.forEach(iteration => {
|
|
if (iteration.metrics) {
|
|
avgMetrics.diversity.normal += iteration.metrics.diversity.normal;
|
|
avgMetrics.diversity.adversarial += iteration.metrics.diversity.adversarial;
|
|
}
|
|
avgMetrics.performance.normal += iteration.normal.elementsCount / (iteration.normal.duration / 1000);
|
|
avgMetrics.performance.adversarial += iteration.adversarial.elementsCount / (iteration.adversarial.duration / 1000);
|
|
});
|
|
|
|
const iterCount = successfulIterations.length;
|
|
avgMetrics.diversity.normal /= iterCount;
|
|
avgMetrics.diversity.adversarial /= iterCount;
|
|
avgMetrics.performance.normal /= iterCount;
|
|
avgMetrics.performance.adversarial /= iterCount;
|
|
|
|
// Déterminer le gagnant
|
|
const diversityGain = avgMetrics.diversity.adversarial - avgMetrics.diversity.normal;
|
|
const performanceLoss = avgMetrics.performance.normal - avgMetrics.performance.adversarial;
|
|
|
|
// Score composite (favorise diversité avec pénalité performance)
|
|
const adversarialScore = diversityGain * 2 - (performanceLoss * 0.5);
|
|
|
|
return {
|
|
winner: adversarialScore > 5 ? 'adversarial' : 'normal',
|
|
bestScore: Math.max(avgMetrics.diversity.normal, avgMetrics.diversity.adversarial),
|
|
diversityGain,
|
|
performanceLoss,
|
|
avgMetrics,
|
|
summary: `Diversité: +${diversityGain.toFixed(2)}%, Performance: ${performanceLoss > 0 ? '-' : '+'}${Math.abs(performanceLoss).toFixed(2)} elem/s`
|
|
};
|
|
}
|
|
|
|
module.exports = {
|
|
compareNormalVsAdversarial, // ← MAIN ENTRY POINT
|
|
compareMultiDetectors,
|
|
benchmarkPerformance,
|
|
analyzeContentComparison,
|
|
analyzeDiversityScore
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/adversarial-generation/demo-modulaire.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// DÉMONSTRATION ARCHITECTURE MODULAIRE
|
|
// Usage: node lib/adversarial-generation/demo-modulaire.js
|
|
// Objectif: Valider l'intégration modulaire adversariale
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
|
|
// Import modules adversariaux modulaires
|
|
const { applyAdversarialLayer } = require('./AdversarialCore');
|
|
const {
|
|
applyPredefinedStack,
|
|
applyAdaptiveLayers,
|
|
getAvailableStacks
|
|
} = require('./AdversarialLayers');
|
|
const { calculateAntiDetectionScore, evaluateAdversarialImprovement } = require('./AdversarialUtils');
|
|
|
|
/**
|
|
* EXEMPLE D'UTILISATION MODULAIRE
|
|
*/
|
|
async function demoModularAdversarial() {
|
|
console.log('\n🎯 === DÉMONSTRATION ADVERSARIAL MODULAIRE ===\n');
|
|
|
|
// Contenu d'exemple (simulé contenu généré normal)
|
|
const exempleContenu = {
|
|
'|Titre_Principal_1|': 'Guide complet pour choisir votre plaque personnalisée',
|
|
'|Introduction_1|': 'La personnalisation d\'une plaque signalétique représente un enjeu optimal pour votre entreprise. Cette solution comprehensive permet de créer une identité visuelle robuste et seamless.',
|
|
'|Texte_1|': 'Il est important de noter que les matériaux utilisés sont cutting-edge. Par ailleurs, la qualité est optimal. En effet, nos solutions sont comprehensive et robust.',
|
|
'|FAQ_Question_1|': 'Quels sont les matériaux disponibles ?',
|
|
'|FAQ_Reponse_1|': 'Nos matériaux sont optimal : dibond, aluminium, PMMA. Ces solutions comprehensive garantissent une qualité robust et seamless.'
|
|
};
|
|
|
|
console.log('📊 CONTENU ORIGINAL:');
|
|
Object.entries(exempleContenu).forEach(([tag, content]) => {
|
|
console.log(` ${tag}: "${content.substring(0, 60)}..."`);
|
|
});
|
|
|
|
// Analyser contenu original
|
|
const scoreOriginal = calculateAntiDetectionScore(Object.values(exempleContenu).join(' '));
|
|
console.log(`\n📈 Score anti-détection original: ${scoreOriginal}/100`);
|
|
|
|
try {
|
|
// ========================================
|
|
// TEST 1: COUCHE SIMPLE
|
|
// ========================================
|
|
console.log('\n🔧 TEST 1: Application couche adversariale simple');
|
|
|
|
const result1 = await applyAdversarialLayer(exempleContenu, {
|
|
detectorTarget: 'general',
|
|
intensity: 0.8,
|
|
method: 'enhancement'
|
|
});
|
|
|
|
console.log(`✅ Résultat: ${result1.stats.elementsModified}/${result1.stats.elementsProcessed} éléments modifiés`);
|
|
|
|
const scoreAmeliore = calculateAntiDetectionScore(Object.values(result1.content).join(' '));
|
|
console.log(`📈 Score anti-détection amélioré: ${scoreAmeliore}/100 (+${scoreAmeliore - scoreOriginal})`);
|
|
|
|
// ========================================
|
|
// TEST 2: STACK PRÉDÉFINI
|
|
// ========================================
|
|
console.log('\n📦 TEST 2: Application stack prédéfini');
|
|
|
|
// Lister stacks disponibles
|
|
const stacks = getAvailableStacks();
|
|
console.log(' Stacks disponibles:');
|
|
stacks.forEach(stack => {
|
|
console.log(` - ${stack.name}: ${stack.description} (${stack.layersCount} couches)`);
|
|
});
|
|
|
|
const result2 = await applyPredefinedStack(exempleContenu, 'standardDefense', {
|
|
csvData: {
|
|
personality: { nom: 'Marc', style: 'technique' },
|
|
mc0: 'plaque personnalisée'
|
|
}
|
|
});
|
|
|
|
console.log(`✅ Stack standard: ${result2.stats.totalModifications} modifications totales`);
|
|
console.log(` 📊 Couches appliquées: ${result2.stats.layers.filter(l => l.success).length}/${result2.stats.layers.length}`);
|
|
|
|
const scoreStack = calculateAntiDetectionScore(Object.values(result2.content).join(' '));
|
|
console.log(`📈 Score anti-détection stack: ${scoreStack}/100 (+${scoreStack - scoreOriginal})`);
|
|
|
|
// ========================================
|
|
// TEST 3: COUCHES ADAPTATIVES
|
|
// ========================================
|
|
console.log('\n🧠 TEST 3: Application couches adaptatives');
|
|
|
|
const result3 = await applyAdaptiveLayers(exempleContenu, {
|
|
targetDetectors: ['gptZero', 'originality'],
|
|
maxIntensity: 1.2
|
|
});
|
|
|
|
if (result3.stats.adaptive) {
|
|
console.log(`✅ Adaptatif: ${result3.stats.layersApplied || result3.stats.totalModifications} modifications`);
|
|
|
|
const scoreAdaptatif = calculateAntiDetectionScore(Object.values(result3.content).join(' '));
|
|
console.log(`📈 Score anti-détection adaptatif: ${scoreAdaptatif}/100 (+${scoreAdaptatif - scoreOriginal})`);
|
|
}
|
|
|
|
// ========================================
|
|
// COMPARAISON FINALE
|
|
// ========================================
|
|
console.log('\n📊 COMPARAISON FINALE:');
|
|
|
|
const evaluation = evaluateAdversarialImprovement(
|
|
Object.values(exempleContenu).join(' '),
|
|
Object.values(result2.content).join(' '),
|
|
'general'
|
|
);
|
|
|
|
console.log(` 🔹 Réduction empreintes IA: ${evaluation.fingerprintReduction.toFixed(2)}%`);
|
|
console.log(` 🔹 Augmentation diversité: ${evaluation.diversityIncrease.toFixed(2)}%`);
|
|
console.log(` 🔹 Amélioration variation: ${evaluation.variationIncrease.toFixed(2)}%`);
|
|
console.log(` 🔹 Score amélioration global: ${evaluation.improvementScore}`);
|
|
console.log(` 🔹 Taux modification: ${evaluation.modificationRate.toFixed(2)}%`);
|
|
console.log(` 💡 Recommandation: ${evaluation.recommendation}`);
|
|
|
|
// ========================================
|
|
// EXEMPLES DE CONTENU TRANSFORMÉ
|
|
// ========================================
|
|
console.log('\n✨ EXEMPLES DE TRANSFORMATION:');
|
|
|
|
const exempleTransforme = result2.content['|Introduction_1|'] || result1.content['|Introduction_1|'];
|
|
console.log('\n📝 AVANT:');
|
|
console.log(` "${exempleContenu['|Introduction_1|']}"`);
|
|
console.log('\n📝 APRÈS:');
|
|
console.log(` "${exempleTransforme}"`);
|
|
|
|
console.log('\n✅ === DÉMONSTRATION MODULAIRE TERMINÉE ===\n');
|
|
|
|
return {
|
|
success: true,
|
|
originalScore: scoreOriginal,
|
|
improvedScore: Math.max(scoreAmeliore, scoreStack),
|
|
improvement: evaluation.improvementScore
|
|
};
|
|
|
|
} catch (error) {
|
|
console.error('\n❌ ERREUR DÉMONSTRATION:', error.message);
|
|
return { success: false, error: error.message };
|
|
}
|
|
}
|
|
|
|
/**
|
|
* EXEMPLE D'INTÉGRATION AVEC PIPELINE NORMALE
|
|
*/
|
|
async function demoIntegrationPipeline() {
|
|
console.log('\n🔗 === DÉMONSTRATION INTÉGRATION PIPELINE ===\n');
|
|
|
|
// Simuler résultat pipeline normale (Level 1)
|
|
const contenuNormal = {
|
|
'|Titre_H1_1|': 'Solutions de plaques personnalisées professionnelles',
|
|
'|Intro_1|': 'Notre expertise en signalétique permet de créer des plaques sur mesure adaptées à vos besoins spécifiques.',
|
|
'|Texte_1|': 'Les matériaux proposés incluent l\'aluminium, le dibond et le PMMA. Chaque solution présente des avantages particuliers selon l\'usage prévu.'
|
|
};
|
|
|
|
console.log('💼 SCÉNARIO: Application adversarial post-pipeline normale');
|
|
|
|
try {
|
|
// Exemple Level 6 - Post-processing adversarial
|
|
console.log('\n🎯 Étape 1: Contenu généré par pipeline normale');
|
|
console.log(' ✅ Contenu de base: qualité préservée');
|
|
|
|
console.log('\n🎯 Étape 2: Application couche adversariale modulaire');
|
|
const resultAdversarial = await applyAdversarialLayer(contenuNormal, {
|
|
detectorTarget: 'gptZero',
|
|
intensity: 0.9,
|
|
method: 'hybrid',
|
|
preserveStructure: true
|
|
});
|
|
|
|
console.log(` ✅ Couche adversariale: ${resultAdversarial.stats.elementsModified} éléments modifiés`);
|
|
|
|
console.log('\n📊 RÉSULTAT FINAL:');
|
|
Object.entries(resultAdversarial.content).forEach(([tag, content]) => {
|
|
console.log(` ${tag}:`);
|
|
console.log(` AVANT: "${contenuNormal[tag]}"`);
|
|
console.log(` APRÈS: "${content}"`);
|
|
console.log('');
|
|
});
|
|
|
|
return { success: true, result: resultAdversarial };
|
|
|
|
} catch (error) {
|
|
console.error('❌ ERREUR INTÉGRATION:', error.message);
|
|
return { success: false, error: error.message };
|
|
}
|
|
}
|
|
|
|
// Exécuter démonstrations si fichier appelé directement
|
|
if (require.main === module) {
|
|
(async () => {
|
|
await demoModularAdversarial();
|
|
await demoIntegrationPipeline();
|
|
})().catch(console.error);
|
|
}
|
|
|
|
module.exports = {
|
|
demoModularAdversarial,
|
|
demoIntegrationPipeline
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/selective-enhancement/SelectiveUtils.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// SELECTIVE UTILS - UTILITAIRES MODULAIRES
|
|
// Responsabilité: Fonctions utilitaires partagées par tous les modules selective
|
|
// Architecture: Helper functions réutilisables et composables
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
|
|
/**
|
|
* ANALYSEURS DE CONTENU SELECTIVE
|
|
*/
|
|
|
|
/**
|
|
* Analyser qualité technique d'un contenu
|
|
*/
|
|
function analyzeTechnicalQuality(content, contextualTerms = []) {
|
|
if (!content || typeof content !== 'string') return { score: 0, details: {} };
|
|
|
|
const analysis = {
|
|
score: 0,
|
|
details: {
|
|
technicalTermsFound: 0,
|
|
technicalTermsExpected: contextualTerms.length,
|
|
genericWordsCount: 0,
|
|
hasSpecifications: false,
|
|
hasDimensions: false,
|
|
contextIntegration: 0
|
|
}
|
|
};
|
|
|
|
const lowerContent = content.toLowerCase();
|
|
|
|
// 1. Compter termes techniques présents
|
|
contextualTerms.forEach(term => {
|
|
if (lowerContent.includes(term.toLowerCase())) {
|
|
analysis.details.technicalTermsFound++;
|
|
}
|
|
});
|
|
|
|
// 2. Détecter mots génériques
|
|
const genericWords = ['produit', 'solution', 'service', 'offre', 'article', 'élément'];
|
|
analysis.details.genericWordsCount = genericWords.filter(word =>
|
|
lowerContent.includes(word)
|
|
).length;
|
|
|
|
// 3. Vérifier spécifications techniques
|
|
analysis.details.hasSpecifications = /\b(norme|iso|din|ce)\b/i.test(content);
|
|
|
|
// 4. Vérifier dimensions/données techniques
|
|
analysis.details.hasDimensions = /\d+\s*(mm|cm|m|%|°|kg|g)\b/i.test(content);
|
|
|
|
// 5. Calculer score global (0-100)
|
|
const termRatio = contextualTerms.length > 0 ?
|
|
(analysis.details.technicalTermsFound / contextualTerms.length) * 40 : 20;
|
|
const genericPenalty = Math.min(20, analysis.details.genericWordsCount * 5);
|
|
const specificationBonus = analysis.details.hasSpecifications ? 15 : 0;
|
|
const dimensionBonus = analysis.details.hasDimensions ? 15 : 0;
|
|
const lengthBonus = content.length > 100 ? 10 : 0;
|
|
|
|
analysis.score = Math.max(0, Math.min(100,
|
|
termRatio + specificationBonus + dimensionBonus + lengthBonus - genericPenalty
|
|
));
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* Analyser fluidité des transitions
|
|
*/
|
|
function analyzeTransitionFluidity(content) {
|
|
if (!content || typeof content !== 'string') return { score: 0, details: {} };
|
|
|
|
const sentences = content.split(/[.!?]+/)
|
|
.map(s => s.trim())
|
|
.filter(s => s.length > 5);
|
|
|
|
if (sentences.length < 2) {
|
|
return { score: 100, details: { reason: 'Contenu trop court pour analyse transitions' } };
|
|
}
|
|
|
|
const analysis = {
|
|
score: 0,
|
|
details: {
|
|
sentencesCount: sentences.length,
|
|
connectorsFound: 0,
|
|
repetitiveConnectors: 0,
|
|
abruptTransitions: 0,
|
|
averageSentenceLength: 0,
|
|
lengthVariation: 0
|
|
}
|
|
};
|
|
|
|
// 1. Analyser connecteurs
|
|
const commonConnectors = ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc', 'ensuite'];
|
|
const connectorCounts = {};
|
|
|
|
commonConnectors.forEach(connector => {
|
|
const matches = (content.match(new RegExp(`\\b${connector}\\b`, 'gi')) || []);
|
|
connectorCounts[connector] = matches.length;
|
|
analysis.details.connectorsFound += matches.length;
|
|
if (matches.length > 1) analysis.details.repetitiveConnectors++;
|
|
});
|
|
|
|
// 2. Détecter transitions abruptes
|
|
for (let i = 1; i < sentences.length; i++) {
|
|
const sentence = sentences[i].toLowerCase().trim();
|
|
const hasConnector = commonConnectors.some(connector =>
|
|
sentence.startsWith(connector) || sentence.includes(` ${connector} `)
|
|
);
|
|
|
|
if (!hasConnector && sentence.length > 20) {
|
|
analysis.details.abruptTransitions++;
|
|
}
|
|
}
|
|
|
|
// 3. Analyser variation de longueur
|
|
const lengths = sentences.map(s => s.split(/\s+/).length);
|
|
analysis.details.averageSentenceLength = lengths.reduce((a, b) => a + b, 0) / lengths.length;
|
|
|
|
const variance = lengths.reduce((acc, len) =>
|
|
acc + Math.pow(len - analysis.details.averageSentenceLength, 2), 0
|
|
) / lengths.length;
|
|
analysis.details.lengthVariation = Math.sqrt(variance);
|
|
|
|
// 4. Calculer score fluidité (0-100)
|
|
const connectorScore = Math.min(30, (analysis.details.connectorsFound / sentences.length) * 100);
|
|
const repetitionPenalty = Math.min(20, analysis.details.repetitiveConnectors * 5);
|
|
const abruptPenalty = Math.min(30, (analysis.details.abruptTransitions / sentences.length) * 50);
|
|
const variationScore = Math.min(20, analysis.details.lengthVariation * 2);
|
|
|
|
analysis.score = Math.max(0, Math.min(100,
|
|
connectorScore + variationScore - repetitionPenalty - abruptPenalty + 50
|
|
));
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* Analyser cohérence de style
|
|
*/
|
|
function analyzeStyleConsistency(content, expectedPersonality = null) {
|
|
if (!content || typeof content !== 'string') return { score: 0, details: {} };
|
|
|
|
const analysis = {
|
|
score: 0,
|
|
details: {
|
|
personalityAlignment: 0,
|
|
toneConsistency: 0,
|
|
vocabularyLevel: 'standard',
|
|
formalityScore: 0,
|
|
personalityWordsFound: 0
|
|
}
|
|
};
|
|
|
|
// 1. Analyser alignement personnalité
|
|
if (expectedPersonality && expectedPersonality.vocabulairePref) {
|
|
const personalityWords = expectedPersonality.vocabulairePref.toLowerCase().split(',');
|
|
const contentLower = content.toLowerCase();
|
|
|
|
personalityWords.forEach(word => {
|
|
if (word.trim() && contentLower.includes(word.trim())) {
|
|
analysis.details.personalityWordsFound++;
|
|
}
|
|
});
|
|
|
|
analysis.details.personalityAlignment = personalityWords.length > 0 ?
|
|
(analysis.details.personalityWordsFound / personalityWords.length) * 100 : 0;
|
|
}
|
|
|
|
// 2. Analyser niveau vocabulaire
|
|
const technicalWords = content.match(/\b\w{8,}\b/g) || [];
|
|
const totalWords = content.split(/\s+/).length;
|
|
const techRatio = technicalWords.length / totalWords;
|
|
|
|
if (techRatio > 0.15) analysis.details.vocabularyLevel = 'expert';
|
|
else if (techRatio < 0.05) analysis.details.vocabularyLevel = 'accessible';
|
|
else analysis.details.vocabularyLevel = 'standard';
|
|
|
|
// 3. Analyser formalité
|
|
const formalIndicators = ['il convient de', 'par conséquent', 'néanmoins', 'toutefois'];
|
|
const casualIndicators = ['du coup', 'sympa', 'cool', 'nickel'];
|
|
|
|
let formalCount = formalIndicators.filter(indicator =>
|
|
content.toLowerCase().includes(indicator)
|
|
).length;
|
|
|
|
let casualCount = casualIndicators.filter(indicator =>
|
|
content.toLowerCase().includes(indicator)
|
|
).length;
|
|
|
|
analysis.details.formalityScore = formalCount - casualCount; // Positif = formel, négatif = casual
|
|
|
|
// 4. Calculer score cohérence (0-100)
|
|
let baseScore = 50;
|
|
|
|
if (expectedPersonality) {
|
|
baseScore += analysis.details.personalityAlignment * 0.3;
|
|
|
|
// Ajustements selon niveau technique attendu
|
|
const expectedLevel = expectedPersonality.niveauTechnique || 'standard';
|
|
if (expectedLevel === analysis.details.vocabularyLevel) {
|
|
baseScore += 20;
|
|
} else {
|
|
baseScore -= 10;
|
|
}
|
|
}
|
|
|
|
// Bonus cohérence tonale
|
|
const sentences = content.split(/[.!?]+/).filter(s => s.length > 10);
|
|
if (sentences.length > 1) {
|
|
baseScore += Math.min(20, analysis.details.lengthVariation || 10);
|
|
}
|
|
|
|
analysis.score = Math.max(0, Math.min(100, baseScore));
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* COMPARATEURS ET MÉTRIQUES
|
|
*/
|
|
|
|
/**
|
|
* Comparer deux contenus et calculer taux amélioration
|
|
*/
|
|
function compareContentImprovement(original, enhanced, analysisType = 'general') {
|
|
if (!original || !enhanced) return { improvementRate: 0, details: {} };
|
|
|
|
const comparison = {
|
|
improvementRate: 0,
|
|
details: {
|
|
lengthChange: ((enhanced.length - original.length) / original.length) * 100,
|
|
wordCountChange: 0,
|
|
structuralChanges: 0,
|
|
contentPreserved: true
|
|
}
|
|
};
|
|
|
|
// 1. Analyser changements structurels
|
|
const originalSentences = original.split(/[.!?]+/).length;
|
|
const enhancedSentences = enhanced.split(/[.!?]+/).length;
|
|
comparison.details.structuralChanges = Math.abs(enhancedSentences - originalSentences);
|
|
|
|
// 2. Analyser changements de mots
|
|
const originalWords = original.toLowerCase().split(/\s+/).filter(w => w.length > 2);
|
|
const enhancedWords = enhanced.toLowerCase().split(/\s+/).filter(w => w.length > 2);
|
|
comparison.details.wordCountChange = enhancedWords.length - originalWords.length;
|
|
|
|
// 3. Vérifier préservation du contenu principal
|
|
const originalKeyWords = originalWords.filter(w => w.length > 4);
|
|
const preservedWords = originalKeyWords.filter(w => enhanced.toLowerCase().includes(w));
|
|
comparison.details.contentPreserved = (preservedWords.length / originalKeyWords.length) > 0.7;
|
|
|
|
// 4. Calculer taux amélioration selon type d'analyse
|
|
switch (analysisType) {
|
|
case 'technical':
|
|
const originalTech = analyzeTechnicalQuality(original);
|
|
const enhancedTech = analyzeTechnicalQuality(enhanced);
|
|
comparison.improvementRate = enhancedTech.score - originalTech.score;
|
|
break;
|
|
|
|
case 'transitions':
|
|
const originalFluid = analyzeTransitionFluidity(original);
|
|
const enhancedFluid = analyzeTransitionFluidity(enhanced);
|
|
comparison.improvementRate = enhancedFluid.score - originalFluid.score;
|
|
break;
|
|
|
|
case 'style':
|
|
const originalStyle = analyzeStyleConsistency(original);
|
|
const enhancedStyle = analyzeStyleConsistency(enhanced);
|
|
comparison.improvementRate = enhancedStyle.score - originalStyle.score;
|
|
break;
|
|
|
|
default:
|
|
// Amélioration générale (moyenne pondérée)
|
|
comparison.improvementRate = Math.min(50, Math.abs(comparison.details.lengthChange) * 0.1 +
|
|
(comparison.details.contentPreserved ? 20 : -20) +
|
|
Math.min(15, Math.abs(comparison.details.wordCountChange)));
|
|
}
|
|
|
|
return comparison;
|
|
}
|
|
|
|
/**
|
|
* UTILITAIRES DE CONTENU
|
|
*/
|
|
|
|
/**
|
|
* Nettoyer contenu généré par LLM
|
|
*/
|
|
function cleanGeneratedContent(content, cleaningLevel = 'standard') {
|
|
if (!content || typeof content !== 'string') return content;
|
|
|
|
let cleaned = content.trim();
|
|
|
|
// Nettoyage de base
|
|
cleaned = cleaned.replace(/^(voici\s+)?le\s+contenu\s+(amélioré|modifié|réécrit)[:\s]*/gi, '');
|
|
cleaned = cleaned.replace(/^(bon,?\s*)?(alors,?\s*)?(voici\s+)?/gi, '');
|
|
cleaned = cleaned.replace(/^(avec\s+les?\s+)?améliorations?\s*[:\s]*/gi, '');
|
|
|
|
// Nettoyage formatage
|
|
cleaned = cleaned.replace(/\*\*([^*]+)\*\*/g, '$1'); // Gras markdown → texte normal
|
|
cleaned = cleaned.replace(/\s{2,}/g, ' '); // Espaces multiples
|
|
cleaned = cleaned.replace(/([.!?])\s*([.!?])/g, '$1 '); // Double ponctuation
|
|
|
|
if (cleaningLevel === 'intensive') {
|
|
// Nettoyage intensif
|
|
cleaned = cleaned.replace(/^\s*[-*+]\s*/gm, ''); // Puces en début de ligne
|
|
cleaned = cleaned.replace(/^(pour\s+)?(ce\s+)?(contenu\s*)?[,:]?\s*/gi, '');
|
|
cleaned = cleaned.replace(/\([^)]*\)/g, ''); // Parenthèses et contenu
|
|
}
|
|
|
|
// Nettoyage final
|
|
cleaned = cleaned.replace(/^[,.\s]+/, ''); // Début
|
|
cleaned = cleaned.replace(/[,\s]+$/, ''); // Fin
|
|
cleaned = cleaned.trim();
|
|
|
|
return cleaned;
|
|
}
|
|
|
|
/**
|
|
* Valider contenu selective
|
|
*/
|
|
function validateSelectiveContent(content, originalContent, criteria = {}) {
|
|
const validation = {
|
|
isValid: true,
|
|
score: 0,
|
|
issues: [],
|
|
suggestions: []
|
|
};
|
|
|
|
const {
|
|
minLength = 20,
|
|
maxLengthChange = 50, // % de changement maximum
|
|
preserveContent = true,
|
|
checkTechnicalTerms = true
|
|
} = criteria;
|
|
|
|
// 1. Vérifier longueur
|
|
if (!content || content.length < minLength) {
|
|
validation.isValid = false;
|
|
validation.issues.push('Contenu trop court');
|
|
validation.suggestions.push('Augmenter la longueur du contenu généré');
|
|
} else {
|
|
validation.score += 25;
|
|
}
|
|
|
|
// 2. Vérifier changements de longueur
|
|
if (originalContent) {
|
|
const lengthChange = Math.abs((content.length - originalContent.length) / originalContent.length) * 100;
|
|
|
|
if (lengthChange > maxLengthChange) {
|
|
validation.issues.push('Changement de longueur excessif');
|
|
validation.suggestions.push('Réduire l\'intensité d\'amélioration');
|
|
} else {
|
|
validation.score += 25;
|
|
}
|
|
|
|
// 3. Vérifier préservation du contenu
|
|
if (preserveContent) {
|
|
const preservation = compareContentImprovement(originalContent, content);
|
|
|
|
if (!preservation.details.contentPreserved) {
|
|
validation.isValid = false;
|
|
validation.issues.push('Contenu original non préservé');
|
|
validation.suggestions.push('Améliorer conservation du sens original');
|
|
} else {
|
|
validation.score += 25;
|
|
}
|
|
}
|
|
}
|
|
|
|
// 4. Vérifications spécifiques
|
|
if (checkTechnicalTerms) {
|
|
const technicalQuality = analyzeTechnicalQuality(content);
|
|
|
|
if (technicalQuality.score > 60) {
|
|
validation.score += 25;
|
|
} else if (technicalQuality.score < 30) {
|
|
validation.issues.push('Qualité technique insuffisante');
|
|
validation.suggestions.push('Ajouter plus de termes techniques spécialisés');
|
|
}
|
|
}
|
|
|
|
// Score final et validation
|
|
validation.score = Math.min(100, validation.score);
|
|
validation.isValid = validation.isValid && validation.score >= 60;
|
|
|
|
return validation;
|
|
}
|
|
|
|
/**
|
|
* UTILITAIRES TECHNIQUES
|
|
*/
|
|
|
|
/**
|
|
* Chunk array avec gestion intelligente
|
|
*/
|
|
function chunkArray(array, chunkSize, smartChunking = false) {
|
|
if (!Array.isArray(array)) return [];
|
|
if (array.length <= chunkSize) return [array];
|
|
|
|
const chunks = [];
|
|
|
|
if (smartChunking) {
|
|
// Chunking intelligent : éviter de séparer éléments liés
|
|
let currentChunk = [];
|
|
|
|
for (let i = 0; i < array.length; i++) {
|
|
currentChunk.push(array[i]);
|
|
|
|
// Conditions de fin de chunk intelligente
|
|
const isChunkFull = currentChunk.length >= chunkSize;
|
|
const isLastElement = i === array.length - 1;
|
|
const nextElementRelated = i < array.length - 1 &&
|
|
array[i].tag && array[i + 1].tag &&
|
|
array[i].tag.includes('FAQ') && array[i + 1].tag.includes('FAQ');
|
|
|
|
if ((isChunkFull && !nextElementRelated) || isLastElement) {
|
|
chunks.push([...currentChunk]);
|
|
currentChunk = [];
|
|
}
|
|
}
|
|
|
|
// Ajouter chunk restant si non vide
|
|
if (currentChunk.length > 0) {
|
|
if (chunks.length > 0 && chunks[chunks.length - 1].length + currentChunk.length <= chunkSize * 1.2) {
|
|
// Merger avec dernier chunk si pas trop gros
|
|
chunks[chunks.length - 1].push(...currentChunk);
|
|
} else {
|
|
chunks.push(currentChunk);
|
|
}
|
|
}
|
|
} else {
|
|
// Chunking standard
|
|
for (let i = 0; i < array.length; i += chunkSize) {
|
|
chunks.push(array.slice(i, i + chunkSize));
|
|
}
|
|
}
|
|
|
|
return chunks;
|
|
}
|
|
|
|
/**
|
|
* Sleep avec logging optionnel
|
|
*/
|
|
async function sleep(ms, logMessage = null) {
|
|
if (logMessage) {
|
|
logSh(`⏳ ${logMessage} (${ms}ms)`, 'DEBUG');
|
|
}
|
|
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
/**
|
|
* Mesurer performance d'opération
|
|
*/
|
|
function measurePerformance(operationName, startTime = Date.now()) {
|
|
const endTime = Date.now();
|
|
const duration = endTime - startTime;
|
|
|
|
const performance = {
|
|
operationName,
|
|
startTime,
|
|
endTime,
|
|
duration,
|
|
durationFormatted: formatDuration(duration)
|
|
};
|
|
|
|
return performance;
|
|
}
|
|
|
|
/**
|
|
* Formater durée en format lisible
|
|
*/
|
|
function formatDuration(ms) {
|
|
if (ms < 1000) return `${ms}ms`;
|
|
if (ms < 60000) return `${(ms / 1000).toFixed(1)}s`;
|
|
return `${Math.floor(ms / 60000)}m ${Math.floor((ms % 60000) / 1000)}s`;
|
|
}
|
|
|
|
/**
|
|
* STATISTIQUES ET RAPPORTS
|
|
*/
|
|
|
|
/**
|
|
* Générer rapport amélioration
|
|
*/
|
|
function generateImprovementReport(originalContent, enhancedContent, layerType = 'general') {
|
|
const report = {
|
|
layerType,
|
|
timestamp: new Date().toISOString(),
|
|
summary: {
|
|
elementsProcessed: 0,
|
|
elementsImproved: 0,
|
|
averageImprovement: 0,
|
|
totalExecutionTime: 0
|
|
},
|
|
details: {
|
|
byElement: [],
|
|
qualityMetrics: {},
|
|
recommendations: []
|
|
}
|
|
};
|
|
|
|
// Analyser chaque élément
|
|
Object.keys(originalContent).forEach(tag => {
|
|
const original = originalContent[tag];
|
|
const enhanced = enhancedContent[tag];
|
|
|
|
if (original && enhanced) {
|
|
report.summary.elementsProcessed++;
|
|
|
|
const improvement = compareContentImprovement(original, enhanced, layerType);
|
|
|
|
if (improvement.improvementRate > 0) {
|
|
report.summary.elementsImproved++;
|
|
}
|
|
|
|
report.summary.averageImprovement += improvement.improvementRate;
|
|
|
|
report.details.byElement.push({
|
|
tag,
|
|
improvementRate: improvement.improvementRate,
|
|
lengthChange: improvement.details.lengthChange,
|
|
contentPreserved: improvement.details.contentPreserved
|
|
});
|
|
}
|
|
});
|
|
|
|
// Calculer moyennes
|
|
if (report.summary.elementsProcessed > 0) {
|
|
report.summary.averageImprovement = report.summary.averageImprovement / report.summary.elementsProcessed;
|
|
}
|
|
|
|
// Métriques qualité globales
|
|
const fullOriginal = Object.values(originalContent).join(' ');
|
|
const fullEnhanced = Object.values(enhancedContent).join(' ');
|
|
|
|
report.details.qualityMetrics = {
|
|
technical: analyzeTechnicalQuality(fullEnhanced),
|
|
transitions: analyzeTransitionFluidity(fullEnhanced),
|
|
style: analyzeStyleConsistency(fullEnhanced)
|
|
};
|
|
|
|
// Recommandations
|
|
if (report.summary.averageImprovement < 10) {
|
|
report.details.recommendations.push('Augmenter l\'intensité d\'amélioration');
|
|
}
|
|
|
|
if (report.details.byElement.some(e => !e.contentPreserved)) {
|
|
report.details.recommendations.push('Améliorer préservation du contenu original');
|
|
}
|
|
|
|
return report;
|
|
}
|
|
|
|
module.exports = {
|
|
// Analyseurs
|
|
analyzeTechnicalQuality,
|
|
analyzeTransitionFluidity,
|
|
analyzeStyleConsistency,
|
|
|
|
// Comparateurs
|
|
compareContentImprovement,
|
|
|
|
// Utilitaires contenu
|
|
cleanGeneratedContent,
|
|
validateSelectiveContent,
|
|
|
|
// Utilitaires techniques
|
|
chunkArray,
|
|
sleep,
|
|
measurePerformance,
|
|
formatDuration,
|
|
|
|
// Rapports
|
|
generateImprovementReport
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/selective-enhancement/TechnicalLayer.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// TECHNICAL LAYER - COUCHE TECHNIQUE MODULAIRE
|
|
// Responsabilité: Amélioration technique modulaire réutilisable
|
|
// LLM: GPT-4o-mini (précision technique optimale)
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { chunkArray, sleep } = require('./SelectiveUtils');
|
|
|
|
/**
|
|
* COUCHE TECHNIQUE MODULAIRE
|
|
*/
|
|
class TechnicalLayer {
|
|
constructor() {
|
|
this.name = 'TechnicalEnhancement';
|
|
this.defaultLLM = 'openai';
|
|
this.priority = 1; // Haute priorité - appliqué en premier généralement
|
|
}
|
|
|
|
/**
|
|
* MAIN METHOD - Appliquer amélioration technique
|
|
*/
|
|
async apply(content, config = {}) {
|
|
return await tracer.run('TechnicalLayer.apply()', async () => {
|
|
const {
|
|
llmProvider = this.defaultLLM,
|
|
intensity = 1.0, // 0.0-2.0 intensité d'amélioration
|
|
analysisMode = true, // Analyser avant d'appliquer
|
|
csvData = null,
|
|
preserveStructure = true,
|
|
targetTerms = null // Termes techniques ciblés
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
technicalLayer: true,
|
|
llmProvider,
|
|
intensity,
|
|
elementsCount: Object.keys(content).length,
|
|
mc0: csvData?.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`⚙️ TECHNICAL LAYER: Amélioration technique (${llmProvider})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments | Intensité: ${intensity}`, 'INFO');
|
|
|
|
try {
|
|
let enhancedContent = {};
|
|
let elementsProcessed = 0;
|
|
let elementsEnhanced = 0;
|
|
|
|
if (analysisMode) {
|
|
// 1. Analyser éléments nécessitant amélioration technique
|
|
const analysis = await this.analyzeTechnicalNeeds(content, csvData, targetTerms);
|
|
|
|
logSh(` 📋 Analyse: ${analysis.candidates.length}/${Object.keys(content).length} éléments candidats`, 'DEBUG');
|
|
|
|
if (analysis.candidates.length === 0) {
|
|
logSh(`✅ TECHNICAL LAYER: Aucune amélioration nécessaire`, 'INFO');
|
|
return {
|
|
content,
|
|
stats: {
|
|
processed: Object.keys(content).length,
|
|
enhanced: 0,
|
|
analysisSkipped: true,
|
|
duration: Date.now() - startTime
|
|
}
|
|
};
|
|
}
|
|
|
|
// 2. Améliorer les éléments sélectionnés
|
|
const improvedResults = await this.enhanceTechnicalElements(
|
|
analysis.candidates,
|
|
csvData,
|
|
{ llmProvider, intensity, preserveStructure }
|
|
);
|
|
|
|
// 3. Merger avec contenu original
|
|
enhancedContent = { ...content };
|
|
Object.keys(improvedResults).forEach(tag => {
|
|
if (improvedResults[tag] !== content[tag]) {
|
|
enhancedContent[tag] = improvedResults[tag];
|
|
elementsEnhanced++;
|
|
}
|
|
});
|
|
|
|
elementsProcessed = analysis.candidates.length;
|
|
|
|
} else {
|
|
// Mode direct : améliorer tous les éléments
|
|
enhancedContent = await this.enhanceAllElementsDirect(
|
|
content,
|
|
csvData,
|
|
{ llmProvider, intensity, preserveStructure }
|
|
);
|
|
|
|
elementsProcessed = Object.keys(content).length;
|
|
elementsEnhanced = this.countDifferences(content, enhancedContent);
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: elementsProcessed,
|
|
enhanced: elementsEnhanced,
|
|
total: Object.keys(content).length,
|
|
enhancementRate: (elementsEnhanced / Math.max(elementsProcessed, 1)) * 100,
|
|
duration,
|
|
llmProvider,
|
|
intensity
|
|
};
|
|
|
|
logSh(`✅ TECHNICAL LAYER TERMINÉE: ${elementsEnhanced}/${elementsProcessed} améliorés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Technical layer appliquée', stats);
|
|
|
|
return { content: enhancedContent, stats };
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ TECHNICAL LAYER ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}, { content: Object.keys(content), config });
|
|
}
|
|
|
|
/**
|
|
* ANALYSER BESOINS TECHNIQUES
|
|
*/
|
|
async analyzeTechnicalNeeds(content, csvData, targetTerms = null) {
|
|
logSh(`🔍 Analyse besoins techniques`, 'DEBUG');
|
|
|
|
const analysis = {
|
|
candidates: [],
|
|
technicalTermsFound: [],
|
|
missingTerms: [],
|
|
globalScore: 0
|
|
};
|
|
|
|
// Définir termes techniques selon contexte
|
|
const contextualTerms = this.getContextualTechnicalTerms(csvData?.mc0, targetTerms);
|
|
|
|
// Analyser chaque élément
|
|
Object.entries(content).forEach(([tag, text]) => {
|
|
const elementAnalysis = this.analyzeTechnicalElement(text, contextualTerms, csvData);
|
|
|
|
if (elementAnalysis.needsImprovement) {
|
|
analysis.candidates.push({
|
|
tag,
|
|
content: text,
|
|
technicalTerms: elementAnalysis.foundTerms,
|
|
missingTerms: elementAnalysis.missingTerms,
|
|
score: elementAnalysis.score,
|
|
improvements: elementAnalysis.improvements
|
|
});
|
|
|
|
analysis.globalScore += elementAnalysis.score;
|
|
}
|
|
|
|
analysis.technicalTermsFound.push(...elementAnalysis.foundTerms);
|
|
});
|
|
|
|
analysis.globalScore = analysis.globalScore / Math.max(Object.keys(content).length, 1);
|
|
analysis.technicalTermsFound = [...new Set(analysis.technicalTermsFound)];
|
|
|
|
logSh(` 📊 Score global technique: ${analysis.globalScore.toFixed(2)}`, 'DEBUG');
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* AMÉLIORER ÉLÉMENTS TECHNIQUES SÉLECTIONNÉS
|
|
*/
|
|
async enhanceTechnicalElements(candidates, csvData, config) {
|
|
logSh(`🛠️ Amélioration ${candidates.length} éléments techniques`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(candidates, 4); // Chunks de 4 pour GPT-4
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(` 📦 Chunk technique ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
const enhancementPrompt = this.createTechnicalEnhancementPrompt(chunk, csvData, config);
|
|
|
|
const response = await callLLM(config.llmProvider, enhancementPrompt, {
|
|
temperature: 0.4, // Précision technique
|
|
maxTokens: 3000
|
|
}, csvData?.personality);
|
|
|
|
const chunkResults = this.parseTechnicalResponse(response, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk technique ${chunkIndex + 1}: ${Object.keys(chunkResults).length} améliorés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk technique ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: conserver contenu original
|
|
chunk.forEach(element => {
|
|
results[element.tag] = element.content;
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* AMÉLIORER TOUS ÉLÉMENTS MODE DIRECT
|
|
*/
|
|
async enhanceAllElementsDirect(content, csvData, config) {
|
|
const allElements = Object.entries(content).map(([tag, text]) => ({
|
|
tag,
|
|
content: text,
|
|
technicalTerms: [],
|
|
improvements: ['amélioration_générale_technique']
|
|
}));
|
|
|
|
return await this.enhanceTechnicalElements(allElements, csvData, config);
|
|
}
|
|
|
|
// ============= HELPER METHODS =============
|
|
|
|
/**
|
|
* Analyser élément technique individuel
|
|
*/
|
|
analyzeTechnicalElement(text, contextualTerms, csvData) {
|
|
let score = 0;
|
|
const foundTerms = [];
|
|
const missingTerms = [];
|
|
const improvements = [];
|
|
|
|
// 1. Détecter termes techniques présents
|
|
contextualTerms.forEach(term => {
|
|
if (text.toLowerCase().includes(term.toLowerCase())) {
|
|
foundTerms.push(term);
|
|
} else if (text.length > 100) { // Seulement pour textes longs
|
|
missingTerms.push(term);
|
|
}
|
|
});
|
|
|
|
// 2. Évaluer manque de précision technique
|
|
if (foundTerms.length === 0 && text.length > 80) {
|
|
score += 0.4;
|
|
improvements.push('ajout_termes_techniques');
|
|
}
|
|
|
|
// 3. Détecter vocabulaire trop générique
|
|
const genericWords = ['produit', 'solution', 'service', 'offre', 'article'];
|
|
const genericCount = genericWords.filter(word =>
|
|
text.toLowerCase().includes(word)
|
|
).length;
|
|
|
|
if (genericCount > 1) {
|
|
score += 0.3;
|
|
improvements.push('spécialisation_vocabulaire');
|
|
}
|
|
|
|
// 4. Manque de données techniques (dimensions, etc.)
|
|
if (text.length > 50 && !(/\d+\s*(mm|cm|m|%|°|kg|g)/.test(text))) {
|
|
score += 0.2;
|
|
improvements.push('ajout_données_techniques');
|
|
}
|
|
|
|
// 5. Contexte métier spécifique
|
|
if (csvData?.mc0 && !text.toLowerCase().includes(csvData.mc0.toLowerCase().split(' ')[0])) {
|
|
score += 0.1;
|
|
improvements.push('intégration_contexte_métier');
|
|
}
|
|
|
|
return {
|
|
needsImprovement: score > 0.3,
|
|
score,
|
|
foundTerms,
|
|
missingTerms: missingTerms.slice(0, 3), // Limiter à 3 termes manquants
|
|
improvements
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Obtenir termes techniques contextuels
|
|
*/
|
|
getContextualTechnicalTerms(mc0, targetTerms) {
|
|
// Termes de base signalétique
|
|
const baseTerms = [
|
|
'dibond', 'aluminium', 'PMMA', 'acrylique', 'plexiglas',
|
|
'impression', 'gravure', 'découpe', 'fraisage', 'perçage',
|
|
'adhésif', 'fixation', 'visserie', 'support'
|
|
];
|
|
|
|
// Termes spécifiques selon contexte
|
|
const contextualTerms = [];
|
|
|
|
if (mc0) {
|
|
const mc0Lower = mc0.toLowerCase();
|
|
|
|
if (mc0Lower.includes('plaque')) {
|
|
contextualTerms.push('épaisseur 3mm', 'format standard', 'finition brossée', 'anodisation');
|
|
}
|
|
|
|
if (mc0Lower.includes('signalétique')) {
|
|
contextualTerms.push('norme ISO', 'pictogramme', 'contraste visuel', 'lisibilité');
|
|
}
|
|
|
|
if (mc0Lower.includes('personnalisée')) {
|
|
contextualTerms.push('découpe forme', 'impression numérique', 'quadrichromie', 'pantone');
|
|
}
|
|
}
|
|
|
|
// Ajouter termes ciblés si fournis
|
|
if (targetTerms && Array.isArray(targetTerms)) {
|
|
contextualTerms.push(...targetTerms);
|
|
}
|
|
|
|
return [...baseTerms, ...contextualTerms];
|
|
}
|
|
|
|
/**
|
|
* Créer prompt amélioration technique
|
|
*/
|
|
createTechnicalEnhancementPrompt(chunk, csvData, config) {
|
|
const personality = csvData?.personality;
|
|
|
|
let prompt = `MISSION: Améliore UNIQUEMENT la précision technique de ces contenus.
|
|
|
|
CONTEXTE: ${csvData?.mc0 || 'Signalétique personnalisée'} - Secteur: impression/signalétique
|
|
${personality ? `PERSONNALITÉ: ${personality.nom} (${personality.style})` : ''}
|
|
INTENSITÉ: ${config.intensity} (0.5=léger, 1.0=standard, 1.5=intensif)
|
|
|
|
ÉLÉMENTS À AMÉLIORER TECHNIQUEMENT:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
CONTENU: "${item.content}"
|
|
AMÉLIORATIONS: ${item.improvements.join(', ')}
|
|
${item.missingTerms.length > 0 ? `TERMES À INTÉGRER: ${item.missingTerms.join(', ')}` : ''}`).join('\n\n')}
|
|
|
|
CONSIGNES TECHNIQUES:
|
|
- GARDE exactement le même message et ton${personality ? ` ${personality.style}` : ''}
|
|
- AJOUTE précision technique naturelle et vocabulaire spécialisé
|
|
- INTÈGRE termes métier : matériaux, procédés, normes, dimensions
|
|
- REMPLACE vocabulaire générique par termes techniques appropriés
|
|
- ÉVITE jargon incompréhensible, reste accessible
|
|
- PRESERVE longueur approximative (±15%)
|
|
|
|
VOCABULAIRE TECHNIQUE RECOMMANDÉ:
|
|
- Matériaux: dibond, aluminium anodisé, PMMA coulé, PVC expansé
|
|
- Procédés: impression UV, gravure laser, découpe numérique, fraisage CNC
|
|
- Finitions: brossé, poli, texturé, laqué
|
|
- Fixations: perçage, adhésif double face, vis inox, plots de fixation
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu avec amélioration technique précise
|
|
[2] Contenu avec amélioration technique précise
|
|
etc...
|
|
|
|
IMPORTANT: Réponse DIRECTE par les contenus améliorés, pas d'explication.`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse technique
|
|
*/
|
|
parseTechnicalResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let technicalContent = match[2].trim();
|
|
const element = chunk[index];
|
|
|
|
// Nettoyer contenu technique
|
|
technicalContent = this.cleanTechnicalContent(technicalContent);
|
|
|
|
if (technicalContent && technicalContent.length > 10) {
|
|
results[element.tag] = technicalContent;
|
|
logSh(`✅ Amélioré technique [${element.tag}]: "${technicalContent.substring(0, 60)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content; // Fallback
|
|
logSh(`⚠️ Fallback technique [${element.tag}]: amélioration invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < chunk.length) {
|
|
const element = chunk[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Nettoyer contenu technique généré
|
|
*/
|
|
cleanTechnicalContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(voici\s+)?le\s+contenu\s+amélioré\s*[:.]?\s*/gi, '');
|
|
content = content.replace(/^(avec\s+)?amélioration\s+technique\s*[:.]?\s*/gi, '');
|
|
content = content.replace(/^(bon,?\s*)?(alors,?\s*)?pour\s+/gi, '');
|
|
|
|
// Nettoyer formatage
|
|
content = content.replace(/\*\*[^*]+\*\*/g, ''); // Gras markdown
|
|
content = content.replace(/\s{2,}/g, ' '); // Espaces multiples
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
/**
|
|
* Compter différences entre contenus
|
|
*/
|
|
countDifferences(original, enhanced) {
|
|
let count = 0;
|
|
|
|
Object.keys(original).forEach(tag => {
|
|
if (enhanced[tag] && enhanced[tag] !== original[tag]) {
|
|
count++;
|
|
}
|
|
});
|
|
|
|
return count;
|
|
}
|
|
}
|
|
|
|
module.exports = { TechnicalLayer };
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/selective-enhancement/TransitionLayer.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// TRANSITION LAYER - COUCHE TRANSITIONS MODULAIRE
|
|
// Responsabilité: Amélioration fluidité modulaire réutilisable
|
|
// LLM: Gemini (fluidité linguistique optimale)
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { chunkArray, sleep } = require('./SelectiveUtils');
|
|
|
|
/**
|
|
* COUCHE TRANSITIONS MODULAIRE
|
|
*/
|
|
class TransitionLayer {
|
|
constructor() {
|
|
this.name = 'TransitionEnhancement';
|
|
this.defaultLLM = 'gemini';
|
|
this.priority = 2; // Priorité moyenne - appliqué après technique
|
|
}
|
|
|
|
/**
|
|
* MAIN METHOD - Appliquer amélioration transitions
|
|
*/
|
|
async apply(content, config = {}) {
|
|
return await tracer.run('TransitionLayer.apply()', async () => {
|
|
const {
|
|
llmProvider = this.defaultLLM,
|
|
intensity = 1.0, // 0.0-2.0 intensité d'amélioration
|
|
analysisMode = true, // Analyser avant d'appliquer
|
|
csvData = null,
|
|
preserveStructure = true,
|
|
targetIssues = null // Issues spécifiques à corriger
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
transitionLayer: true,
|
|
llmProvider,
|
|
intensity,
|
|
elementsCount: Object.keys(content).length,
|
|
mc0: csvData?.mc0
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🔗 TRANSITION LAYER: Amélioration fluidité (${llmProvider})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments | Intensité: ${intensity}`, 'INFO');
|
|
|
|
try {
|
|
let enhancedContent = {};
|
|
let elementsProcessed = 0;
|
|
let elementsEnhanced = 0;
|
|
|
|
if (analysisMode) {
|
|
// 1. Analyser éléments nécessitant amélioration transitions
|
|
const analysis = await this.analyzeTransitionNeeds(content, csvData, targetIssues);
|
|
|
|
logSh(` 📋 Analyse: ${analysis.candidates.length}/${Object.keys(content).length} éléments candidats`, 'DEBUG');
|
|
|
|
if (analysis.candidates.length === 0) {
|
|
logSh(`✅ TRANSITION LAYER: Fluidité déjà optimale`, 'INFO');
|
|
return {
|
|
content,
|
|
stats: {
|
|
processed: Object.keys(content).length,
|
|
enhanced: 0,
|
|
analysisSkipped: true,
|
|
duration: Date.now() - startTime
|
|
}
|
|
};
|
|
}
|
|
|
|
// 2. Améliorer les éléments sélectionnés
|
|
const improvedResults = await this.enhanceTransitionElements(
|
|
analysis.candidates,
|
|
csvData,
|
|
{ llmProvider, intensity, preserveStructure }
|
|
);
|
|
|
|
// 3. Merger avec contenu original
|
|
enhancedContent = { ...content };
|
|
Object.keys(improvedResults).forEach(tag => {
|
|
if (improvedResults[tag] !== content[tag]) {
|
|
enhancedContent[tag] = improvedResults[tag];
|
|
elementsEnhanced++;
|
|
}
|
|
});
|
|
|
|
elementsProcessed = analysis.candidates.length;
|
|
|
|
} else {
|
|
// Mode direct : améliorer tous les éléments longs
|
|
const longElements = Object.entries(content)
|
|
.filter(([tag, text]) => text.length > 150)
|
|
.map(([tag, text]) => ({ tag, content: text, issues: ['amélioration_générale'] }));
|
|
|
|
if (longElements.length === 0) {
|
|
return { content, stats: { processed: 0, enhanced: 0, duration: Date.now() - startTime } };
|
|
}
|
|
|
|
const improvedResults = await this.enhanceTransitionElements(
|
|
longElements,
|
|
csvData,
|
|
{ llmProvider, intensity, preserveStructure }
|
|
);
|
|
|
|
enhancedContent = { ...content };
|
|
Object.keys(improvedResults).forEach(tag => {
|
|
if (improvedResults[tag] !== content[tag]) {
|
|
enhancedContent[tag] = improvedResults[tag];
|
|
elementsEnhanced++;
|
|
}
|
|
});
|
|
|
|
elementsProcessed = longElements.length;
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: elementsProcessed,
|
|
enhanced: elementsEnhanced,
|
|
total: Object.keys(content).length,
|
|
enhancementRate: (elementsEnhanced / Math.max(elementsProcessed, 1)) * 100,
|
|
duration,
|
|
llmProvider,
|
|
intensity
|
|
};
|
|
|
|
logSh(`✅ TRANSITION LAYER TERMINÉE: ${elementsEnhanced}/${elementsProcessed} fluidifiés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Transition layer appliquée', stats);
|
|
|
|
return { content: enhancedContent, stats };
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ TRANSITION LAYER ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback gracieux : retourner contenu original
|
|
logSh(`🔄 Fallback: contenu original préservé`, 'WARNING');
|
|
return {
|
|
content,
|
|
stats: { fallback: true, duration },
|
|
error: error.message
|
|
};
|
|
}
|
|
}, { content: Object.keys(content), config });
|
|
}
|
|
|
|
/**
|
|
* ANALYSER BESOINS TRANSITIONS
|
|
*/
|
|
async analyzeTransitionNeeds(content, csvData, targetIssues = null) {
|
|
logSh(`🔍 Analyse besoins transitions`, 'DEBUG');
|
|
|
|
const analysis = {
|
|
candidates: [],
|
|
globalScore: 0,
|
|
issuesFound: {
|
|
repetitiveConnectors: 0,
|
|
abruptTransitions: 0,
|
|
uniformSentences: 0,
|
|
formalityImbalance: 0
|
|
}
|
|
};
|
|
|
|
// Analyser chaque élément
|
|
Object.entries(content).forEach(([tag, text]) => {
|
|
const elementAnalysis = this.analyzeTransitionElement(text, csvData);
|
|
|
|
if (elementAnalysis.needsImprovement) {
|
|
analysis.candidates.push({
|
|
tag,
|
|
content: text,
|
|
issues: elementAnalysis.issues,
|
|
score: elementAnalysis.score,
|
|
improvements: elementAnalysis.improvements
|
|
});
|
|
|
|
analysis.globalScore += elementAnalysis.score;
|
|
|
|
// Compter types d'issues
|
|
elementAnalysis.issues.forEach(issue => {
|
|
if (analysis.issuesFound.hasOwnProperty(issue)) {
|
|
analysis.issuesFound[issue]++;
|
|
}
|
|
});
|
|
}
|
|
});
|
|
|
|
analysis.globalScore = analysis.globalScore / Math.max(Object.keys(content).length, 1);
|
|
|
|
logSh(` 📊 Score global transitions: ${analysis.globalScore.toFixed(2)}`, 'DEBUG');
|
|
logSh(` 🔍 Issues trouvées: ${JSON.stringify(analysis.issuesFound)}`, 'DEBUG');
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* AMÉLIORER ÉLÉMENTS TRANSITIONS SÉLECTIONNÉS
|
|
*/
|
|
async enhanceTransitionElements(candidates, csvData, config) {
|
|
logSh(`🔄 Amélioration ${candidates.length} éléments transitions`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(candidates, 6); // Chunks plus petits pour Gemini
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(` 📦 Chunk transitions ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
const enhancementPrompt = this.createTransitionEnhancementPrompt(chunk, csvData, config);
|
|
|
|
const response = await callLLM(config.llmProvider, enhancementPrompt, {
|
|
temperature: 0.6, // Créativité modérée pour fluidité
|
|
maxTokens: 2500
|
|
}, csvData?.personality);
|
|
|
|
const chunkResults = this.parseTransitionResponse(response, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk transitions ${chunkIndex + 1}: ${Object.keys(chunkResults).length} fluidifiés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk transitions ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: conserver contenu original
|
|
chunk.forEach(element => {
|
|
results[element.tag] = element.content;
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER METHODS =============
|
|
|
|
/**
|
|
* Analyser élément transition individuel
|
|
*/
|
|
analyzeTransitionElement(text, csvData) {
|
|
const sentences = text.split(/[.!?]+/).filter(s => s.trim().length > 10);
|
|
|
|
if (sentences.length < 2) {
|
|
return { needsImprovement: false, score: 0, issues: [], improvements: [] };
|
|
}
|
|
|
|
let score = 0;
|
|
const issues = [];
|
|
const improvements = [];
|
|
|
|
// 1. Analyser connecteurs répétitifs
|
|
const repetitiveScore = this.analyzeRepetitiveConnectors(text);
|
|
if (repetitiveScore > 0.3) {
|
|
score += 0.3;
|
|
issues.push('repetitiveConnectors');
|
|
improvements.push('varier_connecteurs');
|
|
}
|
|
|
|
// 2. Analyser transitions abruptes
|
|
const abruptScore = this.analyzeAbruptTransitions(sentences);
|
|
if (abruptScore > 0.4) {
|
|
score += 0.4;
|
|
issues.push('abruptTransitions');
|
|
improvements.push('ajouter_transitions_fluides');
|
|
}
|
|
|
|
// 3. Analyser uniformité des phrases
|
|
const uniformityScore = this.analyzeSentenceUniformity(sentences);
|
|
if (uniformityScore < 0.3) {
|
|
score += 0.2;
|
|
issues.push('uniformSentences');
|
|
improvements.push('varier_longueurs_phrases');
|
|
}
|
|
|
|
// 4. Analyser équilibre formalité
|
|
const formalityScore = this.analyzeFormalityBalance(text);
|
|
if (formalityScore > 0.5) {
|
|
score += 0.1;
|
|
issues.push('formalityImbalance');
|
|
improvements.push('équilibrer_registre_langue');
|
|
}
|
|
|
|
return {
|
|
needsImprovement: score > 0.3,
|
|
score,
|
|
issues,
|
|
improvements
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Analyser connecteurs répétitifs
|
|
*/
|
|
analyzeRepetitiveConnectors(text) {
|
|
const commonConnectors = ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc'];
|
|
let totalConnectors = 0;
|
|
let repetitions = 0;
|
|
|
|
commonConnectors.forEach(connector => {
|
|
const matches = (text.match(new RegExp(`\\b${connector}\\b`, 'gi')) || []);
|
|
totalConnectors += matches.length;
|
|
if (matches.length > 1) repetitions += matches.length - 1;
|
|
});
|
|
|
|
return totalConnectors > 0 ? repetitions / totalConnectors : 0;
|
|
}
|
|
|
|
/**
|
|
* Analyser transitions abruptes
|
|
*/
|
|
analyzeAbruptTransitions(sentences) {
|
|
if (sentences.length < 2) return 0;
|
|
|
|
let abruptCount = 0;
|
|
|
|
for (let i = 1; i < sentences.length; i++) {
|
|
const current = sentences[i].trim().toLowerCase();
|
|
const hasConnector = this.hasTransitionWord(current);
|
|
|
|
if (!hasConnector && current.length > 30) {
|
|
abruptCount++;
|
|
}
|
|
}
|
|
|
|
return abruptCount / (sentences.length - 1);
|
|
}
|
|
|
|
/**
|
|
* Analyser uniformité des phrases
|
|
*/
|
|
analyzeSentenceUniformity(sentences) {
|
|
if (sentences.length < 2) return 1;
|
|
|
|
const lengths = sentences.map(s => s.trim().length);
|
|
const avgLength = lengths.reduce((a, b) => a + b, 0) / lengths.length;
|
|
const variance = lengths.reduce((acc, len) => acc + Math.pow(len - avgLength, 2), 0) / lengths.length;
|
|
const stdDev = Math.sqrt(variance);
|
|
|
|
return Math.min(1, stdDev / avgLength);
|
|
}
|
|
|
|
/**
|
|
* Analyser équilibre formalité
|
|
*/
|
|
analyzeFormalityBalance(text) {
|
|
const formalIndicators = ['il convient de', 'par conséquent', 'néanmoins', 'toutefois', 'cependant'];
|
|
const casualIndicators = ['du coup', 'bon', 'franchement', 'nickel', 'sympa'];
|
|
|
|
let formalCount = 0;
|
|
let casualCount = 0;
|
|
|
|
formalIndicators.forEach(indicator => {
|
|
if (text.toLowerCase().includes(indicator)) formalCount++;
|
|
});
|
|
|
|
casualIndicators.forEach(indicator => {
|
|
if (text.toLowerCase().includes(indicator)) casualCount++;
|
|
});
|
|
|
|
const total = formalCount + casualCount;
|
|
if (total === 0) return 0;
|
|
|
|
// Déséquilibre si trop d'un côté
|
|
return Math.abs(formalCount - casualCount) / total;
|
|
}
|
|
|
|
/**
|
|
* Vérifier présence mots de transition
|
|
*/
|
|
hasTransitionWord(sentence) {
|
|
const transitionWords = [
|
|
'par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc',
|
|
'ensuite', 'puis', 'également', 'aussi', 'néanmoins', 'toutefois',
|
|
'd\'ailleurs', 'en outre', 'par contre', 'en revanche'
|
|
];
|
|
|
|
return transitionWords.some(word => sentence.includes(word));
|
|
}
|
|
|
|
/**
|
|
* Créer prompt amélioration transitions
|
|
*/
|
|
createTransitionEnhancementPrompt(chunk, csvData, config) {
|
|
const personality = csvData?.personality;
|
|
|
|
let prompt = `MISSION: Améliore UNIQUEMENT les transitions et fluidité de ces contenus.
|
|
|
|
CONTEXTE: Article SEO ${csvData?.mc0 || 'signalétique personnalisée'}
|
|
${personality ? `PERSONNALITÉ: ${personality.nom} (${personality.style} web professionnel)` : ''}
|
|
${personality?.connecteursPref ? `CONNECTEURS PRÉFÉRÉS: ${personality.connecteursPref}` : ''}
|
|
INTENSITÉ: ${config.intensity} (0.5=léger, 1.0=standard, 1.5=intensif)
|
|
|
|
CONTENUS À FLUIDIFIER:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
PROBLÈMES: ${item.issues.join(', ')}
|
|
CONTENU: "${item.content}"`).join('\n\n')}
|
|
|
|
OBJECTIFS FLUIDITÉ:
|
|
- Connecteurs plus naturels et variés${personality?.connecteursPref ? `: ${personality.connecteursPref}` : ''}
|
|
- Transitions fluides entre idées et paragraphes
|
|
- Variation naturelle longueurs phrases
|
|
- ÉVITE répétitions excessives ("du coup", "par ailleurs", "en effet")
|
|
- Style ${personality?.style || 'professionnel'} mais naturel web
|
|
|
|
CONSIGNES STRICTES:
|
|
- NE CHANGE PAS le fond du message ni les informations
|
|
- GARDE même structure générale et longueur approximative (±20%)
|
|
- Améliore SEULEMENT la fluidité et les enchaînements
|
|
- RESPECTE le style ${personality?.nom || 'professionnel'}${personality?.style ? ` (${personality.style})` : ''}
|
|
- ÉVITE sur-correction qui rendrait artificiel
|
|
|
|
TECHNIQUES FLUIDITÉ:
|
|
- Varier connecteurs logiques sans répétition
|
|
- Alterner phrases courtes (8-12 mots) et moyennes (15-20 mots)
|
|
- Utiliser pronoms et reprises pour cohésion
|
|
- Ajouter transitions implicites par reformulation
|
|
- Équilibrer registre soutenu/accessible
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu avec transitions améliorées
|
|
[2] Contenu avec transitions améliorées
|
|
etc...
|
|
|
|
IMPORTANT: Réponse DIRECTE par les contenus fluidifiés, pas d'explication.`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse transitions
|
|
*/
|
|
parseTransitionResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let fluidContent = match[2].trim();
|
|
const element = chunk[index];
|
|
|
|
// Nettoyer contenu fluidifié
|
|
fluidContent = this.cleanTransitionContent(fluidContent);
|
|
|
|
if (fluidContent && fluidContent.length > 10) {
|
|
results[element.tag] = fluidContent;
|
|
logSh(`✅ Fluidifié [${element.tag}]: "${fluidContent.substring(0, 60)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content; // Fallback
|
|
logSh(`⚠️ Fallback transitions [${element.tag}]: amélioration invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < chunk.length) {
|
|
const element = chunk[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Nettoyer contenu transitions généré
|
|
*/
|
|
cleanTransitionContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(voici\s+)?le\s+contenu\s+(fluidifié|amélioré)\s*[:.]?\s*/gi, '');
|
|
content = content.replace(/^(avec\s+)?transitions\s+améliorées\s*[:.]?\s*/gi, '');
|
|
content = content.replace(/^(bon,?\s*)?(alors,?\s*)?/, '');
|
|
|
|
// Nettoyer formatage
|
|
content = content.replace(/\*\*[^*]+\*\*/g, ''); // Gras markdown
|
|
content = content.replace(/\s{2,}/g, ' '); // Espaces multiples
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
}
|
|
|
|
module.exports = { TransitionLayer };
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/selective-enhancement/StyleLayer.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// STYLE LAYER - COUCHE STYLE MODULAIRE
|
|
// Responsabilité: Adaptation personnalité modulaire réutilisable
|
|
// LLM: Mistral (excellence style et personnalité)
|
|
// ========================================
|
|
|
|
const { callLLM } = require('../LLMManager');
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { chunkArray, sleep } = require('./SelectiveUtils');
|
|
|
|
/**
|
|
* COUCHE STYLE MODULAIRE
|
|
*/
|
|
class StyleLayer {
|
|
constructor() {
|
|
this.name = 'StyleEnhancement';
|
|
this.defaultLLM = 'mistral';
|
|
this.priority = 3; // Priorité basse - appliqué en dernier
|
|
}
|
|
|
|
/**
|
|
* MAIN METHOD - Appliquer amélioration style
|
|
*/
|
|
async apply(content, config = {}) {
|
|
return await tracer.run('StyleLayer.apply()', async () => {
|
|
const {
|
|
llmProvider = this.defaultLLM,
|
|
intensity = 1.0, // 0.0-2.0 intensité d'amélioration
|
|
analysisMode = true, // Analyser avant d'appliquer
|
|
csvData = null,
|
|
preserveStructure = true,
|
|
targetStyle = null // Style spécifique à appliquer
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
styleLayer: true,
|
|
llmProvider,
|
|
intensity,
|
|
elementsCount: Object.keys(content).length,
|
|
personality: csvData?.personality?.nom
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🎨 STYLE LAYER: Amélioration personnalité (${llmProvider})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments | Style: ${csvData?.personality?.nom || 'standard'}`, 'INFO');
|
|
|
|
try {
|
|
let enhancedContent = {};
|
|
let elementsProcessed = 0;
|
|
let elementsEnhanced = 0;
|
|
|
|
// Vérifier présence personnalité
|
|
if (!csvData?.personality && !targetStyle) {
|
|
logSh(`⚠️ STYLE LAYER: Pas de personnalité définie, style générique appliqué`, 'WARNING');
|
|
}
|
|
|
|
if (analysisMode) {
|
|
// 1. Analyser éléments nécessitant amélioration style
|
|
const analysis = await this.analyzeStyleNeeds(content, csvData, targetStyle);
|
|
|
|
logSh(` 📋 Analyse: ${analysis.candidates.length}/${Object.keys(content).length} éléments candidats`, 'DEBUG');
|
|
|
|
if (analysis.candidates.length === 0) {
|
|
logSh(`✅ STYLE LAYER: Style déjà cohérent`, 'INFO');
|
|
return {
|
|
content,
|
|
stats: {
|
|
processed: Object.keys(content).length,
|
|
enhanced: 0,
|
|
analysisSkipped: true,
|
|
duration: Date.now() - startTime
|
|
}
|
|
};
|
|
}
|
|
|
|
// 2. Améliorer les éléments sélectionnés
|
|
const improvedResults = await this.enhanceStyleElements(
|
|
analysis.candidates,
|
|
csvData,
|
|
{ llmProvider, intensity, preserveStructure, targetStyle }
|
|
);
|
|
|
|
// 3. Merger avec contenu original
|
|
enhancedContent = { ...content };
|
|
Object.keys(improvedResults).forEach(tag => {
|
|
if (improvedResults[tag] !== content[tag]) {
|
|
enhancedContent[tag] = improvedResults[tag];
|
|
elementsEnhanced++;
|
|
}
|
|
});
|
|
|
|
elementsProcessed = analysis.candidates.length;
|
|
|
|
} else {
|
|
// Mode direct : améliorer tous les éléments textuels
|
|
const textualElements = Object.entries(content)
|
|
.filter(([tag, text]) => text.length > 50 && !tag.includes('FAQ_Question'))
|
|
.map(([tag, text]) => ({ tag, content: text, styleIssues: ['adaptation_générale'] }));
|
|
|
|
if (textualElements.length === 0) {
|
|
return { content, stats: { processed: 0, enhanced: 0, duration: Date.now() - startTime } };
|
|
}
|
|
|
|
const improvedResults = await this.enhanceStyleElements(
|
|
textualElements,
|
|
csvData,
|
|
{ llmProvider, intensity, preserveStructure, targetStyle }
|
|
);
|
|
|
|
enhancedContent = { ...content };
|
|
Object.keys(improvedResults).forEach(tag => {
|
|
if (improvedResults[tag] !== content[tag]) {
|
|
enhancedContent[tag] = improvedResults[tag];
|
|
elementsEnhanced++;
|
|
}
|
|
});
|
|
|
|
elementsProcessed = textualElements.length;
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
processed: elementsProcessed,
|
|
enhanced: elementsEnhanced,
|
|
total: Object.keys(content).length,
|
|
enhancementRate: (elementsEnhanced / Math.max(elementsProcessed, 1)) * 100,
|
|
duration,
|
|
llmProvider,
|
|
intensity,
|
|
personalityApplied: csvData?.personality?.nom || targetStyle || 'générique'
|
|
};
|
|
|
|
logSh(`✅ STYLE LAYER TERMINÉE: ${elementsEnhanced}/${elementsProcessed} stylisés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Style layer appliquée', stats);
|
|
|
|
return { content: enhancedContent, stats };
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ STYLE LAYER ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback gracieux : retourner contenu original
|
|
logSh(`🔄 Fallback: style original préservé`, 'WARNING');
|
|
return {
|
|
content,
|
|
stats: { fallback: true, duration },
|
|
error: error.message
|
|
};
|
|
}
|
|
}, { content: Object.keys(content), config });
|
|
}
|
|
|
|
/**
|
|
* ANALYSER BESOINS STYLE
|
|
*/
|
|
async analyzeStyleNeeds(content, csvData, targetStyle = null) {
|
|
logSh(`🎨 Analyse besoins style`, 'DEBUG');
|
|
|
|
const analysis = {
|
|
candidates: [],
|
|
globalScore: 0,
|
|
styleIssues: {
|
|
genericLanguage: 0,
|
|
personalityMismatch: 0,
|
|
inconsistentTone: 0,
|
|
missingVocabulary: 0
|
|
}
|
|
};
|
|
|
|
const personality = csvData?.personality;
|
|
const expectedStyle = targetStyle || personality;
|
|
|
|
// Analyser chaque élément
|
|
Object.entries(content).forEach(([tag, text]) => {
|
|
const elementAnalysis = this.analyzeStyleElement(text, expectedStyle, csvData);
|
|
|
|
if (elementAnalysis.needsImprovement) {
|
|
analysis.candidates.push({
|
|
tag,
|
|
content: text,
|
|
styleIssues: elementAnalysis.issues,
|
|
score: elementAnalysis.score,
|
|
improvements: elementAnalysis.improvements
|
|
});
|
|
|
|
analysis.globalScore += elementAnalysis.score;
|
|
|
|
// Compter types d'issues
|
|
elementAnalysis.issues.forEach(issue => {
|
|
if (analysis.styleIssues.hasOwnProperty(issue)) {
|
|
analysis.styleIssues[issue]++;
|
|
}
|
|
});
|
|
}
|
|
});
|
|
|
|
analysis.globalScore = analysis.globalScore / Math.max(Object.keys(content).length, 1);
|
|
|
|
logSh(` 📊 Score global style: ${analysis.globalScore.toFixed(2)}`, 'DEBUG');
|
|
logSh(` 🎭 Issues style: ${JSON.stringify(analysis.styleIssues)}`, 'DEBUG');
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* AMÉLIORER ÉLÉMENTS STYLE SÉLECTIONNÉS
|
|
*/
|
|
async enhanceStyleElements(candidates, csvData, config) {
|
|
logSh(`🎨 Amélioration ${candidates.length} éléments style`, 'DEBUG');
|
|
|
|
const results = {};
|
|
const chunks = chunkArray(candidates, 5); // Chunks optimisés pour Mistral
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(` 📦 Chunk style ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
const enhancementPrompt = this.createStyleEnhancementPrompt(chunk, csvData, config);
|
|
|
|
const response = await callLLM(config.llmProvider, enhancementPrompt, {
|
|
temperature: 0.8, // Créativité élevée pour style
|
|
maxTokens: 3000
|
|
}, csvData?.personality);
|
|
|
|
const chunkResults = this.parseStyleResponse(response, chunk);
|
|
Object.assign(results, chunkResults);
|
|
|
|
logSh(` ✅ Chunk style ${chunkIndex + 1}: ${Object.keys(chunkResults).length} stylisés`, 'DEBUG');
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1800);
|
|
}
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Chunk style ${chunkIndex + 1} échoué: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: conserver contenu original
|
|
chunk.forEach(element => {
|
|
results[element.tag] = element.content;
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER METHODS =============
|
|
|
|
/**
|
|
* Analyser élément style individuel
|
|
*/
|
|
analyzeStyleElement(text, expectedStyle, csvData) {
|
|
let score = 0;
|
|
const issues = [];
|
|
const improvements = [];
|
|
|
|
// Si pas de style attendu, score faible
|
|
if (!expectedStyle) {
|
|
return { needsImprovement: false, score: 0.1, issues: ['pas_style_défini'], improvements: [] };
|
|
}
|
|
|
|
// 1. Analyser langage générique
|
|
const genericScore = this.analyzeGenericLanguage(text);
|
|
if (genericScore > 0.4) {
|
|
score += 0.3;
|
|
issues.push('genericLanguage');
|
|
improvements.push('personnaliser_vocabulaire');
|
|
}
|
|
|
|
// 2. Analyser adéquation personnalité
|
|
if (expectedStyle.vocabulairePref) {
|
|
const personalityScore = this.analyzePersonalityAlignment(text, expectedStyle);
|
|
if (personalityScore < 0.3) {
|
|
score += 0.4;
|
|
issues.push('personalityMismatch');
|
|
improvements.push('appliquer_style_personnalité');
|
|
}
|
|
}
|
|
|
|
// 3. Analyser cohérence de ton
|
|
const toneScore = this.analyzeToneConsistency(text, expectedStyle);
|
|
if (toneScore > 0.5) {
|
|
score += 0.2;
|
|
issues.push('inconsistentTone');
|
|
improvements.push('unifier_ton');
|
|
}
|
|
|
|
// 4. Analyser vocabulaire spécialisé
|
|
if (expectedStyle.niveauTechnique) {
|
|
const vocabScore = this.analyzeVocabularyLevel(text, expectedStyle);
|
|
if (vocabScore > 0.4) {
|
|
score += 0.1;
|
|
issues.push('missingVocabulary');
|
|
improvements.push('ajuster_niveau_vocabulaire');
|
|
}
|
|
}
|
|
|
|
return {
|
|
needsImprovement: score > 0.3,
|
|
score,
|
|
issues,
|
|
improvements
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Analyser langage générique
|
|
*/
|
|
analyzeGenericLanguage(text) {
|
|
const genericPhrases = [
|
|
'nos solutions', 'notre expertise', 'notre savoir-faire',
|
|
'nous vous proposons', 'nous mettons à votre disposition',
|
|
'qualité optimale', 'service de qualité', 'expertise reconnue'
|
|
];
|
|
|
|
let genericCount = 0;
|
|
genericPhrases.forEach(phrase => {
|
|
if (text.toLowerCase().includes(phrase)) genericCount++;
|
|
});
|
|
|
|
const wordCount = text.split(/\s+/).length;
|
|
return Math.min(1, (genericCount / Math.max(wordCount / 50, 1)));
|
|
}
|
|
|
|
/**
|
|
* Analyser alignement personnalité
|
|
*/
|
|
analyzePersonalityAlignment(text, personality) {
|
|
if (!personality.vocabulairePref) return 1;
|
|
|
|
const preferredWords = personality.vocabulairePref.toLowerCase().split(',');
|
|
const contentLower = text.toLowerCase();
|
|
|
|
let alignmentScore = 0;
|
|
preferredWords.forEach(word => {
|
|
if (word.trim() && contentLower.includes(word.trim())) {
|
|
alignmentScore++;
|
|
}
|
|
});
|
|
|
|
return Math.min(1, alignmentScore / Math.max(preferredWords.length, 1));
|
|
}
|
|
|
|
/**
|
|
* Analyser cohérence de ton
|
|
*/
|
|
analyzeToneConsistency(text, expectedStyle) {
|
|
const sentences = text.split(/[.!?]+/).filter(s => s.trim().length > 10);
|
|
if (sentences.length < 2) return 0;
|
|
|
|
const tones = sentences.map(sentence => this.detectSentenceTone(sentence));
|
|
const expectedTone = this.getExpectedTone(expectedStyle);
|
|
|
|
let inconsistencies = 0;
|
|
tones.forEach(tone => {
|
|
if (tone !== expectedTone && tone !== 'neutral') {
|
|
inconsistencies++;
|
|
}
|
|
});
|
|
|
|
return inconsistencies / tones.length;
|
|
}
|
|
|
|
/**
|
|
* Analyser niveau vocabulaire
|
|
*/
|
|
analyzeVocabularyLevel(text, expectedStyle) {
|
|
const technicalWords = text.match(/\b\w{8,}\b/g) || [];
|
|
const expectedLevel = expectedStyle.niveauTechnique || 'standard';
|
|
|
|
const techRatio = technicalWords.length / text.split(/\s+/).length;
|
|
|
|
switch (expectedLevel) {
|
|
case 'accessible':
|
|
return techRatio > 0.1 ? techRatio : 0; // Trop technique
|
|
case 'expert':
|
|
return techRatio < 0.05 ? 1 - techRatio : 0; // Pas assez technique
|
|
default:
|
|
return techRatio > 0.15 || techRatio < 0.02 ? Math.abs(0.08 - techRatio) : 0;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Détecter ton de phrase
|
|
*/
|
|
detectSentenceTone(sentence) {
|
|
const lowerSentence = sentence.toLowerCase();
|
|
|
|
if (/\b(excellent|remarquable|exceptionnel|parfait)\b/.test(lowerSentence)) return 'enthusiastic';
|
|
if (/\b(il convient|nous recommandons|il est conseillé)\b/.test(lowerSentence)) return 'formal';
|
|
if (/\b(sympa|cool|nickel|top)\b/.test(lowerSentence)) return 'casual';
|
|
if (/\b(technique|précision|spécification)\b/.test(lowerSentence)) return 'technical';
|
|
|
|
return 'neutral';
|
|
}
|
|
|
|
/**
|
|
* Obtenir ton attendu selon personnalité
|
|
*/
|
|
getExpectedTone(personality) {
|
|
if (!personality || !personality.style) return 'neutral';
|
|
|
|
const style = personality.style.toLowerCase();
|
|
|
|
if (style.includes('technique') || style.includes('expert')) return 'technical';
|
|
if (style.includes('commercial') || style.includes('vente')) return 'enthusiastic';
|
|
if (style.includes('décontracté') || style.includes('moderne')) return 'casual';
|
|
if (style.includes('professionnel') || style.includes('formel')) return 'formal';
|
|
|
|
return 'neutral';
|
|
}
|
|
|
|
/**
|
|
* Créer prompt amélioration style
|
|
*/
|
|
createStyleEnhancementPrompt(chunk, csvData, config) {
|
|
const personality = csvData?.personality || config.targetStyle;
|
|
|
|
let prompt = `MISSION: Adapte UNIQUEMENT le style et la personnalité de ces contenus.
|
|
|
|
CONTEXTE: Article SEO ${csvData?.mc0 || 'signalétique personnalisée'}
|
|
${personality ? `PERSONNALITÉ CIBLE: ${personality.nom} (${personality.style})` : 'STYLE: Professionnel standard'}
|
|
${personality?.description ? `DESCRIPTION: ${personality.description}` : ''}
|
|
INTENSITÉ: ${config.intensity} (0.5=léger, 1.0=standard, 1.5=intensif)
|
|
|
|
CONTENUS À STYLISER:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
PROBLÈMES: ${item.styleIssues.join(', ')}
|
|
CONTENU: "${item.content}"`).join('\n\n')}
|
|
|
|
PROFIL PERSONNALITÉ ${personality?.nom || 'Standard'}:
|
|
${personality ? `- Style: ${personality.style}
|
|
- Niveau: ${personality.niveauTechnique || 'standard'}
|
|
- Vocabulaire préféré: ${personality.vocabulairePref || 'professionnel'}
|
|
- Connecteurs: ${personality.connecteursPref || 'variés'}
|
|
${personality.specificites ? `- Spécificités: ${personality.specificites}` : ''}` : '- Style professionnel web standard'}
|
|
|
|
OBJECTIFS STYLE:
|
|
- Appliquer personnalité ${personality?.nom || 'standard'} de façon naturelle
|
|
- Utiliser vocabulaire et expressions caractéristiques
|
|
- Maintenir cohérence de ton sur tout le contenu
|
|
- Adapter niveau technique selon profil (${personality?.niveauTechnique || 'standard'})
|
|
- Style web ${personality?.style || 'professionnel'} mais authentique
|
|
|
|
CONSIGNES STRICTES:
|
|
- NE CHANGE PAS le fond du message ni les informations factuelles
|
|
- GARDE même structure et longueur approximative (±15%)
|
|
- Applique SEULEMENT style et personnalité sur la forme
|
|
- RESPECTE impérativement le niveau ${personality?.niveauTechnique || 'standard'}
|
|
- ÉVITE exagération qui rendrait artificiel
|
|
|
|
TECHNIQUES STYLE:
|
|
${personality?.vocabulairePref ? `- Intégrer naturellement: ${personality.vocabulairePref}` : '- Vocabulaire professionnel équilibré'}
|
|
- Adapter registre de langue selon ${personality?.style || 'professionnel'}
|
|
- Expressions et tournures caractéristiques personnalité
|
|
- Ton cohérent: ${this.getExpectedTone(personality)} mais naturel
|
|
- Connecteurs préférés: ${personality?.connecteursPref || 'variés et naturels'}
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu avec style personnalisé
|
|
[2] Contenu avec style personnalisé
|
|
etc...
|
|
|
|
IMPORTANT: Réponse DIRECTE par les contenus stylisés, pas d'explication.`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse style
|
|
*/
|
|
parseStyleResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let styledContent = match[2].trim();
|
|
const element = chunk[index];
|
|
|
|
// Nettoyer contenu stylisé
|
|
styledContent = this.cleanStyleContent(styledContent);
|
|
|
|
if (styledContent && styledContent.length > 10) {
|
|
results[element.tag] = styledContent;
|
|
logSh(`✅ Stylisé [${element.tag}]: "${styledContent.substring(0, 60)}..."`, 'DEBUG');
|
|
} else {
|
|
results[element.tag] = element.content; // Fallback
|
|
logSh(`⚠️ Fallback style [${element.tag}]: amélioration invalide`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Compléter les manquants
|
|
while (index < chunk.length) {
|
|
const element = chunk[index];
|
|
results[element.tag] = element.content;
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Nettoyer contenu style généré
|
|
*/
|
|
cleanStyleContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer préfixes indésirables
|
|
content = content.replace(/^(voici\s+)?le\s+contenu\s+(stylisé|adapté|personnalisé)\s*[:.]?\s*/gi, '');
|
|
content = content.replace(/^(avec\s+)?style\s+[^:]*\s*[:.]?\s*/gi, '');
|
|
content = content.replace(/^(dans\s+le\s+style\s+de\s+)[^:]*[:.]?\s*/gi, '');
|
|
|
|
// Nettoyer formatage
|
|
content = content.replace(/\*\*[^*]+\*\*/g, ''); // Gras markdown
|
|
content = content.replace(/\s{2,}/g, ' '); // Espaces multiples
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
}
|
|
|
|
module.exports = { StyleLayer };
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/selective-enhancement/SelectiveCore.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// SELECTIVE CORE - MOTEUR MODULAIRE
|
|
// Responsabilité: Moteur selective enhancement réutilisable sur tout contenu
|
|
// Architecture: Couches applicables à la demande
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
|
|
/**
|
|
* MAIN ENTRY POINT - APPLICATION COUCHE SELECTIVE ENHANCEMENT
|
|
* Input: contenu existant + configuration selective
|
|
* Output: contenu avec couche selective appliquée
|
|
*/
|
|
async function applySelectiveLayer(existingContent, config = {}) {
|
|
return await tracer.run('SelectiveCore.applySelectiveLayer()', async () => {
|
|
const {
|
|
layerType = 'technical', // 'technical' | 'transitions' | 'style' | 'all'
|
|
llmProvider = 'auto', // 'claude' | 'gpt4' | 'gemini' | 'mistral' | 'auto'
|
|
analysisMode = true, // Analyser avant d'appliquer
|
|
preserveStructure = true,
|
|
csvData = null,
|
|
context = {}
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
selectiveLayer: true,
|
|
layerType,
|
|
llmProvider,
|
|
analysisMode,
|
|
elementsCount: Object.keys(existingContent).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🔧 APPLICATION COUCHE SELECTIVE: ${layerType} (${llmProvider})`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(existingContent).length} éléments | Mode: ${analysisMode ? 'analysé' : 'direct'}`, 'INFO');
|
|
|
|
try {
|
|
let enhancedContent = {};
|
|
let layerStats = {};
|
|
|
|
// Sélection automatique du LLM si 'auto'
|
|
const selectedLLM = selectOptimalLLM(layerType, llmProvider);
|
|
|
|
// Application selon type de couche
|
|
switch (layerType) {
|
|
case 'technical':
|
|
const technicalResult = await applyTechnicalEnhancement(existingContent, { ...config, llmProvider: selectedLLM });
|
|
enhancedContent = technicalResult.content;
|
|
layerStats = technicalResult.stats;
|
|
break;
|
|
|
|
case 'transitions':
|
|
const transitionResult = await applyTransitionEnhancement(existingContent, { ...config, llmProvider: selectedLLM });
|
|
enhancedContent = transitionResult.content;
|
|
layerStats = transitionResult.stats;
|
|
break;
|
|
|
|
case 'style':
|
|
const styleResult = await applyStyleEnhancement(existingContent, { ...config, llmProvider: selectedLLM });
|
|
enhancedContent = styleResult.content;
|
|
layerStats = styleResult.stats;
|
|
break;
|
|
|
|
case 'all':
|
|
const allResult = await applyAllSelectiveLayers(existingContent, config);
|
|
enhancedContent = allResult.content;
|
|
layerStats = allResult.stats;
|
|
break;
|
|
|
|
default:
|
|
throw new Error(`Type de couche selective inconnue: ${layerType}`);
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const stats = {
|
|
layerType,
|
|
llmProvider: selectedLLM,
|
|
elementsProcessed: Object.keys(existingContent).length,
|
|
elementsEnhanced: countEnhancedElements(existingContent, enhancedContent),
|
|
duration,
|
|
...layerStats
|
|
};
|
|
|
|
logSh(`✅ COUCHE SELECTIVE APPLIQUÉE: ${stats.elementsEnhanced}/${stats.elementsProcessed} améliorés (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Couche selective appliquée', stats);
|
|
|
|
return {
|
|
content: enhancedContent,
|
|
stats,
|
|
original: existingContent,
|
|
config: { ...config, llmProvider: selectedLLM }
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ COUCHE SELECTIVE ÉCHOUÉE après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
// Fallback: retourner contenu original
|
|
logSh(`🔄 Fallback: contenu original conservé`, 'WARNING');
|
|
return {
|
|
content: existingContent,
|
|
stats: { fallback: true, duration },
|
|
original: existingContent,
|
|
config,
|
|
error: error.message
|
|
};
|
|
}
|
|
}, { existingContent: Object.keys(existingContent), config });
|
|
}
|
|
|
|
/**
|
|
* APPLICATION TECHNIQUE MODULAIRE
|
|
*/
|
|
async function applyTechnicalEnhancement(content, config = {}) {
|
|
const { TechnicalLayer } = require('./TechnicalLayer');
|
|
const layer = new TechnicalLayer();
|
|
return await layer.apply(content, config);
|
|
}
|
|
|
|
/**
|
|
* APPLICATION TRANSITIONS MODULAIRE
|
|
*/
|
|
async function applyTransitionEnhancement(content, config = {}) {
|
|
const { TransitionLayer } = require('./TransitionLayer');
|
|
const layer = new TransitionLayer();
|
|
return await layer.apply(content, config);
|
|
}
|
|
|
|
/**
|
|
* APPLICATION STYLE MODULAIRE
|
|
*/
|
|
async function applyStyleEnhancement(content, config = {}) {
|
|
const { StyleLayer } = require('./StyleLayer');
|
|
const layer = new StyleLayer();
|
|
return await layer.apply(content, config);
|
|
}
|
|
|
|
/**
|
|
* APPLICATION TOUTES COUCHES SÉQUENTIELLES
|
|
*/
|
|
async function applyAllSelectiveLayers(content, config = {}) {
|
|
logSh(`🔄 Application séquentielle toutes couches selective`, 'DEBUG');
|
|
|
|
let currentContent = content;
|
|
const allStats = {
|
|
steps: [],
|
|
totalDuration: 0,
|
|
totalEnhancements: 0
|
|
};
|
|
|
|
const steps = [
|
|
{ name: 'technical', llm: 'gpt4' },
|
|
{ name: 'transitions', llm: 'gemini' },
|
|
{ name: 'style', llm: 'mistral' }
|
|
];
|
|
|
|
for (const step of steps) {
|
|
try {
|
|
logSh(` 🔧 Étape: ${step.name} (${step.llm})`, 'DEBUG');
|
|
|
|
const stepResult = await applySelectiveLayer(currentContent, {
|
|
...config,
|
|
layerType: step.name,
|
|
llmProvider: step.llm
|
|
});
|
|
|
|
currentContent = stepResult.content;
|
|
|
|
allStats.steps.push({
|
|
name: step.name,
|
|
llm: step.llm,
|
|
...stepResult.stats
|
|
});
|
|
|
|
allStats.totalDuration += stepResult.stats.duration;
|
|
allStats.totalEnhancements += stepResult.stats.elementsEnhanced;
|
|
|
|
} catch (error) {
|
|
logSh(` ❌ Étape ${step.name} échouée: ${error.message}`, 'ERROR');
|
|
|
|
allStats.steps.push({
|
|
name: step.name,
|
|
llm: step.llm,
|
|
error: error.message,
|
|
duration: 0,
|
|
elementsEnhanced: 0
|
|
});
|
|
}
|
|
}
|
|
|
|
return {
|
|
content: currentContent,
|
|
stats: allStats
|
|
};
|
|
}
|
|
|
|
/**
|
|
* ANALYSE BESOIN D'ENHANCEMENT
|
|
*/
|
|
async function analyzeEnhancementNeeds(content, config = {}) {
|
|
logSh(`🔍 Analyse besoins selective enhancement`, 'DEBUG');
|
|
|
|
const analysis = {
|
|
technical: { needed: false, score: 0, elements: [] },
|
|
transitions: { needed: false, score: 0, elements: [] },
|
|
style: { needed: false, score: 0, elements: [] },
|
|
recommendation: 'none'
|
|
};
|
|
|
|
// Analyser chaque élément
|
|
Object.entries(content).forEach(([tag, text]) => {
|
|
// Analyse technique (termes techniques manquants)
|
|
const technicalNeed = assessTechnicalNeed(text, config.csvData);
|
|
if (technicalNeed.score > 0.3) {
|
|
analysis.technical.needed = true;
|
|
analysis.technical.score += technicalNeed.score;
|
|
analysis.technical.elements.push({ tag, score: technicalNeed.score, reason: technicalNeed.reason });
|
|
}
|
|
|
|
// Analyse transitions (fluidité)
|
|
const transitionNeed = assessTransitionNeed(text);
|
|
if (transitionNeed.score > 0.4) {
|
|
analysis.transitions.needed = true;
|
|
analysis.transitions.score += transitionNeed.score;
|
|
analysis.transitions.elements.push({ tag, score: transitionNeed.score, reason: transitionNeed.reason });
|
|
}
|
|
|
|
// Analyse style (personnalité)
|
|
const styleNeed = assessStyleNeed(text, config.csvData?.personality);
|
|
if (styleNeed.score > 0.3) {
|
|
analysis.style.needed = true;
|
|
analysis.style.score += styleNeed.score;
|
|
analysis.style.elements.push({ tag, score: styleNeed.score, reason: styleNeed.reason });
|
|
}
|
|
});
|
|
|
|
// Normaliser scores
|
|
const elementCount = Object.keys(content).length;
|
|
analysis.technical.score = analysis.technical.score / elementCount;
|
|
analysis.transitions.score = analysis.transitions.score / elementCount;
|
|
analysis.style.score = analysis.style.score / elementCount;
|
|
|
|
// Recommandation
|
|
const scores = [
|
|
{ type: 'technical', score: analysis.technical.score },
|
|
{ type: 'transitions', score: analysis.transitions.score },
|
|
{ type: 'style', score: analysis.style.score }
|
|
].sort((a, b) => b.score - a.score);
|
|
|
|
if (scores[0].score > 0.6) {
|
|
analysis.recommendation = scores[0].type;
|
|
} else if (scores[0].score > 0.4) {
|
|
analysis.recommendation = 'light_' + scores[0].type;
|
|
}
|
|
|
|
logSh(` 📊 Analyse: Tech=${analysis.technical.score.toFixed(2)} | Trans=${analysis.transitions.score.toFixed(2)} | Style=${analysis.style.score.toFixed(2)}`, 'DEBUG');
|
|
logSh(` 💡 Recommandation: ${analysis.recommendation}`, 'DEBUG');
|
|
|
|
return analysis;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
/**
|
|
* Sélectionner LLM optimal selon type de couche
|
|
*/
|
|
function selectOptimalLLM(layerType, llmProvider) {
|
|
if (llmProvider !== 'auto') return llmProvider;
|
|
|
|
const optimalMapping = {
|
|
'technical': 'openai', // OpenAI GPT-4 excellent pour précision technique
|
|
'transitions': 'gemini', // Gemini bon pour fluidité
|
|
'style': 'mistral', // Mistral excellent pour style personnalité
|
|
'all': 'claude' // Claude polyvalent pour tout
|
|
};
|
|
|
|
return optimalMapping[layerType] || 'claude';
|
|
}
|
|
|
|
/**
|
|
* Compter éléments améliorés
|
|
*/
|
|
function countEnhancedElements(original, enhanced) {
|
|
let count = 0;
|
|
|
|
Object.keys(original).forEach(tag => {
|
|
if (enhanced[tag] && enhanced[tag] !== original[tag]) {
|
|
count++;
|
|
}
|
|
});
|
|
|
|
return count;
|
|
}
|
|
|
|
/**
|
|
* Évaluer besoin technique
|
|
*/
|
|
function assessTechnicalNeed(content, csvData) {
|
|
let score = 0;
|
|
let reason = [];
|
|
|
|
// Manque de termes techniques spécifiques
|
|
if (csvData?.mc0) {
|
|
const technicalTerms = ['dibond', 'pmma', 'aluminium', 'fraisage', 'impression', 'gravure', 'découpe'];
|
|
const contentLower = content.toLowerCase();
|
|
const foundTerms = technicalTerms.filter(term => contentLower.includes(term));
|
|
|
|
if (foundTerms.length === 0 && content.length > 100) {
|
|
score += 0.4;
|
|
reason.push('manque_termes_techniques');
|
|
}
|
|
}
|
|
|
|
// Vocabulaire trop générique
|
|
const genericWords = ['produit', 'solution', 'service', 'qualité', 'offre'];
|
|
const genericCount = genericWords.filter(word => content.toLowerCase().includes(word)).length;
|
|
|
|
if (genericCount > 2) {
|
|
score += 0.3;
|
|
reason.push('vocabulaire_générique');
|
|
}
|
|
|
|
// Manque de précision dimensionnelle/technique
|
|
if (content.length > 50 && !(/\d+\s*(mm|cm|m|%|°)/.test(content))) {
|
|
score += 0.2;
|
|
reason.push('manque_précision_technique');
|
|
}
|
|
|
|
return { score: Math.min(1, score), reason: reason.join(',') };
|
|
}
|
|
|
|
/**
|
|
* Évaluer besoin transitions
|
|
*/
|
|
function assessTransitionNeed(content) {
|
|
let score = 0;
|
|
let reason = [];
|
|
|
|
const sentences = content.split(/[.!?]+/).filter(s => s.trim().length > 10);
|
|
|
|
if (sentences.length < 2) return { score: 0, reason: '' };
|
|
|
|
// Connecteurs répétitifs
|
|
const connectors = ['par ailleurs', 'en effet', 'de plus', 'cependant'];
|
|
let repetitiveConnectors = 0;
|
|
|
|
connectors.forEach(connector => {
|
|
const matches = (content.match(new RegExp(connector, 'gi')) || []);
|
|
if (matches.length > 1) repetitiveConnectors++;
|
|
});
|
|
|
|
if (repetitiveConnectors > 1) {
|
|
score += 0.4;
|
|
reason.push('connecteurs_répétitifs');
|
|
}
|
|
|
|
// Transitions abruptes (phrases sans connecteurs logiques)
|
|
let abruptTransitions = 0;
|
|
for (let i = 1; i < sentences.length; i++) {
|
|
const sentence = sentences[i].trim().toLowerCase();
|
|
const hasConnector = connectors.some(conn => sentence.startsWith(conn)) ||
|
|
/^(puis|ensuite|également|aussi|donc|ainsi)/.test(sentence);
|
|
|
|
if (!hasConnector && sentence.length > 30) {
|
|
abruptTransitions++;
|
|
}
|
|
}
|
|
|
|
if (abruptTransitions / sentences.length > 0.6) {
|
|
score += 0.3;
|
|
reason.push('transitions_abruptes');
|
|
}
|
|
|
|
return { score: Math.min(1, score), reason: reason.join(',') };
|
|
}
|
|
|
|
/**
|
|
* Évaluer besoin style
|
|
*/
|
|
function assessStyleNeed(content, personality) {
|
|
let score = 0;
|
|
let reason = [];
|
|
|
|
if (!personality) {
|
|
score += 0.2;
|
|
reason.push('pas_personnalité');
|
|
return { score, reason: reason.join(',') };
|
|
}
|
|
|
|
// Style générique (pas de personnalité visible)
|
|
const personalityWords = (personality.vocabulairePref || '').toLowerCase().split(',');
|
|
const contentLower = content.toLowerCase();
|
|
|
|
const personalityFound = personalityWords.some(word =>
|
|
word.trim() && contentLower.includes(word.trim())
|
|
);
|
|
|
|
if (!personalityFound && content.length > 50) {
|
|
score += 0.4;
|
|
reason.push('style_générique');
|
|
}
|
|
|
|
// Niveau technique inadapté
|
|
if (personality.niveauTechnique === 'accessible' && /\b(optimisation|implémentation|méthodologie)\b/i.test(content)) {
|
|
score += 0.3;
|
|
reason.push('trop_technique');
|
|
}
|
|
|
|
return { score: Math.min(1, score), reason: reason.join(',') };
|
|
}
|
|
|
|
module.exports = {
|
|
applySelectiveLayer, // ← MAIN ENTRY POINT MODULAIRE
|
|
applyTechnicalEnhancement,
|
|
applyTransitionEnhancement,
|
|
applyStyleEnhancement,
|
|
applyAllSelectiveLayers,
|
|
analyzeEnhancementNeeds,
|
|
selectOptimalLLM
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/selective-enhancement/SelectiveLayers.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// SELECTIVE LAYERS - COUCHES COMPOSABLES
|
|
// Responsabilité: Stacks prédéfinis et couches adaptatives pour selective enhancement
|
|
// Architecture: Composable layers avec orchestration intelligente
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
const { tracer } = require('../trace');
|
|
const { applySelectiveLayer } = require('./SelectiveCore');
|
|
|
|
/**
|
|
* STACKS PRÉDÉFINIS SELECTIVE ENHANCEMENT
|
|
*/
|
|
const PREDEFINED_STACKS = {
|
|
// Stack léger - Amélioration technique uniquement
|
|
lightEnhancement: {
|
|
name: 'lightEnhancement',
|
|
description: 'Amélioration technique légère avec OpenAI',
|
|
layers: [
|
|
{ type: 'technical', llm: 'openai', intensity: 0.7 }
|
|
],
|
|
layersCount: 1
|
|
},
|
|
|
|
// Stack standard - Technique + Transitions
|
|
standardEnhancement: {
|
|
name: 'standardEnhancement',
|
|
description: 'Amélioration technique et fluidité (OpenAI + Gemini)',
|
|
layers: [
|
|
{ type: 'technical', llm: 'openai', intensity: 0.9 },
|
|
{ type: 'transitions', llm: 'gemini', intensity: 0.8 }
|
|
],
|
|
layersCount: 2
|
|
},
|
|
|
|
// Stack complet - Toutes couches séquentielles
|
|
fullEnhancement: {
|
|
name: 'fullEnhancement',
|
|
description: 'Enhancement complet multi-LLM (OpenAI + Gemini + Mistral)',
|
|
layers: [
|
|
{ type: 'technical', llm: 'openai', intensity: 1.0 },
|
|
{ type: 'transitions', llm: 'gemini', intensity: 0.9 },
|
|
{ type: 'style', llm: 'mistral', intensity: 0.8 }
|
|
],
|
|
layersCount: 3
|
|
},
|
|
|
|
// Stack personnalité - Style prioritaire
|
|
personalityFocus: {
|
|
name: 'personalityFocus',
|
|
description: 'Focus personnalité et style avec Mistral + technique légère',
|
|
layers: [
|
|
{ type: 'style', llm: 'mistral', intensity: 1.2 },
|
|
{ type: 'technical', llm: 'openai', intensity: 0.6 }
|
|
],
|
|
layersCount: 2
|
|
},
|
|
|
|
// Stack fluidité - Transitions prioritaires
|
|
fluidityFocus: {
|
|
name: 'fluidityFocus',
|
|
description: 'Focus fluidité avec Gemini + enhancements légers',
|
|
layers: [
|
|
{ type: 'transitions', llm: 'gemini', intensity: 1.1 },
|
|
{ type: 'technical', llm: 'openai', intensity: 0.7 },
|
|
{ type: 'style', llm: 'mistral', intensity: 0.6 }
|
|
],
|
|
layersCount: 3
|
|
}
|
|
};
|
|
|
|
/**
|
|
* APPLIQUER STACK PRÉDÉFINI
|
|
*/
|
|
async function applyPredefinedStack(content, stackName, config = {}) {
|
|
return await tracer.run('SelectiveLayers.applyPredefinedStack()', async () => {
|
|
const stack = PREDEFINED_STACKS[stackName];
|
|
|
|
if (!stack) {
|
|
throw new Error(`Stack selective prédéfini inconnu: ${stackName}. Disponibles: ${Object.keys(PREDEFINED_STACKS).join(', ')}`);
|
|
}
|
|
|
|
await tracer.annotate({
|
|
selectivePredefinedStack: true,
|
|
stackName,
|
|
layersCount: stack.layersCount,
|
|
elementsCount: Object.keys(content).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`📦 APPLICATION STACK SELECTIVE: ${stack.name} (${stack.layersCount} couches)`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments | Description: ${stack.description}`, 'INFO');
|
|
|
|
try {
|
|
let currentContent = content;
|
|
const stackStats = {
|
|
stackName,
|
|
layers: [],
|
|
totalModifications: 0,
|
|
totalDuration: 0,
|
|
success: true
|
|
};
|
|
|
|
// Appliquer chaque couche séquentiellement
|
|
for (let i = 0; i < stack.layers.length; i++) {
|
|
const layer = stack.layers[i];
|
|
|
|
try {
|
|
logSh(` 🔧 Couche ${i + 1}/${stack.layersCount}: ${layer.type} (${layer.llm})`, 'DEBUG');
|
|
|
|
const layerResult = await applySelectiveLayer(currentContent, {
|
|
...config,
|
|
layerType: layer.type,
|
|
llmProvider: layer.llm,
|
|
intensity: layer.intensity,
|
|
analysisMode: true
|
|
});
|
|
|
|
currentContent = layerResult.content;
|
|
|
|
stackStats.layers.push({
|
|
order: i + 1,
|
|
type: layer.type,
|
|
llm: layer.llm,
|
|
intensity: layer.intensity,
|
|
elementsEnhanced: layerResult.stats.elementsEnhanced,
|
|
duration: layerResult.stats.duration,
|
|
success: !layerResult.stats.fallback
|
|
});
|
|
|
|
stackStats.totalModifications += layerResult.stats.elementsEnhanced;
|
|
stackStats.totalDuration += layerResult.stats.duration;
|
|
|
|
logSh(` ✅ Couche ${layer.type}: ${layerResult.stats.elementsEnhanced} améliorations`, 'DEBUG');
|
|
|
|
} catch (layerError) {
|
|
logSh(` ❌ Couche ${layer.type} échouée: ${layerError.message}`, 'ERROR');
|
|
|
|
stackStats.layers.push({
|
|
order: i + 1,
|
|
type: layer.type,
|
|
llm: layer.llm,
|
|
error: layerError.message,
|
|
duration: 0,
|
|
success: false
|
|
});
|
|
|
|
// Continuer avec les autres couches
|
|
}
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const successfulLayers = stackStats.layers.filter(l => l.success).length;
|
|
|
|
logSh(`✅ STACK SELECTIVE ${stackName}: ${successfulLayers}/${stack.layersCount} couches | ${stackStats.totalModifications} modifications (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Stack selective appliqué', { ...stackStats, totalDuration: duration });
|
|
|
|
return {
|
|
content: currentContent,
|
|
stats: { ...stackStats, totalDuration: duration },
|
|
original: content,
|
|
stackApplied: stackName
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ STACK SELECTIVE ${stackName} ÉCHOUÉ après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
return {
|
|
content,
|
|
stats: { stackName, error: error.message, duration, success: false },
|
|
original: content,
|
|
fallback: true
|
|
};
|
|
}
|
|
}, { content: Object.keys(content), stackName, config });
|
|
}
|
|
|
|
/**
|
|
* APPLIQUER COUCHES ADAPTATIVES
|
|
*/
|
|
async function applyAdaptiveLayers(content, config = {}) {
|
|
return await tracer.run('SelectiveLayers.applyAdaptiveLayers()', async () => {
|
|
const {
|
|
maxIntensity = 1.0,
|
|
analysisThreshold = 0.4,
|
|
csvData = null
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
selectiveAdaptiveLayers: true,
|
|
maxIntensity,
|
|
analysisThreshold,
|
|
elementsCount: Object.keys(content).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🧠 APPLICATION COUCHES ADAPTATIVES SELECTIVE`, 'INFO');
|
|
logSh(` 📊 ${Object.keys(content).length} éléments | Seuil: ${analysisThreshold}`, 'INFO');
|
|
|
|
try {
|
|
// 1. Analyser besoins de chaque type de couche
|
|
const needsAnalysis = await analyzeSelectiveNeeds(content, csvData);
|
|
|
|
logSh(` 📋 Analyse besoins: Tech=${needsAnalysis.technical.score.toFixed(2)} | Trans=${needsAnalysis.transitions.score.toFixed(2)} | Style=${needsAnalysis.style.score.toFixed(2)}`, 'DEBUG');
|
|
|
|
// 2. Déterminer couches à appliquer selon scores
|
|
const layersToApply = [];
|
|
|
|
if (needsAnalysis.technical.needed && needsAnalysis.technical.score > analysisThreshold) {
|
|
layersToApply.push({
|
|
type: 'technical',
|
|
llm: 'openai',
|
|
intensity: Math.min(maxIntensity, needsAnalysis.technical.score * 1.2),
|
|
priority: 1
|
|
});
|
|
}
|
|
|
|
if (needsAnalysis.transitions.needed && needsAnalysis.transitions.score > analysisThreshold) {
|
|
layersToApply.push({
|
|
type: 'transitions',
|
|
llm: 'gemini',
|
|
intensity: Math.min(maxIntensity, needsAnalysis.transitions.score * 1.1),
|
|
priority: 2
|
|
});
|
|
}
|
|
|
|
if (needsAnalysis.style.needed && needsAnalysis.style.score > analysisThreshold) {
|
|
layersToApply.push({
|
|
type: 'style',
|
|
llm: 'mistral',
|
|
intensity: Math.min(maxIntensity, needsAnalysis.style.score),
|
|
priority: 3
|
|
});
|
|
}
|
|
|
|
if (layersToApply.length === 0) {
|
|
logSh(`✅ COUCHES ADAPTATIVES: Aucune amélioration nécessaire`, 'INFO');
|
|
return {
|
|
content,
|
|
stats: {
|
|
adaptive: true,
|
|
layersApplied: 0,
|
|
analysisOnly: true,
|
|
duration: Date.now() - startTime
|
|
}
|
|
};
|
|
}
|
|
|
|
// 3. Appliquer couches par ordre de priorité
|
|
layersToApply.sort((a, b) => a.priority - b.priority);
|
|
logSh(` 🎯 Couches sélectionnées: ${layersToApply.map(l => `${l.type}(${l.intensity.toFixed(1)})`).join(' → ')}`, 'INFO');
|
|
|
|
let currentContent = content;
|
|
const adaptiveStats = {
|
|
layersAnalyzed: 3,
|
|
layersApplied: layersToApply.length,
|
|
layers: [],
|
|
totalModifications: 0,
|
|
adaptive: true
|
|
};
|
|
|
|
for (const layer of layersToApply) {
|
|
try {
|
|
logSh(` 🔧 Couche adaptative: ${layer.type} (intensité: ${layer.intensity.toFixed(1)})`, 'DEBUG');
|
|
|
|
const layerResult = await applySelectiveLayer(currentContent, {
|
|
...config,
|
|
layerType: layer.type,
|
|
llmProvider: layer.llm,
|
|
intensity: layer.intensity,
|
|
analysisMode: true
|
|
});
|
|
|
|
currentContent = layerResult.content;
|
|
|
|
adaptiveStats.layers.push({
|
|
type: layer.type,
|
|
llm: layer.llm,
|
|
intensity: layer.intensity,
|
|
elementsEnhanced: layerResult.stats.elementsEnhanced,
|
|
duration: layerResult.stats.duration,
|
|
success: !layerResult.stats.fallback
|
|
});
|
|
|
|
adaptiveStats.totalModifications += layerResult.stats.elementsEnhanced;
|
|
|
|
} catch (layerError) {
|
|
logSh(` ❌ Couche adaptative ${layer.type} échouée: ${layerError.message}`, 'ERROR');
|
|
|
|
adaptiveStats.layers.push({
|
|
type: layer.type,
|
|
error: layerError.message,
|
|
success: false
|
|
});
|
|
}
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const successfulLayers = adaptiveStats.layers.filter(l => l.success).length;
|
|
|
|
logSh(`✅ COUCHES ADAPTATIVES: ${successfulLayers}/${layersToApply.length} appliquées | ${adaptiveStats.totalModifications} modifications (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Couches adaptatives appliquées', { ...adaptiveStats, totalDuration: duration });
|
|
|
|
return {
|
|
content: currentContent,
|
|
stats: { ...adaptiveStats, totalDuration: duration },
|
|
original: content
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ COUCHES ADAPTATIVES ÉCHOUÉES après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
return {
|
|
content,
|
|
stats: { adaptive: true, error: error.message, duration },
|
|
original: content,
|
|
fallback: true
|
|
};
|
|
}
|
|
}, { content: Object.keys(content), config });
|
|
}
|
|
|
|
/**
|
|
* PIPELINE COUCHES PERSONNALISÉ
|
|
*/
|
|
async function applyLayerPipeline(content, layerSequence, config = {}) {
|
|
return await tracer.run('SelectiveLayers.applyLayerPipeline()', async () => {
|
|
if (!Array.isArray(layerSequence) || layerSequence.length === 0) {
|
|
throw new Error('Séquence de couches invalide ou vide');
|
|
}
|
|
|
|
await tracer.annotate({
|
|
selectiveLayerPipeline: true,
|
|
pipelineLength: layerSequence.length,
|
|
elementsCount: Object.keys(content).length
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🔄 PIPELINE COUCHES SELECTIVE PERSONNALISÉ: ${layerSequence.length} étapes`, 'INFO');
|
|
|
|
try {
|
|
let currentContent = content;
|
|
const pipelineStats = {
|
|
pipelineLength: layerSequence.length,
|
|
steps: [],
|
|
totalModifications: 0,
|
|
success: true
|
|
};
|
|
|
|
for (let i = 0; i < layerSequence.length; i++) {
|
|
const step = layerSequence[i];
|
|
|
|
try {
|
|
logSh(` 📍 Étape ${i + 1}/${layerSequence.length}: ${step.type} (${step.llm || 'auto'})`, 'DEBUG');
|
|
|
|
const stepResult = await applySelectiveLayer(currentContent, {
|
|
...config,
|
|
...step
|
|
});
|
|
|
|
currentContent = stepResult.content;
|
|
|
|
pipelineStats.steps.push({
|
|
order: i + 1,
|
|
...step,
|
|
elementsEnhanced: stepResult.stats.elementsEnhanced,
|
|
duration: stepResult.stats.duration,
|
|
success: !stepResult.stats.fallback
|
|
});
|
|
|
|
pipelineStats.totalModifications += stepResult.stats.elementsEnhanced;
|
|
|
|
} catch (stepError) {
|
|
logSh(` ❌ Étape ${i + 1} échouée: ${stepError.message}`, 'ERROR');
|
|
|
|
pipelineStats.steps.push({
|
|
order: i + 1,
|
|
...step,
|
|
error: stepError.message,
|
|
success: false
|
|
});
|
|
}
|
|
}
|
|
|
|
const duration = Date.now() - startTime;
|
|
const successfulSteps = pipelineStats.steps.filter(s => s.success).length;
|
|
|
|
logSh(`✅ PIPELINE SELECTIVE: ${successfulSteps}/${layerSequence.length} étapes | ${pipelineStats.totalModifications} modifications (${duration}ms)`, 'INFO');
|
|
|
|
await tracer.event('Pipeline selective appliqué', { ...pipelineStats, totalDuration: duration });
|
|
|
|
return {
|
|
content: currentContent,
|
|
stats: { ...pipelineStats, totalDuration: duration },
|
|
original: content
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ PIPELINE SELECTIVE ÉCHOUÉ après ${duration}ms: ${error.message}`, 'ERROR');
|
|
|
|
return {
|
|
content,
|
|
stats: { error: error.message, duration, success: false },
|
|
original: content,
|
|
fallback: true
|
|
};
|
|
}
|
|
}, { content: Object.keys(content), layerSequence, config });
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
/**
|
|
* Analyser besoins selective enhancement
|
|
*/
|
|
async function analyzeSelectiveNeeds(content, csvData) {
|
|
const analysis = {
|
|
technical: { needed: false, score: 0, elements: [] },
|
|
transitions: { needed: false, score: 0, elements: [] },
|
|
style: { needed: false, score: 0, elements: [] }
|
|
};
|
|
|
|
// Analyser chaque élément pour tous types de besoins
|
|
Object.entries(content).forEach(([tag, text]) => {
|
|
// Analyse technique (import depuis SelectiveCore logic)
|
|
const technicalNeed = assessTechnicalNeed(text, csvData);
|
|
if (technicalNeed.score > 0.3) {
|
|
analysis.technical.needed = true;
|
|
analysis.technical.score += technicalNeed.score;
|
|
analysis.technical.elements.push({ tag, score: technicalNeed.score });
|
|
}
|
|
|
|
// Analyse transitions
|
|
const transitionNeed = assessTransitionNeed(text);
|
|
if (transitionNeed.score > 0.3) {
|
|
analysis.transitions.needed = true;
|
|
analysis.transitions.score += transitionNeed.score;
|
|
analysis.transitions.elements.push({ tag, score: transitionNeed.score });
|
|
}
|
|
|
|
// Analyse style
|
|
const styleNeed = assessStyleNeed(text, csvData?.personality);
|
|
if (styleNeed.score > 0.3) {
|
|
analysis.style.needed = true;
|
|
analysis.style.score += styleNeed.score;
|
|
analysis.style.elements.push({ tag, score: styleNeed.score });
|
|
}
|
|
});
|
|
|
|
// Normaliser scores
|
|
const elementCount = Object.keys(content).length;
|
|
analysis.technical.score = analysis.technical.score / elementCount;
|
|
analysis.transitions.score = analysis.transitions.score / elementCount;
|
|
analysis.style.score = analysis.style.score / elementCount;
|
|
|
|
return analysis;
|
|
}
|
|
|
|
/**
|
|
* Évaluer besoin technique (simplifié de SelectiveCore)
|
|
*/
|
|
function assessTechnicalNeed(content, csvData) {
|
|
let score = 0;
|
|
|
|
// Manque de termes techniques spécifiques
|
|
if (csvData?.mc0) {
|
|
const technicalTerms = ['dibond', 'pmma', 'aluminium', 'fraisage', 'impression', 'gravure'];
|
|
const foundTerms = technicalTerms.filter(term => content.toLowerCase().includes(term));
|
|
|
|
if (foundTerms.length === 0 && content.length > 100) {
|
|
score += 0.4;
|
|
}
|
|
}
|
|
|
|
// Vocabulaire générique
|
|
const genericWords = ['produit', 'solution', 'service', 'qualité'];
|
|
const genericCount = genericWords.filter(word => content.toLowerCase().includes(word)).length;
|
|
|
|
if (genericCount > 2) score += 0.3;
|
|
|
|
return { score: Math.min(1, score) };
|
|
}
|
|
|
|
/**
|
|
* Évaluer besoin transitions (simplifié)
|
|
*/
|
|
function assessTransitionNeed(content) {
|
|
const sentences = content.split(/[.!?]+/).filter(s => s.trim().length > 10);
|
|
if (sentences.length < 2) return { score: 0 };
|
|
|
|
let score = 0;
|
|
|
|
// Connecteurs répétitifs
|
|
const connectors = ['par ailleurs', 'en effet', 'de plus'];
|
|
let repetitions = 0;
|
|
|
|
connectors.forEach(connector => {
|
|
const matches = (content.match(new RegExp(connector, 'gi')) || []);
|
|
if (matches.length > 1) repetitions++;
|
|
});
|
|
|
|
if (repetitions > 1) score += 0.4;
|
|
|
|
return { score: Math.min(1, score) };
|
|
}
|
|
|
|
/**
|
|
* Évaluer besoin style (simplifié)
|
|
*/
|
|
function assessStyleNeed(content, personality) {
|
|
let score = 0;
|
|
|
|
if (!personality) {
|
|
score += 0.2;
|
|
return { score };
|
|
}
|
|
|
|
// Style générique
|
|
const personalityWords = (personality.vocabulairePref || '').toLowerCase().split(',');
|
|
const personalityFound = personalityWords.some(word =>
|
|
word.trim() && content.toLowerCase().includes(word.trim())
|
|
);
|
|
|
|
if (!personalityFound && content.length > 50) score += 0.4;
|
|
|
|
return { score: Math.min(1, score) };
|
|
}
|
|
|
|
/**
|
|
* Obtenir stacks disponibles
|
|
*/
|
|
function getAvailableStacks() {
|
|
return Object.values(PREDEFINED_STACKS);
|
|
}
|
|
|
|
module.exports = {
|
|
// Main functions
|
|
applyPredefinedStack,
|
|
applyAdaptiveLayers,
|
|
applyLayerPipeline,
|
|
|
|
// Utils
|
|
getAvailableStacks,
|
|
analyzeSelectiveNeeds,
|
|
|
|
// Constants
|
|
PREDEFINED_STACKS
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/main_modulaire.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// MAIN MODULAIRE - PIPELINE ARCHITECTURALE MODERNE
|
|
// Responsabilité: Orchestration workflow avec architecture modulaire complète
|
|
// Usage: node main_modulaire.js [rowNumber] [stackType]
|
|
// ========================================
|
|
|
|
const { logSh } = require('./ErrorReporting');
|
|
const { tracer } = require('./trace');
|
|
|
|
// Imports pipeline de base
|
|
const { readInstructionsData, selectPersonalityWithAI, getPersonalities } = require('./BrainConfig');
|
|
const { extractElements, buildSmartHierarchy } = require('./ElementExtraction');
|
|
const { generateMissingKeywords } = require('./MissingKeywords');
|
|
const { generateSimple } = require('./ContentGeneration');
|
|
const { injectGeneratedContent } = require('./ContentAssembly');
|
|
const { compileAndStoreArticle } = require('./ArticleStorage');
|
|
|
|
// Imports modules modulaires
|
|
const { applySelectiveLayer } = require('./selective-enhancement/SelectiveCore');
|
|
const {
|
|
applyPredefinedStack,
|
|
applyAdaptiveLayers,
|
|
getAvailableStacks
|
|
} = require('./selective-enhancement/SelectiveLayers');
|
|
const {
|
|
applyAdversarialLayer
|
|
} = require('./adversarial-generation/AdversarialCore');
|
|
const {
|
|
applyPredefinedStack: applyAdversarialStack
|
|
} = require('./adversarial-generation/AdversarialLayers');
|
|
|
|
/**
|
|
* WORKFLOW MODULAIRE PRINCIPAL
|
|
*/
|
|
async function handleModularWorkflow(config = {}) {
|
|
return await tracer.run('MainModulaire.handleModularWorkflow()', async () => {
|
|
const {
|
|
rowNumber = 2,
|
|
selectiveStack = 'standardEnhancement', // lightEnhancement, standardEnhancement, fullEnhancement, personalityFocus, fluidityFocus, adaptive
|
|
adversarialMode = 'light', // none, light, standard, heavy, adaptive
|
|
source = 'main_modulaire'
|
|
} = config;
|
|
|
|
await tracer.annotate({
|
|
modularWorkflow: true,
|
|
rowNumber,
|
|
selectiveStack,
|
|
adversarialMode,
|
|
source
|
|
});
|
|
|
|
const startTime = Date.now();
|
|
logSh(`🚀 WORKFLOW MODULAIRE DÉMARRÉ`, 'INFO');
|
|
logSh(` 📊 Ligne: ${rowNumber} | Selective: ${selectiveStack} | Adversarial: ${adversarialMode}`, 'INFO');
|
|
|
|
try {
|
|
// ========================================
|
|
// PHASE 1: PRÉPARATION DONNÉES
|
|
// ========================================
|
|
logSh(`📋 PHASE 1: Préparation données`, 'INFO');
|
|
|
|
const csvData = await readInstructionsData(rowNumber);
|
|
if (!csvData) {
|
|
throw new Error(`Impossible de lire les données ligne ${rowNumber}`);
|
|
}
|
|
|
|
const personalities = await getPersonalities();
|
|
const selectedPersonality = await selectPersonalityWithAI(
|
|
csvData.mc0,
|
|
csvData.t0,
|
|
personalities
|
|
);
|
|
|
|
csvData.personality = selectedPersonality;
|
|
|
|
logSh(` ✅ Données: ${csvData.mc0} | Personnalité: ${selectedPersonality.nom}`, 'DEBUG');
|
|
|
|
// ========================================
|
|
// PHASE 2: EXTRACTION ÉLÉMENTS
|
|
// ========================================
|
|
logSh(`📝 PHASE 2: Extraction éléments XML`, 'INFO');
|
|
|
|
const elements = await extractElements(csvData.xmlTemplate, csvData);
|
|
logSh(` ✅ ${elements.length} éléments extraits`, 'DEBUG');
|
|
|
|
// ========================================
|
|
// PHASE 3: GÉNÉRATION MOTS-CLÉS MANQUANTS
|
|
// ========================================
|
|
logSh(`🔍 PHASE 3: Génération mots-clés manquants`, 'INFO');
|
|
|
|
const finalElements = await generateMissingKeywords(elements, csvData);
|
|
logSh(` ✅ Mots-clés complétés`, 'DEBUG');
|
|
|
|
// ========================================
|
|
// PHASE 4: CONSTRUCTION HIÉRARCHIE
|
|
// ========================================
|
|
logSh(`🏗️ PHASE 4: Construction hiérarchie`, 'INFO');
|
|
|
|
const hierarchy = await buildSmartHierarchy(finalElements);
|
|
logSh(` ✅ ${Object.keys(hierarchy).length} sections hiérarchisées`, 'DEBUG');
|
|
|
|
// ========================================
|
|
// PHASE 5: GÉNÉRATION CONTENU DE BASE
|
|
// ========================================
|
|
logSh(`💫 PHASE 5: Génération contenu de base`, 'INFO');
|
|
|
|
const generatedContent = await generateSimple(hierarchy, csvData);
|
|
|
|
logSh(` ✅ ${Object.keys(generatedContent).length} éléments générés`, 'DEBUG');
|
|
|
|
// ========================================
|
|
// PHASE 6: SELECTIVE ENHANCEMENT MODULAIRE
|
|
// ========================================
|
|
logSh(`🔧 PHASE 6: Selective Enhancement Modulaire (${selectiveStack})`, 'INFO');
|
|
|
|
let selectiveResult;
|
|
|
|
switch (selectiveStack) {
|
|
case 'adaptive':
|
|
selectiveResult = await applyAdaptiveLayers(generatedContent, {
|
|
maxIntensity: 1.1,
|
|
analysisThreshold: 0.3,
|
|
csvData
|
|
});
|
|
break;
|
|
|
|
case 'technical':
|
|
case 'transitions':
|
|
case 'style':
|
|
selectiveResult = await applySelectiveLayer(generatedContent, {
|
|
layerType: selectiveStack,
|
|
llmProvider: 'auto',
|
|
intensity: 1.0,
|
|
csvData
|
|
});
|
|
break;
|
|
|
|
default:
|
|
// Stack prédéfini
|
|
selectiveResult = await applyPredefinedStack(generatedContent, selectiveStack, {
|
|
csvData,
|
|
analysisMode: true
|
|
});
|
|
}
|
|
|
|
const enhancedContent = selectiveResult.content;
|
|
|
|
logSh(` ✅ Selective: ${selectiveResult.stats.elementsEnhanced || selectiveResult.stats.totalModifications || 0} améliorations`, 'INFO');
|
|
|
|
// ========================================
|
|
// PHASE 7: ADVERSARIAL ENHANCEMENT (OPTIONNEL)
|
|
// ========================================
|
|
let finalContent = enhancedContent;
|
|
let adversarialStats = null;
|
|
|
|
if (adversarialMode !== 'none') {
|
|
logSh(`🎯 PHASE 7: Adversarial Enhancement (${adversarialMode})`, 'INFO');
|
|
|
|
let adversarialResult;
|
|
|
|
switch (adversarialMode) {
|
|
case 'adaptive':
|
|
// Utiliser adversarial adaptatif
|
|
adversarialResult = await applyAdversarialLayer(enhancedContent, {
|
|
detectorTarget: 'general',
|
|
method: 'hybrid',
|
|
intensity: 0.8,
|
|
analysisMode: true
|
|
});
|
|
break;
|
|
|
|
case 'light':
|
|
case 'standard':
|
|
case 'heavy':
|
|
// Utiliser stack adversarial prédéfini
|
|
const stackMapping = {
|
|
light: 'lightDefense',
|
|
standard: 'standardDefense',
|
|
heavy: 'heavyDefense'
|
|
};
|
|
|
|
adversarialResult = await applyAdversarialStack(enhancedContent, stackMapping[adversarialMode], {
|
|
csvData
|
|
});
|
|
break;
|
|
}
|
|
|
|
if (adversarialResult && !adversarialResult.fallback) {
|
|
finalContent = adversarialResult.content;
|
|
adversarialStats = adversarialResult.stats;
|
|
|
|
logSh(` ✅ Adversarial: ${adversarialStats.elementsModified || adversarialStats.totalModifications || 0} modifications`, 'INFO');
|
|
} else {
|
|
logSh(` ⚠️ Adversarial fallback: contenu selective préservé`, 'WARNING');
|
|
}
|
|
}
|
|
|
|
// ========================================
|
|
// PHASE 8: ASSEMBLAGE ET STOCKAGE
|
|
// ========================================
|
|
logSh(`🔗 PHASE 8: Assemblage et stockage`, 'INFO');
|
|
|
|
const assembledContent = await injectGeneratedContent(finalContent, csvData.xmlTemplate);
|
|
|
|
const storageResult = await compileAndStoreArticle(assembledContent, {
|
|
...csvData,
|
|
source: `${source}_${selectiveStack}${adversarialMode !== 'none' ? `_${adversarialMode}` : ''}`
|
|
});
|
|
|
|
logSh(` ✅ Stocké: ${storageResult.compiledLength} caractères`, 'DEBUG');
|
|
|
|
// ========================================
|
|
// RÉSUMÉ FINAL
|
|
// ========================================
|
|
const totalDuration = Date.now() - startTime;
|
|
const finalStats = {
|
|
rowNumber,
|
|
selectiveStack,
|
|
adversarialMode,
|
|
totalDuration,
|
|
elementsGenerated: Object.keys(generatedContent).length,
|
|
selectiveEnhancements: selectiveResult.stats.elementsEnhanced || selectiveResult.stats.totalModifications || 0,
|
|
adversarialModifications: adversarialStats?.elementsModified || adversarialStats?.totalModifications || 0,
|
|
finalLength: storageResult.compiledLength,
|
|
personality: selectedPersonality.nom,
|
|
source
|
|
};
|
|
|
|
logSh(`✅ WORKFLOW MODULAIRE TERMINÉ (${totalDuration}ms)`, 'INFO');
|
|
logSh(` 📊 ${finalStats.elementsGenerated} générés | ${finalStats.selectiveEnhancements} selective | ${finalStats.adversarialModifications} adversarial`, 'INFO');
|
|
logSh(` 🎭 Personnalité: ${finalStats.personality} | Taille finale: ${finalStats.finalLength} chars`, 'INFO');
|
|
|
|
await tracer.event('Workflow modulaire terminé', finalStats);
|
|
|
|
return {
|
|
success: true,
|
|
stats: finalStats,
|
|
content: finalContent,
|
|
assembledContent,
|
|
storageResult,
|
|
selectiveResult,
|
|
adversarialResult: adversarialStats ? { stats: adversarialStats } : null
|
|
};
|
|
|
|
} catch (error) {
|
|
const duration = Date.now() - startTime;
|
|
logSh(`❌ WORKFLOW MODULAIRE ÉCHOUÉ après ${duration}ms: ${error.message}`, 'ERROR');
|
|
logSh(`Stack trace: ${error.stack}`, 'ERROR');
|
|
|
|
await tracer.event('Workflow modulaire échoué', {
|
|
error: error.message,
|
|
duration,
|
|
rowNumber,
|
|
selectiveStack,
|
|
adversarialMode
|
|
});
|
|
|
|
throw error;
|
|
}
|
|
}, { config });
|
|
}
|
|
|
|
/**
|
|
* BENCHMARK COMPARATIF STACKS
|
|
*/
|
|
async function benchmarkStacks(rowNumber = 2) {
|
|
console.log('\n⚡ === BENCHMARK STACKS MODULAIRES ===\n');
|
|
|
|
const stacks = getAvailableStacks();
|
|
const adversarialModes = ['none', 'light', 'standard'];
|
|
|
|
const results = [];
|
|
|
|
for (const stack of stacks.slice(0, 3)) { // Tester 3 stacks principaux
|
|
for (const advMode of adversarialModes.slice(0, 2)) { // 2 modes adversarial
|
|
|
|
console.log(`🧪 Test: ${stack.name} + adversarial ${advMode}`);
|
|
|
|
try {
|
|
const startTime = Date.now();
|
|
|
|
const result = await handleModularWorkflow({
|
|
rowNumber,
|
|
selectiveStack: stack.name,
|
|
adversarialMode: advMode,
|
|
source: 'benchmark'
|
|
});
|
|
|
|
const duration = Date.now() - startTime;
|
|
|
|
results.push({
|
|
stack: stack.name,
|
|
adversarial: advMode,
|
|
duration,
|
|
success: true,
|
|
selectiveEnhancements: result.stats.selectiveEnhancements,
|
|
adversarialModifications: result.stats.adversarialModifications,
|
|
finalLength: result.stats.finalLength
|
|
});
|
|
|
|
console.log(` ✅ ${duration}ms | ${result.stats.selectiveEnhancements} selective | ${result.stats.adversarialModifications} adversarial`);
|
|
|
|
} catch (error) {
|
|
results.push({
|
|
stack: stack.name,
|
|
adversarial: advMode,
|
|
success: false,
|
|
error: error.message
|
|
});
|
|
|
|
console.log(` ❌ Échoué: ${error.message}`);
|
|
}
|
|
}
|
|
}
|
|
|
|
// Résumé benchmark
|
|
console.log('\n📊 RÉSUMÉ BENCHMARK:');
|
|
|
|
const successful = results.filter(r => r.success);
|
|
if (successful.length > 0) {
|
|
const avgDuration = successful.reduce((sum, r) => sum + r.duration, 0) / successful.length;
|
|
const bestPerf = successful.reduce((best, r) => r.duration < best.duration ? r : best);
|
|
const mostEnhancements = successful.reduce((best, r) =>
|
|
(r.selectiveEnhancements + r.adversarialModifications) > (best.selectiveEnhancements + best.adversarialModifications) ? r : best
|
|
);
|
|
|
|
console.log(` ⚡ Durée moyenne: ${avgDuration.toFixed(0)}ms`);
|
|
console.log(` 🏆 Meilleure perf: ${bestPerf.stack} + ${bestPerf.adversarial} (${bestPerf.duration}ms)`);
|
|
console.log(` 🔥 Plus d'améliorations: ${mostEnhancements.stack} + ${mostEnhancements.adversarial} (${mostEnhancements.selectiveEnhancements + mostEnhancements.adversarialModifications})`);
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* INTERFACE LIGNE DE COMMANDE
|
|
*/
|
|
async function main() {
|
|
const args = process.argv.slice(2);
|
|
const command = args[0] || 'workflow';
|
|
|
|
try {
|
|
switch (command) {
|
|
case 'workflow':
|
|
const rowNumber = parseInt(args[1]) || 2;
|
|
const selectiveStack = args[2] || 'standardEnhancement';
|
|
const adversarialMode = args[3] || 'light';
|
|
|
|
console.log(`\n🚀 Exécution workflow modulaire:`);
|
|
console.log(` 📊 Ligne: ${rowNumber}`);
|
|
console.log(` 🔧 Stack selective: ${selectiveStack}`);
|
|
console.log(` 🎯 Mode adversarial: ${adversarialMode}`);
|
|
|
|
const result = await handleModularWorkflow({
|
|
rowNumber,
|
|
selectiveStack,
|
|
adversarialMode
|
|
});
|
|
|
|
console.log('\n✅ WORKFLOW MODULAIRE RÉUSSI');
|
|
console.log(`📈 Stats: ${JSON.stringify(result.stats, null, 2)}`);
|
|
break;
|
|
|
|
case 'benchmark':
|
|
const benchRowNumber = parseInt(args[1]) || 2;
|
|
|
|
console.log(`\n⚡ Benchmark stacks (ligne ${benchRowNumber})`);
|
|
const benchResults = await benchmarkStacks(benchRowNumber);
|
|
|
|
console.log('\n📊 Résultats complets:');
|
|
console.table(benchResults);
|
|
break;
|
|
|
|
case 'stacks':
|
|
console.log('\n📦 STACKS SELECTIVE DISPONIBLES:');
|
|
const availableStacks = getAvailableStacks();
|
|
availableStacks.forEach(stack => {
|
|
console.log(`\n 🔧 ${stack.name}:`);
|
|
console.log(` 📝 ${stack.description}`);
|
|
console.log(` 📊 ${stack.layersCount} couches`);
|
|
console.log(` 🎯 Couches: ${stack.layers ? stack.layers.map(l => `${l.type}(${l.llm})`).join(' → ') : 'N/A'}`);
|
|
});
|
|
|
|
console.log('\n🎯 MODES ADVERSARIAL DISPONIBLES:');
|
|
console.log(' - none: Pas d\'adversarial');
|
|
console.log(' - light: Défense légère');
|
|
console.log(' - standard: Défense standard');
|
|
console.log(' - heavy: Défense intensive');
|
|
console.log(' - adaptive: Adaptatif intelligent');
|
|
break;
|
|
|
|
case 'help':
|
|
default:
|
|
console.log('\n🔧 === MAIN MODULAIRE - USAGE ===');
|
|
console.log('\nCommandes disponibles:');
|
|
console.log(' workflow [ligne] [stack] [adversarial] - Exécuter workflow complet');
|
|
console.log(' benchmark [ligne] - Benchmark stacks');
|
|
console.log(' stacks - Lister stacks disponibles');
|
|
console.log(' help - Afficher cette aide');
|
|
console.log('\nExemples:');
|
|
console.log(' node main_modulaire.js workflow 2 fullEnhancement standard');
|
|
console.log(' node main_modulaire.js workflow 3 adaptive light');
|
|
console.log(' node main_modulaire.js benchmark 2');
|
|
console.log(' node main_modulaire.js stacks');
|
|
break;
|
|
}
|
|
|
|
} catch (error) {
|
|
console.error('\n❌ ERREUR MAIN MODULAIRE:', error.message);
|
|
console.error(error.stack);
|
|
process.exit(1);
|
|
}
|
|
}
|
|
|
|
// Export pour usage programmatique
|
|
module.exports = {
|
|
handleModularWorkflow,
|
|
benchmarkStacks
|
|
};
|
|
|
|
// Exécution CLI si appelé directement
|
|
if (require.main === module) {
|
|
main().catch(error => {
|
|
console.error('❌ ERREUR FATALE:', error.message);
|
|
process.exit(1);
|
|
});
|
|
}
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/ManualTrigger.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
const { logSh } = require('./ErrorReporting'); // Using unified logSh from ErrorReporting
|
|
|
|
/**
|
|
* 🚀 TRIGGER MANUEL - Lit ligne 2 et lance le workflow
|
|
* Exécute cette fonction depuis l'éditeur Apps Script
|
|
*/
|
|
function runWorkflowLigne(numeroLigne = 2) {
|
|
cleanLogSheet(); // Nettoie les logs pour ce test
|
|
|
|
try {
|
|
logSh('🎬 >>> DÉMARRAGE WORKFLOW MANUEL <<<', 'INFO');
|
|
|
|
// 1. LIRE AUTOMATIQUEMENT LA LIGNE INDIQUÉ
|
|
const csvData = readCSVDataFromRow(numeroLigne);
|
|
logSh(`✅ Données lues - MC0: ${csvData.mc0}`, 'INFO');
|
|
logSh(`✅ Titre: ${csvData.t0}`, 'INFO');
|
|
logSh(`✅ Personnalité: ${csvData.personality.nom}`, 'INFO');
|
|
|
|
// 2. XML TEMPLATE SIMPLE POUR TEST (ou lit depuis Digital Ocean si configuré)
|
|
const xmlTemplate = getXMLTemplateForTest(csvData);
|
|
logSh(`✅ XML template: ${xmlTemplate.length} caractères`, 'INFO');
|
|
|
|
// 3. 🎯 LANCER LE WORKFLOW PRINCIPAL
|
|
const workflowData = {
|
|
csvData: csvData,
|
|
xmlTemplate: Utilities.base64Encode(xmlTemplate),
|
|
source: 'manuel_ligne2'
|
|
};
|
|
|
|
const result = handleFullWorkflow(workflowData);
|
|
|
|
logSh('🏆 === WORKFLOW MANUEL TERMINÉ ===', 'INFO');
|
|
|
|
// ← EXTRAIRE LES VRAIES DONNÉES
|
|
let actualData;
|
|
if (result && result.getContentText) {
|
|
// C'est un ContentService, extraire le JSON
|
|
actualData = JSON.parse(result.getContentText());
|
|
} else {
|
|
actualData = result;
|
|
}
|
|
|
|
logSh(`Type result: ${typeof result}`, 'DEBUG');
|
|
logSh(`Result keys: ${Object.keys(result || {})}`, 'DEBUG');
|
|
logSh(`ActualData keys: ${Object.keys(actualData || {})}`, 'DEBUG');
|
|
logSh(`ActualData: ${JSON.stringify(actualData)}`, 'DEBUG');
|
|
|
|
if (actualData && actualData.stats) {
|
|
logSh(`📊 Éléments générés: ${actualData.stats.contentPieces}`, 'INFO');
|
|
logSh(`📝 Nombre de mots: ${actualData.stats.wordCount}`, 'INFO');
|
|
} else {
|
|
logSh('⚠️ Format résultat inattendu', 'WARNING');
|
|
logSh('ActualData: ' + JSON.stringify(actualData, null, 2), 'DEBUG'); // Using logSh instead of console.log
|
|
}
|
|
|
|
return actualData;
|
|
|
|
} catch (error) {
|
|
logSh(`❌ ERREUR WORKFLOW MANUEL: ${error.toString()}`, 'ERROR');
|
|
logSh(`Stack: ${error.stack}`, 'ERROR');
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* HELPER - Lire CSV depuis une ligne spécifique
|
|
*/
|
|
function readCSVDataFromRow(rowNumber) {
|
|
const sheetId = '1iA2GvWeUxX-vpnAMfVm3ZMG9LhaC070SdGssEcXAh2c';
|
|
const spreadsheet = SpreadsheetApp.openById(sheetId);
|
|
const articlesSheet = spreadsheet.getSheetByName('instructions');
|
|
|
|
// Lire la ligne complète (colonnes A à H)
|
|
const range = articlesSheet.getRange(rowNumber, 1, 1, 9);
|
|
const [slug, t0, mc0, tMinus1, lMinus1, mcPlus1, tPlus1, lPlus1, xmlFileName] = range.getValues()[0];
|
|
|
|
logSh(`📖 Lecture ligne ${rowNumber}: ${slug}`, 'DEBUG');
|
|
|
|
// Récupérer personnalités et sélectionner automatiquement
|
|
const personalitiesSheet = spreadsheet.getSheetByName('Personnalites');
|
|
const personalities = getPersonalities(personalitiesSheet);
|
|
const selectedPersonality = selectPersonalityWithAI(mc0, t0, personalities);
|
|
|
|
return {
|
|
rowNumber: rowNumber,
|
|
slug: slug || 'test-slug',
|
|
t0: t0 || 'Titre par défaut',
|
|
mc0: mc0 || 'mot-clé test',
|
|
tMinus1: tMinus1 || 'parent',
|
|
lMinus1: lMinus1 || '/parent',
|
|
mcPlus1: mcPlus1 || 'mot1,mot2,mot3,mot4',
|
|
tPlus1: tPlus1 || 'Titre1,Titre2,Titre3,Titre4',
|
|
lPlus1: lPlus1 || '/lien1,/lien2,/lien3,/lien4',
|
|
personality: selectedPersonality,
|
|
xmlFileName: xmlFileName ? xmlFileName.toString().trim() : null
|
|
};
|
|
}
|
|
|
|
/**
|
|
* HELPER - XML Template simple pour test (ou depuis Digital Ocean)
|
|
*/
|
|
function getXMLTemplateForTest(csvData) {
|
|
logSh("csvData.xmlFileName: " + csvData.xmlFileName, 'DEBUG'); // Using logSh instead of console.log
|
|
|
|
if (csvData.xmlFileName) {
|
|
logSh("Tentative Digital Ocean...", 'INFO'); // Using logSh instead of console.log
|
|
try {
|
|
return fetchXMLFromDigitalOceanSimple(csvData.xmlFileName);
|
|
} catch (error) {
|
|
// ← ENLÈVE LE CATCH SILENCIEUX
|
|
logSh("Erreur DO: " + error.toString(), 'WARNING'); // Using logSh instead of console.log
|
|
logSh(`❌ ERREUR DO DÉTAILLÉE: ${error.toString()}`, 'ERROR');
|
|
|
|
// Continue sans Digital Ocean
|
|
}
|
|
}
|
|
|
|
logSh("❌ FATAL: Aucun template XML disponible", 'ERROR');
|
|
throw new Error("FATAL: Template XML indisponible (Digital Ocean inaccessible + pas de fallback) - arrêt du workflow");
|
|
}
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/selective-enhancement/demo-modulaire.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// DÉMONSTRATION ARCHITECTURE MODULAIRE SELECTIVE
|
|
// Usage: node lib/selective-enhancement/demo-modulaire.js
|
|
// Objectif: Valider l'intégration modulaire selective enhancement
|
|
// ========================================
|
|
|
|
const { logSh } = require('../ErrorReporting');
|
|
|
|
// Import modules selective modulaires
|
|
const { applySelectiveLayer } = require('./SelectiveCore');
|
|
const {
|
|
applyPredefinedStack,
|
|
applyAdaptiveLayers,
|
|
getAvailableStacks
|
|
} = require('./SelectiveLayers');
|
|
const {
|
|
analyzeTechnicalQuality,
|
|
analyzeTransitionFluidity,
|
|
analyzeStyleConsistency,
|
|
generateImprovementReport
|
|
} = require('./SelectiveUtils');
|
|
|
|
/**
|
|
* EXEMPLE D'UTILISATION MODULAIRE SELECTIVE
|
|
*/
|
|
async function demoModularSelective() {
|
|
console.log('\n🔧 === DÉMONSTRATION SELECTIVE MODULAIRE ===\n');
|
|
|
|
// Contenu d'exemple avec problèmes de qualité
|
|
const exempleContenu = {
|
|
'|Titre_Principal_1|': 'Guide complet pour choisir votre plaque personnalisée',
|
|
'|Introduction_1|': 'La personnalisation d\'une plaque signalétique représente un enjeu important pour votre entreprise. Cette solution permet de créer une identité visuelle.',
|
|
'|Texte_1|': 'Il est important de noter que les matériaux utilisés sont de qualité. Par ailleurs, la qualité est bonne. En effet, nos solutions sont bonnes et robustes. Par ailleurs, cela fonctionne bien.',
|
|
'|FAQ_Question_1|': 'Quels sont les matériaux disponibles ?',
|
|
'|FAQ_Reponse_1|': 'Nos matériaux sont de qualité : ils conviennent parfaitement. Ces solutions garantissent une qualité et un rendu optimal.'
|
|
};
|
|
|
|
console.log('📊 CONTENU ORIGINAL:');
|
|
Object.entries(exempleContenu).forEach(([tag, content]) => {
|
|
console.log(` ${tag}: "${content}"`);
|
|
});
|
|
|
|
// Analyser qualité originale
|
|
const fullOriginal = Object.values(exempleContenu).join(' ');
|
|
const qualiteOriginale = {
|
|
technical: analyzeTechnicalQuality(fullOriginal, ['dibond', 'aluminium', 'pmma', 'impression']),
|
|
transitions: analyzeTransitionFluidity(fullOriginal),
|
|
style: analyzeStyleConsistency(fullOriginal)
|
|
};
|
|
|
|
console.log(`\n📈 QUALITÉ ORIGINALE:`);
|
|
console.log(` 🔧 Technique: ${qualiteOriginale.technical.score}/100`);
|
|
console.log(` 🔗 Transitions: ${qualiteOriginale.transitions.score}/100`);
|
|
console.log(` 🎨 Style: ${qualiteOriginale.style.score}/100`);
|
|
|
|
try {
|
|
// ========================================
|
|
// TEST 1: COUCHE TECHNIQUE SEULE
|
|
// ========================================
|
|
console.log('\n🔧 TEST 1: Application couche technique');
|
|
|
|
const result1 = await applySelectiveLayer(exempleContenu, {
|
|
layerType: 'technical',
|
|
llmProvider: 'gpt4',
|
|
intensity: 0.9,
|
|
csvData: {
|
|
personality: { nom: 'Marc', style: 'technique' },
|
|
mc0: 'plaque personnalisée'
|
|
}
|
|
});
|
|
|
|
console.log(`✅ Résultat: ${result1.stats.enhanced}/${result1.stats.processed} éléments améliorés`);
|
|
console.log(` ⏱️ Durée: ${result1.stats.duration}ms`);
|
|
|
|
// ========================================
|
|
// TEST 2: STACK PRÉDÉFINI
|
|
// ========================================
|
|
console.log('\n📦 TEST 2: Application stack prédéfini');
|
|
|
|
// Lister stacks disponibles
|
|
const stacks = getAvailableStacks();
|
|
console.log(' Stacks disponibles:');
|
|
stacks.forEach(stack => {
|
|
console.log(` - ${stack.name}: ${stack.description}`);
|
|
});
|
|
|
|
const result2 = await applyPredefinedStack(exempleContenu, 'standardEnhancement', {
|
|
csvData: {
|
|
personality: {
|
|
nom: 'Sophie',
|
|
style: 'professionnel',
|
|
vocabulairePref: 'signalétique,personnalisation,qualité,expertise',
|
|
niveauTechnique: 'standard'
|
|
},
|
|
mc0: 'plaque personnalisée'
|
|
}
|
|
});
|
|
|
|
console.log(`✅ Stack standard: ${result2.stats.totalModifications} modifications totales`);
|
|
console.log(` 📊 Couches: ${result2.stats.layers.filter(l => l.success).length}/${result2.stats.layers.length} réussies`);
|
|
|
|
// ========================================
|
|
// TEST 3: COUCHES ADAPTATIVES
|
|
// ========================================
|
|
console.log('\n🧠 TEST 3: Application couches adaptatives');
|
|
|
|
const result3 = await applyAdaptiveLayers(exempleContenu, {
|
|
maxIntensity: 1.2,
|
|
analysisThreshold: 0.3,
|
|
csvData: {
|
|
personality: {
|
|
nom: 'Laurent',
|
|
style: 'commercial',
|
|
vocabulairePref: 'expertise,solution,performance,innovation',
|
|
niveauTechnique: 'accessible'
|
|
},
|
|
mc0: 'signalétique personnalisée'
|
|
}
|
|
});
|
|
|
|
if (result3.stats.adaptive) {
|
|
console.log(`✅ Adaptatif: ${result3.stats.layersApplied} couches appliquées`);
|
|
console.log(` 📊 Modifications: ${result3.stats.totalModifications}`);
|
|
}
|
|
|
|
// ========================================
|
|
// COMPARAISON QUALITÉ FINALE
|
|
// ========================================
|
|
console.log('\n📊 ANALYSE QUALITÉ FINALE:');
|
|
|
|
const contenuFinal = result2.content; // Prendre résultat stack standard
|
|
const fullEnhanced = Object.values(contenuFinal).join(' ');
|
|
|
|
const qualiteFinale = {
|
|
technical: analyzeTechnicalQuality(fullEnhanced, ['dibond', 'aluminium', 'pmma', 'impression']),
|
|
transitions: analyzeTransitionFluidity(fullEnhanced),
|
|
style: analyzeStyleConsistency(fullEnhanced, result2.csvData?.personality)
|
|
};
|
|
|
|
console.log('\n📈 AMÉLIORATION QUALITÉ:');
|
|
console.log(` 🔧 Technique: ${qualiteOriginale.technical.score} → ${qualiteFinale.technical.score} (+${(qualiteFinale.technical.score - qualiteOriginale.technical.score).toFixed(1)})`);
|
|
console.log(` 🔗 Transitions: ${qualiteOriginale.transitions.score} → ${qualiteFinale.transitions.score} (+${(qualiteFinale.transitions.score - qualiteOriginale.transitions.score).toFixed(1)})`);
|
|
console.log(` 🎨 Style: ${qualiteOriginale.style.score} → ${qualiteFinale.style.score} (+${(qualiteFinale.style.score - qualiteOriginale.style.score).toFixed(1)})`);
|
|
|
|
// Rapport détaillé
|
|
const rapport = generateImprovementReport(exempleContenu, contenuFinal, 'selective');
|
|
|
|
console.log('\n📋 RAPPORT AMÉLIORATION:');
|
|
console.log(` 📈 Amélioration moyenne: ${rapport.summary.averageImprovement.toFixed(1)}%`);
|
|
console.log(` ✅ Éléments améliorés: ${rapport.summary.elementsImproved}/${rapport.summary.elementsProcessed}`);
|
|
|
|
if (rapport.details.recommendations.length > 0) {
|
|
console.log(` 💡 Recommandations: ${rapport.details.recommendations.join(', ')}`);
|
|
}
|
|
|
|
// ========================================
|
|
// EXEMPLES DE TRANSFORMATION
|
|
// ========================================
|
|
console.log('\n✨ EXEMPLES DE TRANSFORMATION:');
|
|
|
|
console.log('\n📝 INTRODUCTION:');
|
|
console.log('AVANT:', `"${exempleContenu['|Introduction_1|']}"`);
|
|
console.log('APRÈS:', `"${contenuFinal['|Introduction_1|']}"`);
|
|
|
|
console.log('\n📝 TEXTE PRINCIPAL:');
|
|
console.log('AVANT:', `"${exempleContenu['|Texte_1|']}"`);
|
|
console.log('APRÈS:', `"${contenuFinal['|Texte_1|']}"`);
|
|
|
|
console.log('\n✅ === DÉMONSTRATION SELECTIVE MODULAIRE TERMINÉE ===\n');
|
|
|
|
return {
|
|
success: true,
|
|
originalQuality: qualiteOriginale,
|
|
finalQuality: qualiteFinale,
|
|
improvementReport: rapport
|
|
};
|
|
|
|
} catch (error) {
|
|
console.error('\n❌ ERREUR DÉMONSTRATION:', error.message);
|
|
console.error(error.stack);
|
|
return { success: false, error: error.message };
|
|
}
|
|
}
|
|
|
|
/**
|
|
* EXEMPLE D'INTÉGRATION AVEC PIPELINE EXISTANTE
|
|
*/
|
|
async function demoIntegrationExistante() {
|
|
console.log('\n🔗 === DÉMONSTRATION INTÉGRATION PIPELINE ===\n');
|
|
|
|
// Simuler contenu venant de ContentGeneration.js (Level 1)
|
|
const contenuExistant = {
|
|
'|Titre_H1_1|': 'Solutions de plaques personnalisées professionnelles',
|
|
'|Meta_Description_1|': 'Découvrez notre gamme complète de plaques personnalisées pour tous vos besoins de signalétique professionnelle.',
|
|
'|Introduction_1|': 'Dans le domaine de la signalétique personnalisée, le choix des matériaux et des techniques de fabrication constitue un élément déterminant.',
|
|
'|Texte_Avantages_1|': 'Les avantages de nos solutions incluent la durabilité, la résistance aux intempéries et la possibilité de personnalisation complète.'
|
|
};
|
|
|
|
console.log('💼 SCÉNARIO: Application selective post-génération normale');
|
|
|
|
try {
|
|
console.log('\n🎯 Étape 1: Contenu généré par pipeline Level 1');
|
|
console.log(' ✅ Contenu de base: qualité préservée');
|
|
|
|
console.log('\n🎯 Étape 2: Application selective enhancement modulaire');
|
|
|
|
// Test avec couche technique puis style
|
|
let contenuEnhanced = contenuExistant;
|
|
|
|
// Amélioration technique
|
|
const resultTechnique = await applySelectiveLayer(contenuEnhanced, {
|
|
layerType: 'technical',
|
|
llmProvider: 'gpt4',
|
|
intensity: 1.0,
|
|
analysisMode: true,
|
|
csvData: {
|
|
personality: { nom: 'Marc', style: 'technique' },
|
|
mc0: 'plaque personnalisée'
|
|
}
|
|
});
|
|
|
|
contenuEnhanced = resultTechnique.content;
|
|
console.log(` ✅ Couche technique: ${resultTechnique.stats.enhanced} éléments améliorés`);
|
|
|
|
// Amélioration style
|
|
const resultStyle = await applySelectiveLayer(contenuEnhanced, {
|
|
layerType: 'style',
|
|
llmProvider: 'mistral',
|
|
intensity: 0.8,
|
|
analysisMode: true,
|
|
csvData: {
|
|
personality: {
|
|
nom: 'Sophie',
|
|
style: 'professionnel moderne',
|
|
vocabulairePref: 'innovation,expertise,personnalisation,qualité',
|
|
niveauTechnique: 'accessible'
|
|
}
|
|
}
|
|
});
|
|
|
|
contenuEnhanced = resultStyle.content;
|
|
console.log(` ✅ Couche style: ${resultStyle.stats.enhanced} éléments stylisés`);
|
|
|
|
console.log('\n📊 RÉSULTAT FINAL INTÉGRÉ:');
|
|
Object.entries(contenuEnhanced).forEach(([tag, content]) => {
|
|
console.log(`\n ${tag}:`);
|
|
console.log(` ORIGINAL: "${contenuExistant[tag]}"`);
|
|
console.log(` ENHANCED: "${content}"`);
|
|
});
|
|
|
|
return {
|
|
success: true,
|
|
techniqueResult: resultTechnique,
|
|
styleResult: resultStyle,
|
|
finalContent: contenuEnhanced
|
|
};
|
|
|
|
} catch (error) {
|
|
console.error('❌ ERREUR INTÉGRATION:', error.message);
|
|
return { success: false, error: error.message };
|
|
}
|
|
}
|
|
|
|
/**
|
|
* TEST PERFORMANCE ET BENCHMARKS
|
|
*/
|
|
async function benchmarkPerformance() {
|
|
console.log('\n⚡ === BENCHMARK PERFORMANCE ===\n');
|
|
|
|
// Contenu de test de taille variable
|
|
const contenuTest = {};
|
|
|
|
// Générer contenu test
|
|
for (let i = 1; i <= 10; i++) {
|
|
contenuTest[`|Element_${i}|`] = `Ceci est un contenu de test numéro ${i} pour valider les performances du système selective enhancement modulaire. ` +
|
|
`Il est important de noter que ce contenu contient du vocabulaire générique et des répétitions. Par ailleurs, les transitions sont basiques. ` +
|
|
`En effet, la qualité technique est faible et le style est générique. Par ailleurs, cela nécessite des améliorations.`.repeat(Math.floor(i/3) + 1);
|
|
}
|
|
|
|
console.log(`📊 Contenu test: ${Object.keys(contenuTest).length} éléments`);
|
|
|
|
try {
|
|
const benchmarks = [];
|
|
|
|
// Test 1: Couche technique seule
|
|
const start1 = Date.now();
|
|
const result1 = await applySelectiveLayer(contenuTest, {
|
|
layerType: 'technical',
|
|
intensity: 0.8
|
|
});
|
|
benchmarks.push({
|
|
test: 'Couche technique seule',
|
|
duration: Date.now() - start1,
|
|
enhanced: result1.stats.enhanced,
|
|
processed: result1.stats.processed
|
|
});
|
|
|
|
// Test 2: Stack complet
|
|
const start2 = Date.now();
|
|
const result2 = await applyPredefinedStack(contenuTest, 'fullEnhancement');
|
|
benchmarks.push({
|
|
test: 'Stack complet (3 couches)',
|
|
duration: Date.now() - start2,
|
|
totalModifications: result2.stats.totalModifications,
|
|
layers: result2.stats.layers.length
|
|
});
|
|
|
|
// Test 3: Adaptatif
|
|
const start3 = Date.now();
|
|
const result3 = await applyAdaptiveLayers(contenuTest, { maxIntensity: 1.0 });
|
|
benchmarks.push({
|
|
test: 'Couches adaptatives',
|
|
duration: Date.now() - start3,
|
|
layersApplied: result3.stats.layersApplied,
|
|
totalModifications: result3.stats.totalModifications
|
|
});
|
|
|
|
console.log('\n📈 RÉSULTATS BENCHMARK:');
|
|
benchmarks.forEach(bench => {
|
|
console.log(`\n ${bench.test}:`);
|
|
console.log(` ⏱️ Durée: ${bench.duration}ms`);
|
|
if (bench.enhanced) console.log(` ✅ Améliorés: ${bench.enhanced}/${bench.processed}`);
|
|
if (bench.totalModifications) console.log(` 🔄 Modifications: ${bench.totalModifications}`);
|
|
if (bench.layers) console.log(` 📦 Couches: ${bench.layers}`);
|
|
if (bench.layersApplied) console.log(` 🧠 Couches adaptées: ${bench.layersApplied}`);
|
|
});
|
|
|
|
return { success: true, benchmarks };
|
|
|
|
} catch (error) {
|
|
console.error('❌ ERREUR BENCHMARK:', error.message);
|
|
return { success: false, error: error.message };
|
|
}
|
|
}
|
|
|
|
// Exécuter démonstrations si fichier appelé directement
|
|
if (require.main === module) {
|
|
(async () => {
|
|
await demoModularSelective();
|
|
await demoIntegrationExistante();
|
|
await benchmarkPerformance();
|
|
})().catch(console.error);
|
|
}
|
|
|
|
module.exports = {
|
|
demoModularSelective,
|
|
demoIntegrationExistante,
|
|
benchmarkPerformance
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/SelectiveEnhancement.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: SelectiveEnhancement.js - Node.js Version
|
|
// Description: Enhancement par batch pour éviter timeouts
|
|
// ========================================
|
|
|
|
const { callLLM } = require('./LLMManager');
|
|
const { logSh } = require('./ErrorReporting');
|
|
const { tracer } = require('./trace.js');
|
|
const { selectMultiplePersonalitiesWithAI, getPersonalities } = require('./BrainConfig');
|
|
|
|
// Utilitaire pour les délais
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
/**
|
|
* NOUVELLE APPROCHE - Multi-Personnalités Batch Enhancement
|
|
* 4 personnalités différentes utilisées dans le pipeline pour maximum d'anti-détection
|
|
*/
|
|
async function generateWithBatchEnhancement(hierarchy, csvData) {
|
|
const totalElements = Object.keys(hierarchy).length;
|
|
|
|
// NOUVEAU: Sélection de 4 personnalités complémentaires
|
|
const personalities = await tracer.run('SelectiveEnhancement.selectMultiplePersonalities()', async () => {
|
|
const allPersonalities = await getPersonalities();
|
|
const selectedPersonalities = await selectMultiplePersonalitiesWithAI(csvData.mc0, csvData.t0, allPersonalities);
|
|
await tracer.event(`4 personnalités sélectionnées: ${selectedPersonalities.map(p => p.nom).join(', ')}`);
|
|
return selectedPersonalities;
|
|
}, { mc0: csvData.mc0, t0: csvData.t0 });
|
|
|
|
await tracer.annotate({
|
|
totalElements,
|
|
personalities: personalities.map(p => `${p.nom}(${p.style})`).join(', '),
|
|
mc0: csvData.mc0
|
|
});
|
|
|
|
// ÉTAPE 1 : Génération base avec IA configurée + Personnalité 1
|
|
const baseContents = await tracer.run('SelectiveEnhancement.generateAllContentBase()', async () => {
|
|
const csvDataWithPersonality1 = { ...csvData, personality: personalities[0] };
|
|
const aiProvider1 = personalities[0].aiEtape1Base;
|
|
const result = await generateAllContentBase(hierarchy, csvDataWithPersonality1, aiProvider1);
|
|
await tracer.event(`${Object.keys(result).length} éléments générés avec ${personalities[0].nom} via ${aiProvider1.toUpperCase()}`);
|
|
return result;
|
|
}, { hierarchyElements: Object.keys(hierarchy).length, personality1: personalities[0].nom, llmProvider: personalities[0].aiEtape1Base, mc0: csvData.mc0 });
|
|
|
|
// ÉTAPE 2 : Enhancement technique avec IA configurée + Personnalité 2
|
|
const technicalEnhanced = await tracer.run('SelectiveEnhancement.enhanceAllTechnicalTerms()', async () => {
|
|
const csvDataWithPersonality2 = { ...csvData, personality: personalities[1] };
|
|
const aiProvider2 = personalities[1].aiEtape2Technique;
|
|
const result = await enhanceAllTechnicalTerms(baseContents, csvDataWithPersonality2, aiProvider2);
|
|
const enhancedCount = Object.keys(result).filter(k => result[k] !== baseContents[k]).length;
|
|
await tracer.event(`${enhancedCount}/${Object.keys(result).length} éléments techniques améliorés avec ${personalities[1].nom} via ${aiProvider2.toUpperCase()}`);
|
|
return result;
|
|
}, { baseElements: Object.keys(baseContents).length, personality2: personalities[1].nom, llmProvider: personalities[1].aiEtape2Technique, mc0: csvData.mc0 });
|
|
|
|
// ÉTAPE 3 : Enhancement transitions avec IA configurée + Personnalité 3
|
|
const transitionsEnhanced = await tracer.run('SelectiveEnhancement.enhanceAllTransitions()', async () => {
|
|
const csvDataWithPersonality3 = { ...csvData, personality: personalities[2] };
|
|
const aiProvider3 = personalities[2].aiEtape3Transitions;
|
|
const result = await enhanceAllTransitions(technicalEnhanced, csvDataWithPersonality3, aiProvider3);
|
|
const enhancedCount = Object.keys(result).filter(k => result[k] !== technicalEnhanced[k]).length;
|
|
await tracer.event(`${enhancedCount}/${Object.keys(result).length} transitions fluidifiées avec ${personalities[2].nom} via ${aiProvider3.toUpperCase()}`);
|
|
return result;
|
|
}, { technicalElements: Object.keys(technicalEnhanced).length, personality3: personalities[2].nom, llmProvider: personalities[2].aiEtape3Transitions });
|
|
|
|
// ÉTAPE 4 : Enhancement style avec IA configurée + Personnalité 4
|
|
const finalContents = await tracer.run('SelectiveEnhancement.enhanceAllPersonalityStyle()', async () => {
|
|
const csvDataWithPersonality4 = { ...csvData, personality: personalities[3] };
|
|
const aiProvider4 = personalities[3].aiEtape4Style;
|
|
const result = await enhanceAllPersonalityStyle(transitionsEnhanced, csvDataWithPersonality4, aiProvider4);
|
|
const enhancedCount = Object.keys(result).filter(k => result[k] !== transitionsEnhanced[k]).length;
|
|
const avgWords = Math.round(Object.values(result).reduce((acc, content) => acc + content.split(' ').length, 0) / Object.keys(result).length);
|
|
await tracer.event(`${enhancedCount}/${Object.keys(result).length} éléments stylisés avec ${personalities[3].nom} via ${aiProvider4.toUpperCase()}`, { avgWordsPerElement: avgWords });
|
|
return result;
|
|
}, { transitionElements: Object.keys(transitionsEnhanced).length, personality4: personalities[3].nom, llmProvider: personalities[3].aiEtape4Style });
|
|
|
|
// Log final du DNA Mixing réussi avec IA configurables
|
|
const aiChain = personalities.map((p, i) => `${p.aiEtape1Base || p.aiEtape2Technique || p.aiEtape3Transitions || p.aiEtape4Style}`.toUpperCase()).join(' → ');
|
|
logSh(`✅ DNA MIXING MULTI-PERSONNALITÉS TERMINÉ:`, 'INFO');
|
|
logSh(` 🎭 4 personnalités utilisées: ${personalities.map(p => p.nom).join(' → ')}`, 'INFO');
|
|
logSh(` 🤖 IA configurées: ${personalities[0].aiEtape1Base.toUpperCase()} → ${personalities[1].aiEtape2Technique.toUpperCase()} → ${personalities[2].aiEtape3Transitions.toUpperCase()} → ${personalities[3].aiEtape4Style.toUpperCase()}`, 'INFO');
|
|
logSh(` 📝 ${Object.keys(finalContents).length} éléments avec style hybride généré`, 'INFO');
|
|
|
|
return finalContents;
|
|
}
|
|
|
|
/**
|
|
* ÉTAPE 1 - Génération base TOUS éléments avec IA configurable
|
|
*/
|
|
async function generateAllContentBase(hierarchy, csvData, aiProvider) {
|
|
logSh('🔍 === DEBUG GÉNÉRATION BASE ===', 'DEBUG');
|
|
|
|
// Debug: logger la hiérarchie complète
|
|
logSh(`🔍 Hiérarchie reçue: ${Object.keys(hierarchy).length} sections`, 'DEBUG');
|
|
Object.keys(hierarchy).forEach((path, i) => {
|
|
const section = hierarchy[path];
|
|
logSh(`🔍 Section ${i+1} [${path}]:`, 'DEBUG');
|
|
logSh(`🔍 - title: ${section.title ? section.title.originalElement?.originalTag : 'AUCUN'}`, 'DEBUG');
|
|
logSh(`🔍 - text: ${section.text ? section.text.originalElement?.originalTag : 'AUCUN'}`, 'DEBUG');
|
|
logSh(`🔍 - questions: ${section.questions?.length || 0}`, 'DEBUG');
|
|
});
|
|
|
|
const allElements = collectAllElements(hierarchy);
|
|
logSh(`🔍 Éléments collectés: ${allElements.length}`, 'DEBUG');
|
|
|
|
// Debug: logger tous les éléments collectés
|
|
allElements.forEach((element, i) => {
|
|
logSh(`🔍 Élément ${i+1}: tag="${element.tag}", type="${element.type}"`, 'DEBUG');
|
|
});
|
|
|
|
// NOUVELLE LOGIQUE : SÉPARER PAIRES FAQ ET AUTRES ÉLÉMENTS
|
|
const results = {};
|
|
|
|
logSh(`🔍 === GÉNÉRATION INTELLIGENTE DE ${allElements.length} ÉLÉMENTS ===`, 'DEBUG');
|
|
logSh(`🔍 Ordre respecté: ${allElements.map(el => el.tag.replace(/\|/g, '')).join(' → ')}`, 'DEBUG');
|
|
|
|
// 1. IDENTIFIER les paires FAQ
|
|
const { faqPairs, otherElements } = separateFAQPairsAndOthers(allElements);
|
|
|
|
logSh(`🔍 ${faqPairs.length} paires FAQ trouvées, ${otherElements.length} autres éléments`, 'INFO');
|
|
|
|
// 2. GÉNÉRER les autres éléments EN BATCH ORDONNÉ (titres d'abord, puis textes avec contexte)
|
|
const groupedElements = groupElementsByType(otherElements);
|
|
|
|
// ORDRE DE GÉNÉRATION : TITRES → TEXTES → INTRO → AUTRES
|
|
const orderedTypes = ['titre', 'texte', 'intro'];
|
|
|
|
for (const type of orderedTypes) {
|
|
const elements = groupedElements[type];
|
|
if (!elements || elements.length === 0) continue;
|
|
|
|
// DÉCOUPER EN CHUNKS DE MAX 4 ÉLÉMENTS POUR ÉVITER TIMEOUTS
|
|
const chunks = chunkArray(elements, 4);
|
|
logSh(`🚀 BATCH ${type.toUpperCase()}: ${elements.length} éléments en ${chunks.length} chunks`, 'INFO');
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
logSh(` Chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
try {
|
|
// Passer les résultats déjà générés pour contexte (titres → textes)
|
|
const batchPrompt = createBatchBasePrompt(chunk, type, csvData, results);
|
|
|
|
const batchResponse = await callLLM(aiProvider, batchPrompt, {
|
|
temperature: 0.7,
|
|
maxTokens: 2000 * chunk.length
|
|
}, csvData.personality);
|
|
|
|
const batchResults = parseBatchResponse(batchResponse, chunk);
|
|
Object.assign(results, batchResults);
|
|
|
|
logSh(`✅ Chunk ${chunkIndex + 1}: ${Object.keys(batchResults).length}/${chunk.length} éléments générés`, 'INFO');
|
|
|
|
} catch (error) {
|
|
logSh(`❌ FATAL: Chunk ${chunkIndex + 1} de ${type} échoué: ${error.message}`, 'ERROR');
|
|
throw new Error(`FATAL: Génération chunk ${chunkIndex + 1} de ${type} échouée - arrêt du workflow: ${error.message}`);
|
|
}
|
|
|
|
// Délai entre chunks pour éviter rate limiting
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
}
|
|
|
|
logSh(`✅ BATCH ${type.toUpperCase()} COMPLET: ${elements.length} éléments générés en ${chunks.length} chunks`, 'INFO');
|
|
}
|
|
|
|
// TRAITER les types restants (autres que titre/texte/intro)
|
|
for (const [type, elements] of Object.entries(groupedElements)) {
|
|
if (orderedTypes.includes(type) || elements.length === 0) continue;
|
|
|
|
// DÉCOUPER EN CHUNKS DE MAX 4 ÉLÉMENTS POUR ÉVITER TIMEOUTS
|
|
const chunks = chunkArray(elements, 4);
|
|
logSh(`🚀 BATCH ${type.toUpperCase()}: ${elements.length} éléments en ${chunks.length} chunks`, 'INFO');
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
logSh(` Chunk ${chunkIndex + 1}/${chunks.length}: ${chunk.length} éléments`, 'DEBUG');
|
|
|
|
try {
|
|
const batchPrompt = createBatchBasePrompt(chunk, type, csvData, results);
|
|
|
|
const batchResponse = await callLLM(aiProvider, batchPrompt, {
|
|
temperature: 0.7,
|
|
maxTokens: 2000 * chunk.length
|
|
}, csvData.personality);
|
|
|
|
const batchResults = parseBatchResponse(batchResponse, chunk);
|
|
Object.assign(results, batchResults);
|
|
|
|
logSh(`✅ Chunk ${chunkIndex + 1}: ${Object.keys(batchResults).length}/${chunk.length} éléments générés`, 'INFO');
|
|
|
|
} catch (error) {
|
|
logSh(`❌ FATAL: Chunk ${chunkIndex + 1} de ${type} échoué: ${error.message}`, 'ERROR');
|
|
throw new Error(`FATAL: Génération chunk ${chunkIndex + 1} de ${type} échouée - arrêt du workflow: ${error.message}`);
|
|
}
|
|
|
|
// Délai entre chunks
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
}
|
|
|
|
logSh(`✅ BATCH ${type.toUpperCase()} COMPLET: ${elements.length} éléments générés en ${chunks.length} chunks`, 'INFO');
|
|
}
|
|
|
|
// 3. GÉNÉRER les paires FAQ ensemble (RESTAURÉ depuis .gs)
|
|
if (faqPairs.length > 0) {
|
|
logSh(`🔍 === GÉNÉRATION PAIRES FAQ (${faqPairs.length} paires) ===`, 'INFO');
|
|
const faqResults = await generateFAQPairsRestored(faqPairs, csvData, aiProvider);
|
|
Object.assign(results, faqResults);
|
|
}
|
|
|
|
logSh(`🔍 === RÉSULTATS FINAUX GÉNÉRATION BASE ===`, 'DEBUG');
|
|
logSh(`🔍 Total généré: ${Object.keys(results).length} éléments`, 'DEBUG');
|
|
Object.keys(results).forEach(tag => {
|
|
logSh(`🔍 [${tag}]: "${results[tag]}"`, 'DEBUG');
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* ÉTAPE 2 - Enhancement technique ÉLÉMENT PAR ÉLÉMENT avec IA configurable
|
|
* NOUVEAU : Traitement individuel pour fiabilité maximale et debug précis
|
|
*/
|
|
async function enhanceAllTechnicalTerms(baseContents, csvData, aiProvider) {
|
|
logSh('🔧 === DÉBUT ENHANCEMENT TECHNIQUE ===', 'INFO');
|
|
logSh('Enhancement technique BATCH TOTAL...', 'DEBUG');
|
|
|
|
const allElements = Object.keys(baseContents);
|
|
if (allElements.length === 0) {
|
|
logSh('⚠️ Aucun élément à analyser techniquement', 'WARNING');
|
|
return baseContents;
|
|
}
|
|
|
|
const analysisStart = Date.now();
|
|
logSh(`📊 Analyse démarrée: ${allElements.length} éléments à examiner`, 'INFO');
|
|
|
|
try {
|
|
// ÉTAPE 1 : Extraction batch TOUS les termes techniques (1 seul appel)
|
|
logSh(`🔍 Analyse technique batch: ${allElements.length} éléments`, 'INFO');
|
|
const technicalAnalysis = await extractAllTechnicalTermsBatch(baseContents, csvData, aiProvider);
|
|
const analysisEnd = Date.now();
|
|
|
|
// ÉTAPE 2 : Enhancement batch TOUS les éléments qui en ont besoin (1 seul appel)
|
|
const elementsNeedingEnhancement = technicalAnalysis.filter(item => item.needsEnhancement);
|
|
|
|
logSh(`📋 Analyse terminée (${analysisEnd - analysisStart}ms):`, 'INFO');
|
|
logSh(` • ${elementsNeedingEnhancement.length}/${allElements.length} éléments nécessitent enhancement`, 'INFO');
|
|
|
|
if (elementsNeedingEnhancement.length === 0) {
|
|
logSh('✅ Aucun élément ne nécessite enhancement technique - contenu déjà optimal', 'INFO');
|
|
return baseContents;
|
|
}
|
|
|
|
// Log détaillé des éléments à améliorer
|
|
elementsNeedingEnhancement.forEach((item, i) => {
|
|
logSh(` ${i+1}. [${item.tag}]: ${item.technicalTerms.join(', ')}`, 'DEBUG');
|
|
});
|
|
|
|
const enhancementStart = Date.now();
|
|
logSh(`🔧 Enhancement technique: ${elementsNeedingEnhancement.length}/${allElements.length} éléments`, 'INFO');
|
|
const enhancedContents = await enhanceAllElementsTechnicalBatch(elementsNeedingEnhancement, csvData, aiProvider);
|
|
const enhancementEnd = Date.now();
|
|
|
|
// ÉTAPE 3 : Merger résultats
|
|
const results = { ...baseContents };
|
|
let actuallyEnhanced = 0;
|
|
Object.keys(enhancedContents).forEach(tag => {
|
|
if (enhancedContents[tag] !== baseContents[tag]) {
|
|
results[tag] = enhancedContents[tag];
|
|
actuallyEnhanced++;
|
|
}
|
|
});
|
|
|
|
logSh(`⚡ Enhancement terminé (${enhancementEnd - enhancementStart}ms):`, 'INFO');
|
|
logSh(` • ${actuallyEnhanced} éléments réellement améliorés`, 'INFO');
|
|
logSh(` • Termes intégrés: dibond, impression UV, fraisage, etc.`, 'DEBUG');
|
|
logSh(`✅ Enhancement technique terminé avec succès`, 'INFO');
|
|
return results;
|
|
|
|
} catch (error) {
|
|
const analysisTotal = Date.now() - analysisStart;
|
|
logSh(`❌ FATAL: Enhancement technique échoué après ${analysisTotal}ms`, 'ERROR');
|
|
logSh(`❌ Message: ${error.message}`, 'ERROR');
|
|
throw new Error(`FATAL: Enhancement technique impossible - arrêt du workflow: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Analyser un seul élément pour détecter les termes techniques
|
|
*/
|
|
async function analyzeSingleElementTechnicalTerms(tag, content, csvData, aiProvider) {
|
|
const prompt = `MISSION: Analyser ce contenu et déterminer s'il contient des termes techniques.
|
|
|
|
CONTEXTE: ${csvData.mc0} - Secteur: signalétique/impression
|
|
|
|
CONTENU À ANALYSER:
|
|
TAG: ${tag}
|
|
CONTENU: "${content}"
|
|
|
|
CONSIGNES:
|
|
- Cherche UNIQUEMENT des vrais termes techniques métier/industrie
|
|
- Évite mots génériques (qualité, service, pratique, personnalisé, etc.)
|
|
- Focus: matériaux, procédés, normes, dimensions, technologies spécifiques
|
|
|
|
EXEMPLES VALIDES: dibond, impression UV, fraisage CNC, épaisseur 3mm, aluminium brossé, anodisation
|
|
EXEMPLES INVALIDES: durable, pratique, personnalisé, moderne, esthétique, haute performance
|
|
|
|
RÉPONSE REQUISE:
|
|
- Si termes techniques trouvés: "OUI - termes: [liste des termes séparés par virgules]"
|
|
- Si aucun terme technique: "NON"
|
|
|
|
EXEMPLE:
|
|
OUI - termes: aluminium composite, impression numérique, gravure laser`;
|
|
|
|
try {
|
|
const response = await callLLM(aiProvider, prompt, { temperature: 0.3 });
|
|
|
|
if (response.toUpperCase().startsWith('OUI')) {
|
|
// Extraire les termes de la réponse
|
|
const termsMatch = response.match(/termes:\s*(.+)/i);
|
|
const terms = termsMatch ? termsMatch[1].trim() : '';
|
|
logSh(`✅ [${tag}] Termes techniques détectés: ${terms}`, 'DEBUG');
|
|
return true;
|
|
} else {
|
|
logSh(`⏭️ [${tag}] Pas de termes techniques`, 'DEBUG');
|
|
return false;
|
|
}
|
|
} catch (error) {
|
|
logSh(`❌ ERREUR analyse ${tag}: ${error.message}`, 'ERROR');
|
|
return false; // En cas d'erreur, on skip l'enhancement
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Enhancer un seul élément techniquement
|
|
*/
|
|
async function enhanceSingleElementTechnical(tag, content, csvData, aiProvider) {
|
|
const prompt = `MISSION: Améliore ce contenu en intégrant des termes techniques précis.
|
|
|
|
CONTEXTE: ${csvData.mc0} - Secteur: signalétique/impression
|
|
|
|
CONTENU À AMÉLIORER:
|
|
TAG: ${tag}
|
|
CONTENU: "${content}"
|
|
|
|
OBJECTIFS:
|
|
- Remplace les termes génériques par des termes techniques précis
|
|
- Ajoute des spécifications techniques réalistes
|
|
- Maintient le même style et longueur
|
|
- Intègre naturellement: matériaux (dibond, aluminium composite), procédés (impression UV, gravure laser), dimensions, normes
|
|
|
|
EXEMPLE DE TRANSFORMATION:
|
|
"matériaux haute performance" → "dibond 3mm ou aluminium composite"
|
|
"impression moderne" → "impression UV haute définition"
|
|
"fixation solide" → "fixation par chevilles inox Ø6mm"
|
|
|
|
CONTRAINTES:
|
|
- GARDE la même structure
|
|
- MÊME longueur approximative
|
|
- Style cohérent avec l'original
|
|
- RÉPONDS DIRECTEMENT par le contenu amélioré, sans préfixe`;
|
|
|
|
try {
|
|
const enhancedContent = await callLLM(aiProvider, prompt, { temperature: 0.7 });
|
|
return enhancedContent.trim();
|
|
} catch (error) {
|
|
logSh(`❌ ERREUR enhancement ${tag}: ${error.message}`, 'ERROR');
|
|
return content; // En cas d'erreur, on retourne le contenu original
|
|
}
|
|
}
|
|
|
|
// ANCIENNES FONCTIONS BATCH SUPPRIMÉES - REMPLACÉES PAR TRAITEMENT INDIVIDUEL
|
|
|
|
/**
|
|
* NOUVELLE FONCTION : Enhancement batch TOUS les éléments
|
|
*/
|
|
// FONCTION SUPPRIMÉE : enhanceAllElementsTechnicalBatch() - Remplacée par traitement individuel
|
|
|
|
/**
|
|
* ÉTAPE 3 - Enhancement transitions BATCH avec IA configurable
|
|
*/
|
|
async function enhanceAllTransitions(baseContents, csvData, aiProvider) {
|
|
logSh('🔗 === DÉBUT ENHANCEMENT TRANSITIONS ===', 'INFO');
|
|
logSh('Enhancement transitions batch...', 'DEBUG');
|
|
|
|
const transitionStart = Date.now();
|
|
const allElements = Object.keys(baseContents);
|
|
logSh(`📊 Analyse transitions: ${allElements.length} éléments à examiner`, 'INFO');
|
|
|
|
// Sélectionner éléments longs qui bénéficient d'amélioration transitions
|
|
const transitionElements = [];
|
|
let analyzedCount = 0;
|
|
Object.keys(baseContents).forEach(tag => {
|
|
const content = baseContents[tag];
|
|
analyzedCount++;
|
|
if (content.length > 150) {
|
|
const needsTransitions = analyzeTransitionNeed(content);
|
|
logSh(` [${tag}]: ${content.length}c, transitions=${needsTransitions ? '✅' : '❌'}`, 'DEBUG');
|
|
if (needsTransitions) {
|
|
transitionElements.push({
|
|
tag: tag,
|
|
content: content
|
|
});
|
|
}
|
|
} else {
|
|
logSh(` [${tag}]: ${content.length}c - trop court, ignoré`, 'DEBUG');
|
|
}
|
|
});
|
|
|
|
logSh(`📋 Analyse transitions terminée:`, 'INFO');
|
|
logSh(` • ${analyzedCount} éléments analysés`, 'INFO');
|
|
logSh(` • ${transitionElements.length} nécessitent amélioration`, 'INFO');
|
|
|
|
if (transitionElements.length === 0) {
|
|
logSh('✅ Pas d\'éléments nécessitant enhancement transitions - fluidité déjà optimale', 'INFO');
|
|
return baseContents;
|
|
}
|
|
|
|
logSh(`${transitionElements.length} éléments à améliorer (transitions)`, 'INFO');
|
|
|
|
const chunks = chunkArray(transitionElements, 6); // Plus petit pour Gemini
|
|
const results = { ...baseContents };
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(`Chunk transitions ${chunkIndex + 1}/${chunks.length} (${chunk.length} éléments)`, 'DEBUG');
|
|
|
|
const batchTransitionsPrompt = `MISSION: Améliore UNIQUEMENT les transitions et fluidité de ces contenus.
|
|
|
|
CONTEXTE: Article SEO professionnel pour site web commercial
|
|
PERSONNALITÉ: ${csvData.personality?.nom} (${csvData.personality?.style} adapté web)
|
|
CONNECTEURS PRÉFÉRÉS: ${csvData.personality?.connecteursPref}
|
|
|
|
CONTENUS:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
"${item.content}"`).join('\n\n')}
|
|
|
|
OBJECTIFS:
|
|
- Connecteurs plus naturels et variés issus de: ${csvData.personality?.connecteursPref}
|
|
- Transitions fluides entre idées
|
|
- ÉVITE répétitions excessives ("franchement", "du coup", "vraiment", "par ailleurs")
|
|
- Style cohérent ${csvData.personality?.style}
|
|
|
|
CONTRAINTES STRICTES:
|
|
- NE CHANGE PAS le fond du message
|
|
- GARDE la même structure et longueur approximative
|
|
- Améliore SEULEMENT la fluidité des transitions
|
|
- RESPECTE le style ${csvData.personality?.nom}
|
|
- RÉPONDS DIRECTEMENT PAR LE CONTENU AMÉLIORÉ, sans préfixe ni tag XML
|
|
|
|
FORMAT DE RÉPONSE:
|
|
[1] Contenu avec transitions améliorées selon ${csvData.personality?.nom}
|
|
[2] Contenu avec transitions améliorées selon ${csvData.personality?.nom}
|
|
etc...`;
|
|
|
|
const improved = await callLLM(aiProvider, batchTransitionsPrompt, {
|
|
temperature: 0.6,
|
|
maxTokens: 2500
|
|
}, csvData.personality);
|
|
|
|
const parsedImprovements = parseTransitionsBatchResponse(improved, chunk);
|
|
|
|
Object.keys(parsedImprovements).forEach(tag => {
|
|
results[tag] = parsedImprovements[tag];
|
|
});
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur chunk transitions ${chunkIndex + 1}: ${error.message}`, 'ERROR');
|
|
}
|
|
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* ÉTAPE 4 - Enhancement style personnalité BATCH avec IA configurable
|
|
*/
|
|
async function enhanceAllPersonalityStyle(baseContents, csvData, aiProvider) {
|
|
const personality = csvData.personality;
|
|
if (!personality) {
|
|
logSh('Pas de personnalité, skip enhancement style', 'DEBUG');
|
|
return baseContents;
|
|
}
|
|
|
|
logSh(`Enhancement style ${personality.nom} batch...`, 'DEBUG');
|
|
|
|
// Tous les éléments bénéficient de l'adaptation personnalité
|
|
const styleElements = Object.keys(baseContents).map(tag => ({
|
|
tag: tag,
|
|
content: baseContents[tag]
|
|
}));
|
|
|
|
const chunks = chunkArray(styleElements, 8);
|
|
const results = { ...baseContents };
|
|
|
|
for (let chunkIndex = 0; chunkIndex < chunks.length; chunkIndex++) {
|
|
const chunk = chunks[chunkIndex];
|
|
|
|
try {
|
|
logSh(`Chunk style ${chunkIndex + 1}/${chunks.length} (${chunk.length} éléments)`, 'DEBUG');
|
|
|
|
const batchStylePrompt = `MISSION: Adapte UNIQUEMENT le style de ces contenus selon ${personality.nom}.
|
|
|
|
CONTEXTE: Finalisation article SEO pour site e-commerce professionnel
|
|
PERSONNALITÉ: ${personality.nom}
|
|
DESCRIPTION: ${personality.description}
|
|
STYLE CIBLE: ${personality.style} adapté au web professionnel
|
|
VOCABULAIRE: ${personality.vocabulairePref}
|
|
CONNECTEURS: ${personality.connecteursPref}
|
|
NIVEAU TECHNIQUE: ${personality.niveauTechnique}
|
|
LONGUEUR PHRASES: ${personality.longueurPhrases}
|
|
|
|
CONTENUS À STYLISER:
|
|
|
|
${chunk.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
"${item.content}"`).join('\n\n')}
|
|
|
|
CONSIGNES STRICTES:
|
|
- GARDE le même contenu informatif et technique
|
|
- Adapte SEULEMENT le ton, les expressions et le vocabulaire selon ${personality.nom}
|
|
- RESPECTE la longueur approximative (même nombre de mots ±20%)
|
|
- ÉVITE les répétitions excessives ("franchement", "du coup", "vraiment")
|
|
- VARIE les expressions et connecteurs selon: ${personality.connecteursPref}
|
|
- Style ${personality.nom} reconnaissable mais NATUREL
|
|
- RÉPONDS DIRECTEMENT PAR LE CONTENU STYLISÉ, sans préfixe ni tag XML
|
|
- PAS de messages d'excuse ou d'incapacité
|
|
|
|
FORMAT DE RÉPONSE:
|
|
[1] Contenu stylisé selon ${personality.nom} (${personality.style})
|
|
[2] Contenu stylisé selon ${personality.nom} (${personality.style})
|
|
etc...`;
|
|
|
|
const styled = await callLLM(aiProvider, batchStylePrompt, {
|
|
temperature: 0.8,
|
|
maxTokens: 3000
|
|
}, personality);
|
|
|
|
const parsedStyles = parseStyleBatchResponse(styled, chunk);
|
|
|
|
Object.keys(parsedStyles).forEach(tag => {
|
|
results[tag] = parsedStyles[tag];
|
|
});
|
|
|
|
} catch (error) {
|
|
logSh(`❌ Erreur chunk style ${chunkIndex + 1}: ${error.message}`, 'ERROR');
|
|
}
|
|
|
|
if (chunkIndex < chunks.length - 1) {
|
|
await sleep(1500);
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= HELPER FUNCTIONS =============
|
|
|
|
/**
|
|
* Sleep function replacement for Utilities.sleep
|
|
*/
|
|
|
|
// FONCTION SUPPRIMÉE : sleep() dupliquée - déjà définie ligne 12
|
|
|
|
/**
|
|
* RESTAURÉ DEPUIS .GS : Génération des paires FAQ cohérentes
|
|
*/
|
|
async function generateFAQPairsRestored(faqPairs, csvData, aiProvider) {
|
|
logSh(`🔍 === GÉNÉRATION PAIRES FAQ (logique .gs restaurée) ===`, 'INFO');
|
|
|
|
if (faqPairs.length === 0) return {};
|
|
|
|
const batchPrompt = createBatchFAQPairsPrompt(faqPairs, csvData);
|
|
logSh(`🔍 Prompt FAQ paires (${batchPrompt.length} chars): "${batchPrompt.substring(0, 300)}..."`, 'DEBUG');
|
|
|
|
try {
|
|
const batchResponse = await callLLM(aiProvider, batchPrompt, {
|
|
temperature: 0.8,
|
|
maxTokens: 3000 // Plus large pour les paires
|
|
}, csvData.personality);
|
|
|
|
logSh(`🔍 Réponse FAQ paires reçue: ${batchResponse.length} caractères`, 'DEBUG');
|
|
logSh(`🔍 Début réponse: "${batchResponse.substring(0, 200)}..."`, 'DEBUG');
|
|
|
|
return parseFAQPairsResponse(batchResponse, faqPairs);
|
|
|
|
} catch (error) {
|
|
logSh(`❌ FATAL: Erreur génération paires FAQ: ${error.message}`, 'ERROR');
|
|
throw new Error(`FATAL: Génération paires FAQ échouée - arrêt du workflow: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* RESTAURÉ DEPUIS .GS : Prompt pour paires FAQ cohérentes
|
|
*/
|
|
function createBatchFAQPairsPrompt(faqPairs, csvData) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `=== 1. CONTEXTE ===
|
|
Entreprise: Autocollant.fr - signalétique personnalisée
|
|
Sujet: ${csvData.mc0}
|
|
Section: FAQ pour article SEO commercial
|
|
|
|
=== 2. PERSONNALITÉ ===
|
|
Rédacteur: ${personality.nom}
|
|
Style: ${personality.style}
|
|
Ton: ${personality.description || 'professionnel'}
|
|
|
|
=== 3. RÈGLES GÉNÉRALES ===
|
|
- Questions naturelles de clients
|
|
- Réponses expertes et rassurantes
|
|
- Langage professionnel mais accessible
|
|
- Textes rédigés humainement et de façon authentique
|
|
- Couvrir: prix, livraison, personnalisation, installation, durabilité
|
|
- IMPÉRATIF: Respecter strictement les contraintes XML
|
|
|
|
=== 4. PAIRES FAQ À GÉNÉRER ===
|
|
|
|
`;
|
|
|
|
faqPairs.forEach((pair, index) => {
|
|
const questionTag = pair.question.tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '');
|
|
const answerTag = pair.answer.tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '');
|
|
|
|
prompt += `${index + 1}. [${questionTag}] + [${answerTag}] - Paire FAQ naturelle
|
|
`;
|
|
});
|
|
|
|
prompt += `
|
|
|
|
FORMAT DE RÉPONSE:
|
|
PAIRE 1:
|
|
[${faqPairs[0].question.tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '')}]
|
|
Question client directe et naturelle sur ${csvData.mc0} ?
|
|
|
|
[${faqPairs[0].answer.tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '')}]
|
|
Réponse utile et rassurante selon le style ${personality.style} de ${personality.nom}.
|
|
`;
|
|
|
|
if (faqPairs.length > 1) {
|
|
prompt += `PAIRE 2:
|
|
etc...
|
|
`;
|
|
}
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* RESTAURÉ DEPUIS .GS : Parser réponse paires FAQ
|
|
*/
|
|
function parseFAQPairsResponse(response, faqPairs) {
|
|
const results = {};
|
|
|
|
logSh(`🔍 Parsing FAQ paires: "${response.substring(0, 300)}..."`, 'DEBUG');
|
|
|
|
// Parser avec regex [TAG] contenu
|
|
const regex = /\[([^\]]+)\]\s*([^[]*?)(?=\[|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
let content = match[2].trim().replace(/\n\s*\n/g, '\n').replace(/^\n+|\n+$/g, '');
|
|
|
|
// NOUVEAU: Appliquer le nettoyage XML pour FAQ aussi
|
|
content = cleanXMLTagsFromContent(content);
|
|
|
|
if (content && content.length > 0) {
|
|
parsedItems[tag] = content;
|
|
logSh(`🔍 Parsé [${tag}]: "${content.substring(0, 100)}..."`, 'DEBUG');
|
|
}
|
|
}
|
|
|
|
// Mapper aux vrais tags FAQ avec |
|
|
let pairesCompletes = 0;
|
|
faqPairs.forEach(pair => {
|
|
const questionCleanTag = pair.question.tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '');
|
|
const answerCleanTag = pair.answer.tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '');
|
|
|
|
const questionContent = parsedItems[questionCleanTag];
|
|
const answerContent = parsedItems[answerCleanTag];
|
|
|
|
if (questionContent && answerContent) {
|
|
results[pair.question.tag] = questionContent;
|
|
results[pair.answer.tag] = answerContent;
|
|
pairesCompletes++;
|
|
logSh(`✅ Paire FAQ ${pair.number} complète: Q="${questionContent}" R="${answerContent.substring(0, 50)}..."`, 'INFO');
|
|
} else {
|
|
logSh(`⚠️ Paire FAQ ${pair.number} incomplète: Q=${!!questionContent} R=${!!answerContent}`, 'WARNING');
|
|
|
|
if (questionContent) results[pair.question.tag] = questionContent;
|
|
if (answerContent) results[pair.answer.tag] = answerContent;
|
|
}
|
|
});
|
|
|
|
logSh(`📊 FAQ parsing: ${pairesCompletes}/${faqPairs.length} paires complètes`, 'INFO');
|
|
|
|
// FATAL si aucune paire complète (comme dans le .gs)
|
|
if (pairesCompletes === 0 && faqPairs.length > 0) {
|
|
logSh(`❌ FATAL: Aucune paire FAQ générée correctement`, 'ERROR');
|
|
throw new Error(`FATAL: Génération FAQ incomplète (0/${faqPairs.length} paires complètes) - arrêt du workflow`);
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* RESTAURÉ DEPUIS .GS : Nettoyer instructions FAQ
|
|
*/
|
|
function cleanFAQInstructions(instructions, csvData) {
|
|
if (!instructions) return '';
|
|
|
|
let cleanInstructions = instructions;
|
|
|
|
// Remplacer variables
|
|
cleanInstructions = cleanInstructions.replace(/\{\{T0\}\}/g, csvData.t0 || '');
|
|
cleanInstructions = cleanInstructions.replace(/\{\{MC0\}\}/g, csvData.mc0 || '');
|
|
cleanInstructions = cleanInstructions.replace(/\{\{T-1\}\}/g, csvData.tMinus1 || '');
|
|
cleanInstructions = cleanInstructions.replace(/\{\{L-1\}\}/g, csvData.lMinus1 || '');
|
|
|
|
// Variables multiples MC+1_X, T+1_X, L+1_X
|
|
if (csvData.mcPlus1) {
|
|
const mcPlus1 = csvData.mcPlus1.split(',').map(s => s.trim());
|
|
for (let i = 1; i <= 6; i++) {
|
|
const mcValue = mcPlus1[i-1] || `[MC+1_${i} non défini]`;
|
|
cleanInstructions = cleanInstructions.replace(new RegExp(`\\{\\{MC\\+1_${i}\\}\\}`, 'g'), mcValue);
|
|
}
|
|
}
|
|
|
|
if (csvData.tPlus1) {
|
|
const tPlus1 = csvData.tPlus1.split(',').map(s => s.trim());
|
|
for (let i = 1; i <= 6; i++) {
|
|
const tValue = tPlus1[i-1] || `[T+1_${i} non défini]`;
|
|
cleanInstructions = cleanInstructions.replace(new RegExp(`\\{\\{T\\+1_${i}\\}\\}`, 'g'), tValue);
|
|
}
|
|
}
|
|
|
|
// Nettoyer HTML
|
|
cleanInstructions = cleanInstructions.replace(/<\/?[^>]+>/g, '');
|
|
cleanInstructions = cleanInstructions.replace(/\s+/g, ' ').trim();
|
|
|
|
return cleanInstructions;
|
|
}
|
|
|
|
/**
|
|
* Collecter tous les éléments dans l'ordre XML original
|
|
* CORRECTION: Suit l'ordre séquentiel XML au lieu de grouper par section
|
|
*/
|
|
function collectAllElements(hierarchy) {
|
|
const allElements = [];
|
|
const tagToElementMap = {};
|
|
|
|
// 1. Créer un mapping de tous les éléments disponibles
|
|
Object.keys(hierarchy).forEach(path => {
|
|
const section = hierarchy[path];
|
|
|
|
if (section.title) {
|
|
tagToElementMap[section.title.originalElement.originalTag] = {
|
|
tag: section.title.originalElement.originalTag,
|
|
element: section.title.originalElement,
|
|
type: 'titre'
|
|
};
|
|
}
|
|
|
|
if (section.text) {
|
|
tagToElementMap[section.text.originalElement.originalTag] = {
|
|
tag: section.text.originalElement.originalTag,
|
|
element: section.text.originalElement,
|
|
type: 'texte'
|
|
};
|
|
}
|
|
|
|
section.questions.forEach(q => {
|
|
tagToElementMap[q.originalElement.originalTag] = {
|
|
tag: q.originalElement.originalTag,
|
|
element: q.originalElement,
|
|
type: q.originalElement.type
|
|
};
|
|
});
|
|
});
|
|
|
|
// 2. Récupérer l'ordre XML original depuis le template global
|
|
logSh(`🔍 Global XML Template disponible: ${!!global.currentXmlTemplate}`, 'DEBUG');
|
|
if (global.currentXmlTemplate && global.currentXmlTemplate.length > 0) {
|
|
logSh(`🔍 Template XML: ${global.currentXmlTemplate.substring(0, 200)}...`, 'DEBUG');
|
|
const regex = /\|([^|]+)\|/g;
|
|
let match;
|
|
|
|
// Parcourir le XML dans l'ordre d'apparition
|
|
while ((match = regex.exec(global.currentXmlTemplate)) !== null) {
|
|
const fullMatch = match[1];
|
|
|
|
// Extraire le nom du tag (sans variables)
|
|
const nameMatch = fullMatch.match(/^([^{]+)/);
|
|
const tagName = nameMatch ? nameMatch[1].trim() : fullMatch.split('{')[0];
|
|
const pureTag = `|${tagName}|`;
|
|
|
|
// Si cet élément existe dans notre mapping, l'ajouter dans l'ordre
|
|
if (tagToElementMap[pureTag]) {
|
|
allElements.push(tagToElementMap[pureTag]);
|
|
logSh(`🔍 Ajouté dans l'ordre: ${pureTag}`, 'DEBUG');
|
|
delete tagToElementMap[pureTag]; // Éviter les doublons
|
|
} else {
|
|
logSh(`🔍 Tag XML non trouvé dans mapping: ${pureTag}`, 'DEBUG');
|
|
}
|
|
}
|
|
}
|
|
|
|
// 3. Ajouter les éléments restants (sécurité)
|
|
const remainingElements = Object.values(tagToElementMap);
|
|
if (remainingElements.length > 0) {
|
|
logSh(`🔍 Éléments restants ajoutés: ${remainingElements.map(el => el.tag).join(', ')}`, 'DEBUG');
|
|
remainingElements.forEach(element => {
|
|
allElements.push(element);
|
|
});
|
|
}
|
|
|
|
logSh(`🔍 ORDRE FINAL: ${allElements.map(el => el.tag.replace(/\|/g, '')).join(' → ')}`, 'INFO');
|
|
|
|
return allElements;
|
|
}
|
|
|
|
/**
|
|
* RESTAURÉ DEPUIS .GS : Séparer les paires FAQ des autres éléments
|
|
*/
|
|
function separateFAQPairsAndOthers(allElements) {
|
|
const faqPairs = [];
|
|
const otherElements = [];
|
|
const faqQuestions = {};
|
|
const faqAnswers = {};
|
|
|
|
// 1. Collecter toutes les questions et réponses FAQ
|
|
allElements.forEach(element => {
|
|
if (element.type === 'faq_question') {
|
|
// Extraire le numéro : |Faq_q_1| → 1
|
|
const numberMatch = element.tag.match(/(\d+)/);
|
|
const faqNumber = numberMatch ? numberMatch[1] : '1';
|
|
faqQuestions[faqNumber] = element;
|
|
logSh(`🔍 Question FAQ ${faqNumber} trouvée: ${element.tag}`, 'DEBUG');
|
|
} else if (element.type === 'faq_reponse') {
|
|
// Extraire le numéro : |Faq_a_1| → 1
|
|
const numberMatch = element.tag.match(/(\d+)/);
|
|
const faqNumber = numberMatch ? numberMatch[1] : '1';
|
|
faqAnswers[faqNumber] = element;
|
|
logSh(`🔍 Réponse FAQ ${faqNumber} trouvée: ${element.tag}`, 'DEBUG');
|
|
} else {
|
|
// Élément normal (titre, texte, intro, etc.)
|
|
otherElements.push(element);
|
|
}
|
|
});
|
|
|
|
// 2. Créer les paires FAQ cohérentes
|
|
Object.keys(faqQuestions).forEach(number => {
|
|
const question = faqQuestions[number];
|
|
const answer = faqAnswers[number];
|
|
|
|
if (question && answer) {
|
|
faqPairs.push({
|
|
number: number,
|
|
question: question,
|
|
answer: answer
|
|
});
|
|
logSh(`✅ Paire FAQ ${number} créée: ${question.tag} + ${answer.tag}`, 'INFO');
|
|
} else if (question) {
|
|
logSh(`⚠️ Question FAQ ${number} sans réponse correspondante`, 'WARNING');
|
|
otherElements.push(question); // Traiter comme élément individuel
|
|
} else if (answer) {
|
|
logSh(`⚠️ Réponse FAQ ${number} sans question correspondante`, 'WARNING');
|
|
otherElements.push(answer); // Traiter comme élément individuel
|
|
}
|
|
});
|
|
|
|
logSh(`🔍 Séparation terminée: ${faqPairs.length} paires FAQ, ${otherElements.length} autres éléments`, 'INFO');
|
|
|
|
return { faqPairs, otherElements };
|
|
}
|
|
|
|
/**
|
|
* Grouper éléments par type
|
|
*/
|
|
function groupElementsByType(elements) {
|
|
const groups = {};
|
|
|
|
elements.forEach(element => {
|
|
const type = element.type;
|
|
if (!groups[type]) {
|
|
groups[type] = [];
|
|
}
|
|
groups[type].push(element);
|
|
});
|
|
|
|
return groups;
|
|
}
|
|
|
|
/**
|
|
* Diviser array en chunks
|
|
*/
|
|
function chunkArray(array, size) {
|
|
const chunks = [];
|
|
for (let i = 0; i < array.length; i += size) {
|
|
chunks.push(array.slice(i, i + size));
|
|
}
|
|
return chunks;
|
|
}
|
|
|
|
/**
|
|
* Trouver le titre associé à un élément texte
|
|
*/
|
|
function findAssociatedTitle(textElement, existingResults) {
|
|
const textName = textElement.element.name || textElement.tag;
|
|
|
|
// STRATÉGIE 1: Correspondance directe (Txt_H2_1 → Titre_H2_1)
|
|
const directMatch = textName.replace(/Txt_/, 'Titre_').replace(/Text_/, 'Titre_');
|
|
const directTitle = existingResults[`|${directMatch}|`] || existingResults[directMatch];
|
|
if (directTitle) return directTitle;
|
|
|
|
// STRATÉGIE 2: Même niveau hiérarchique (H2, H3)
|
|
const levelMatch = textName.match(/(H\d)_(\d+)/);
|
|
if (levelMatch) {
|
|
const [, level, number] = levelMatch;
|
|
const titleTag = `Titre_${level}_${number}`;
|
|
const levelTitle = existingResults[`|${titleTag}|`] || existingResults[titleTag];
|
|
if (levelTitle) return levelTitle;
|
|
}
|
|
|
|
// STRATÉGIE 3: Proximité dans l'ordre (texte suivant un titre)
|
|
const allTitles = Object.entries(existingResults)
|
|
.filter(([tag]) => tag.includes('Titre'))
|
|
.sort(([a], [b]) => a.localeCompare(b));
|
|
|
|
if (allTitles.length > 0) {
|
|
// Retourner le premier titre disponible comme contexte général
|
|
return allTitles[0][1];
|
|
}
|
|
|
|
return null;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt batch de base
|
|
*/
|
|
function createBatchBasePrompt(elements, type, csvData, existingResults = {}) {
|
|
const personality = csvData.personality;
|
|
|
|
let prompt = `=== 1. CONTEXTE ===
|
|
Entreprise: Autocollant.fr - signalétique personnalisée
|
|
Sujet: ${csvData.mc0}
|
|
Type d'article: SEO professionnel pour site commercial
|
|
|
|
=== 2. PERSONNALITÉ ===
|
|
Rédacteur: ${personality.nom}
|
|
Style: ${personality.style}
|
|
Ton: ${personality.description || 'professionnel'}
|
|
|
|
=== 3. RÈGLES GÉNÉRALES ===
|
|
- Contenu SEO optimisé
|
|
- Langage naturel et fluide
|
|
- Éviter répétitions
|
|
- Pas de références techniques dans le contenu
|
|
- Textes rédigés humainement et de façon authentique
|
|
- IMPÉRATIF: Respecter strictement les contraintes XML (nombre de mots, etc.)
|
|
|
|
=== 4. ÉLÉMENTS À GÉNÉRER ===
|
|
`;
|
|
|
|
// AJOUTER CONTEXTE DES TITRES POUR LES TEXTES
|
|
if (type === 'texte' && Object.keys(existingResults).length > 0) {
|
|
const generatedTitles = Object.entries(existingResults)
|
|
.filter(([tag]) => tag.includes('Titre'))
|
|
.map(([tag, title]) => `• ${tag.replace(/\|/g, '')}: "${title}"`)
|
|
.slice(0, 5); // Limiter à 5 titres pour éviter surcharge
|
|
|
|
if (generatedTitles.length > 0) {
|
|
prompt += `
|
|
Titres existants pour contexte:
|
|
${generatedTitles.join('\n')}
|
|
|
|
`;
|
|
}
|
|
}
|
|
|
|
elements.forEach((elementInfo, index) => {
|
|
const cleanTag = elementInfo.tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '');
|
|
|
|
prompt += `${index + 1}. [${cleanTag}] `;
|
|
|
|
// INSTRUCTIONS PROPRES PAR ÉLÉMENT
|
|
if (type === 'titre') {
|
|
if (elementInfo.element.type === 'titre_h1') {
|
|
prompt += `Titre principal accrocheur\n`;
|
|
} else if (elementInfo.element.type === 'titre_h2') {
|
|
prompt += `Titre de section engageant\n`;
|
|
} else if (elementInfo.element.type === 'titre_h3') {
|
|
prompt += `Sous-titre spécialisé\n`;
|
|
} else {
|
|
prompt += `Titre pertinent\n`;
|
|
}
|
|
} else if (type === 'texte') {
|
|
prompt += `Paragraphe informatif\n`;
|
|
|
|
// ASSOCIER LE TITRE CORRESPONDANT AUTOMATIQUEMENT
|
|
const associatedTitle = findAssociatedTitle(elementInfo, existingResults);
|
|
if (associatedTitle) {
|
|
prompt += ` Contexte: "${associatedTitle}"\n`;
|
|
}
|
|
|
|
if (elementInfo.element.resolvedContent) {
|
|
prompt += ` Angle: "${elementInfo.element.resolvedContent}"\n`;
|
|
}
|
|
} else if (type === 'intro') {
|
|
prompt += `Introduction engageante\n`;
|
|
} else {
|
|
prompt += `Contenu pertinent\n`;
|
|
}
|
|
});
|
|
|
|
prompt += `\nSTYLE ${personality.nom.toUpperCase()} - ${personality.style}:
|
|
- Vocabulaire: ${personality.vocabulairePref}
|
|
- Connecteurs: ${personality.connecteursPref}
|
|
- Phrases: ${personality.longueurPhrases}
|
|
- Niveau technique: ${personality.niveauTechnique}
|
|
|
|
CONSIGNES STRICTES POUR ARTICLE SEO:
|
|
- CONTEXTE: Article professionnel pour site e-commerce, destiné aux clients potentiels
|
|
- STYLE: ${personality.style} de ${personality.nom} mais ADAPTÉ au web professionnel
|
|
- INTERDICTION ABSOLUE: expressions trop familières répétées ("du coup", "bon", "franchement", "nickel", "tip-top")
|
|
- VOCABULAIRE: Mélange expertise technique + accessibilité client
|
|
- SEO: Utilise naturellement "${csvData.mc0}" et termes associés
|
|
- POUR LES TITRES: Titre SEO attractif UNIQUEMENT, JAMAIS "Titre_H1_1" ou "Titre_H2_7"
|
|
- EXEMPLE TITRE: "Plaques personnalisées résistantes : guide complet 2024"
|
|
- CONTENU: Informatif, rassurant, incite à l'achat SANS être trop commercial
|
|
- RÉPONDS DIRECTEMENT par le contenu web demandé, SANS préfixe
|
|
|
|
FORMAT DE RÉPONSE ${type === 'titre' ? '(TITRES UNIQUEMENT)' : ''}:
|
|
[${elements[0].tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '')}]
|
|
${type === 'titre' ? 'Titre réel et attractif (PAS "Titre_H1_1")' : 'Contenu rédigé selon le style ' + personality.nom}
|
|
|
|
[${elements[1] ? elements[1].tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '') : 'element2'}]
|
|
${type === 'titre' ? 'Titre réel et attractif (PAS "Titre_H2_1")' : 'Contenu rédigé selon le style ' + personality.nom}
|
|
|
|
etc...`;
|
|
|
|
return prompt;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse batch générique avec nettoyage des tags XML
|
|
*/
|
|
function parseBatchResponse(response, elements) {
|
|
const results = {};
|
|
|
|
// Parser avec regex [TAG] contenu
|
|
const regex = /\[([^\]]+)\]\s*\n([^[]*?)(?=\n\[|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const tag = match[1].trim();
|
|
let content = match[2].trim();
|
|
|
|
// NOUVEAU: Nettoyer les tags XML qui peuvent apparaître dans le contenu
|
|
content = cleanXMLTagsFromContent(content);
|
|
|
|
parsedItems[tag] = content;
|
|
}
|
|
|
|
// Mapper aux vrais tags avec |
|
|
elements.forEach(element => {
|
|
const cleanTag = element.tag.replace(/\|/g, '').replace(/[{}]/g, '').replace(/<\/?strong>/g, '');
|
|
|
|
if (parsedItems[cleanTag] && parsedItems[cleanTag].length > 10) {
|
|
results[element.tag] = parsedItems[cleanTag];
|
|
logSh(`✅ Parsé [${cleanTag}]: "${parsedItems[cleanTag].substring(0, 100)}..."`, 'DEBUG');
|
|
} else {
|
|
// Fallback si parsing échoue ou contenu trop court
|
|
results[element.tag] = `Contenu professionnel pour ${element.element.name}`;
|
|
logSh(`⚠️ Fallback [${cleanTag}]: parsing échoué ou contenu invalide`, 'WARNING');
|
|
}
|
|
});
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* NOUVELLE FONCTION: Nettoyer les tags XML du contenu généré
|
|
*/
|
|
function cleanXMLTagsFromContent(content) {
|
|
if (!content) return content;
|
|
|
|
// Supprimer les tags XML avec **
|
|
content = content.replace(/\*\*[^*]+\*\*/g, '');
|
|
|
|
// Supprimer les préfixes de titres indésirables
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?(voici\s+le\s+topo\s+pour\s+)?Titre_[HU]\d+_\d+[.,\s]*/gi, '');
|
|
content = content.replace(/^(Bon,?\s*)?(alors,?\s*)?pour\s+Titre_[HU]\d+_\d+[.,\s]*/gi, '');
|
|
content = content.replace(/^(Bon,?\s*)?(donc,?\s*)?Titre_[HU]\d+_\d+[.,\s]*/gi, '');
|
|
|
|
// Supprimer les messages d'excuse
|
|
content = content.replace(/Oh là là,?\s*je\s*(suis\s*)?(\w+\s*)?désolée?\s*,?\s*mais\s*je\s*n'ai\s*pas\s*l'information.*?(?=\.|$)/gi, '');
|
|
content = content.replace(/Bon,?\s*passons\s*au\s*suivant.*?(?=\.|$)/gi, '');
|
|
content = content.replace(/je\s*ne\s*sais\s*pas\s*quoi\s*vous\s*dire.*?(?=\.|$)/gi, '');
|
|
content = content.replace(/encore\s*un\s*point\s*où\s*je\s*n'ai\s*pas\s*l'information.*?(?=\.|$)/gi, '');
|
|
|
|
// Réduire les répétitions excessives d'expressions familières
|
|
content = content.replace(/(du coup[,\s]+){3,}/gi, 'du coup ');
|
|
content = content.replace(/(bon[,\s]+){3,}/gi, 'bon ');
|
|
content = content.replace(/(franchement[,\s]+){3,}/gi, 'franchement ');
|
|
content = content.replace(/(alors[,\s]+){3,}/gi, 'alors ');
|
|
content = content.replace(/(nickel[,\s]+){2,}/gi, 'nickel ');
|
|
content = content.replace(/(tip-top[,\s]+){2,}/gi, 'tip-top ');
|
|
content = content.replace(/(costaud[,\s]+){2,}/gi, 'costaud ');
|
|
|
|
// Nettoyer espaces multiples et retours ligne
|
|
content = content.replace(/\s{2,}/g, ' ');
|
|
content = content.replace(/\n{2,}/g, '\n');
|
|
content = content.trim();
|
|
|
|
return content;
|
|
}
|
|
|
|
// ============= PARSING FUNCTIONS =============
|
|
|
|
// FONCTION SUPPRIMÉE : parseAllTechnicalTermsResponse() - Parser batch défaillant remplacé par traitement individuel
|
|
|
|
// FONCTIONS SUPPRIMÉES : parseTechnicalEnhancementBatchResponse() et parseTechnicalBatchResponse() - Remplacées par traitement individuel
|
|
|
|
// Placeholder pour les fonctions de parsing conservées qui suivent
|
|
|
|
function parseTransitionsBatchResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let content = match[2].trim();
|
|
|
|
// Appliquer le nettoyage XML
|
|
content = cleanXMLTagsFromContent(content);
|
|
|
|
if (content && content.length > 10) {
|
|
results[chunk[index].tag] = content;
|
|
} else {
|
|
// Fallback si contenu invalide
|
|
results[chunk[index].tag] = chunk[index].content; // Garder contenu original
|
|
}
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
function parseStyleBatchResponse(response, chunk) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\n\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < chunk.length) {
|
|
let content = match[2].trim();
|
|
|
|
// Appliquer le nettoyage XML
|
|
content = cleanXMLTagsFromContent(content);
|
|
|
|
if (content && content.length > 10) {
|
|
results[chunk[index].tag] = content;
|
|
} else {
|
|
// Fallback si contenu invalide
|
|
results[chunk[index].tag] = chunk[index].content; // Garder contenu original
|
|
}
|
|
index++;
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= ANALYSIS FUNCTIONS =============
|
|
|
|
/**
|
|
* Analyser besoin d'amélioration transitions
|
|
*/
|
|
function analyzeTransitionNeed(content) {
|
|
const sentences = content.split(/[.!?]+/).filter(s => s.trim().length > 10);
|
|
|
|
// Critères multiples d'analyse
|
|
const metrics = {
|
|
repetitiveConnectors: analyzeRepetitiveConnectors(content),
|
|
abruptTransitions: analyzeAbruptTransitions(sentences),
|
|
sentenceVariety: analyzeSentenceVariety(sentences),
|
|
formalityLevel: analyzeFormalityLevel(content),
|
|
overallLength: content.length
|
|
};
|
|
|
|
// Score de besoin (0-1)
|
|
let needScore = 0;
|
|
needScore += metrics.repetitiveConnectors * 0.3;
|
|
needScore += metrics.abruptTransitions * 0.4;
|
|
needScore += (1 - metrics.sentenceVariety) * 0.2;
|
|
needScore += metrics.formalityLevel * 0.1;
|
|
|
|
// Seuil ajustable selon longueur
|
|
const threshold = metrics.overallLength > 300 ? 0.4 : 0.6;
|
|
|
|
logSh(`🔍 Analyse transitions: score=${needScore.toFixed(2)}, seuil=${threshold}`, 'DEBUG');
|
|
|
|
return needScore > threshold;
|
|
}
|
|
|
|
function analyzeRepetitiveConnectors(content) {
|
|
const connectors = ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc'];
|
|
let totalConnectors = 0;
|
|
let repetitions = 0;
|
|
|
|
connectors.forEach(connector => {
|
|
const matches = (content.match(new RegExp(`\\b${connector}\\b`, 'gi')) || []);
|
|
totalConnectors += matches.length;
|
|
if (matches.length > 1) repetitions += matches.length - 1;
|
|
});
|
|
|
|
return totalConnectors > 0 ? repetitions / totalConnectors : 0;
|
|
}
|
|
|
|
function analyzeAbruptTransitions(sentences) {
|
|
if (sentences.length < 2) return 0;
|
|
|
|
let abruptCount = 0;
|
|
|
|
for (let i = 1; i < sentences.length; i++) {
|
|
const current = sentences[i].trim();
|
|
const previous = sentences[i-1].trim();
|
|
|
|
const hasConnector = hasTransitionWord(current);
|
|
const topicContinuity = calculateTopicContinuity(previous, current);
|
|
|
|
// Transition abrupte = pas de connecteur + faible continuité thématique
|
|
if (!hasConnector && topicContinuity < 0.3) {
|
|
abruptCount++;
|
|
}
|
|
}
|
|
|
|
return abruptCount / (sentences.length - 1);
|
|
}
|
|
|
|
function analyzeSentenceVariety(sentences) {
|
|
if (sentences.length < 2) return 1;
|
|
|
|
const lengths = sentences.map(s => s.trim().length);
|
|
const avgLength = lengths.reduce((a, b) => a + b, 0) / lengths.length;
|
|
|
|
// Calculer variance des longueurs
|
|
const variance = lengths.reduce((acc, len) => acc + Math.pow(len - avgLength, 2), 0) / lengths.length;
|
|
const stdDev = Math.sqrt(variance);
|
|
|
|
// Score de variété (0-1) - plus la variance est élevée, plus c'est varié
|
|
return Math.min(1, stdDev / avgLength);
|
|
}
|
|
|
|
function analyzeFormalityLevel(content) {
|
|
const formalIndicators = [
|
|
'il convient de', 'par conséquent', 'néanmoins', 'toutefois',
|
|
'de surcroît', 'en définitive', 'il s\'avère que', 'force est de constater'
|
|
];
|
|
|
|
let formalCount = 0;
|
|
formalIndicators.forEach(indicator => {
|
|
if (content.toLowerCase().includes(indicator)) formalCount++;
|
|
});
|
|
|
|
const sentences = content.split(/[.!?]+/).length;
|
|
return sentences > 0 ? formalCount / sentences : 0;
|
|
}
|
|
|
|
function calculateTopicContinuity(sentence1, sentence2) {
|
|
const stopWords = ['les', 'des', 'une', 'sont', 'avec', 'pour', 'dans', 'cette', 'vous', 'peut', 'tout'];
|
|
|
|
const words1 = extractSignificantWords(sentence1, stopWords);
|
|
const words2 = extractSignificantWords(sentence2, stopWords);
|
|
|
|
if (words1.length === 0 || words2.length === 0) return 0;
|
|
|
|
const commonWords = words1.filter(word => words2.includes(word));
|
|
const semanticSimilarity = commonWords.length / Math.min(words1.length, words2.length);
|
|
|
|
const technicalWords = ['plaque', 'dibond', 'aluminium', 'impression', 'signalétique'];
|
|
const commonTechnical = commonWords.filter(word => technicalWords.includes(word));
|
|
const technicalBonus = commonTechnical.length * 0.2;
|
|
|
|
return Math.min(1, semanticSimilarity + technicalBonus);
|
|
}
|
|
|
|
function extractSignificantWords(sentence, stopWords) {
|
|
return sentence.toLowerCase()
|
|
.match(/\b[a-zàâäéèêëïîôùûüÿç]{4,}\b/g) // Mots 4+ lettres avec accents
|
|
?.filter(word => !stopWords.includes(word)) || [];
|
|
}
|
|
|
|
function hasTransitionWord(sentence) {
|
|
const connectors = ['par ailleurs', 'en effet', 'de plus', 'cependant', 'ainsi', 'donc', 'ensuite', 'puis', 'également', 'aussi', 'toutefois', 'néanmoins', 'alors', 'enfin'];
|
|
return connectors.some(connector => sentence.toLowerCase().includes(connector));
|
|
}
|
|
|
|
/**
|
|
* Instructions de style dynamiques
|
|
*/
|
|
function getPersonalityStyleInstructions(personality) {
|
|
// CORRECTION: Utilisation des VRAIS champs Google Sheets au lieu du hardcodé
|
|
if (!personality) return "Style professionnel standard";
|
|
|
|
const instructions = `STYLE ${personality.nom.toUpperCase()} (${personality.style}):
|
|
- Description: ${personality.description}
|
|
- Vocabulaire préféré: ${personality.vocabulairePref || 'professionnel, qualité'}
|
|
- Connecteurs préférés: ${personality.connecteursPref || 'par ailleurs, en effet'}
|
|
- Mots-clés secteurs: ${personality.motsClesSecteurs || 'technique, qualité'}
|
|
- Longueur phrases: ${personality.longueurPhrases || 'Moyennes (15-25 mots)'}
|
|
- Niveau technique: ${personality.niveauTechnique || 'Accessible'}
|
|
- Style CTA: ${personality.ctaStyle || 'Professionnel'}
|
|
- Défauts simulés: ${personality.defautsSimules || 'Aucun'}
|
|
- Erreurs typiques à éviter: ${personality.erreursTypiques || 'Répétitions, généralités'}`;
|
|
|
|
return instructions;
|
|
}
|
|
|
|
/**
|
|
* Créer prompt pour élément (fonction de base nécessaire)
|
|
*/
|
|
function createPromptForElement(element, csvData) {
|
|
const personality = csvData.personality;
|
|
const styleContext = `Rédige dans le style ${personality.style} de ${personality.nom} (${personality.description}).`;
|
|
|
|
switch (element.type) {
|
|
case 'titre_h1':
|
|
return `${styleContext}
|
|
MISSION: Crée un titre H1 accrocheur pour: ${csvData.mc0}
|
|
Référence: ${csvData.t0}
|
|
CONSIGNES: 10 mots maximum, direct et impactant, optimisé SEO.
|
|
RÉPONDS UNIQUEMENT PAR LE TITRE, sans introduction.`;
|
|
|
|
case 'titre_h2':
|
|
return `${styleContext}
|
|
MISSION: Crée un titre H2 optimisé SEO pour: ${csvData.mc0}
|
|
CONSIGNES: Intègre naturellement le mot-clé, 8 mots maximum.
|
|
RÉPONDS UNIQUEMENT PAR LE TITRE, sans introduction.`;
|
|
|
|
case 'intro':
|
|
if (element.instructions) {
|
|
return `${styleContext}
|
|
MISSION: ${element.instructions}
|
|
Données contextuelles:
|
|
- MC0: ${csvData.mc0}
|
|
- T-1: ${csvData.tMinus1}
|
|
- L-1: ${csvData.lMinus1}
|
|
RÉPONDS UNIQUEMENT PAR LE CONTENU, sans présentation.`;
|
|
}
|
|
return `${styleContext}
|
|
MISSION: Rédige une introduction de 100 mots pour ${csvData.mc0}.
|
|
RÉPONDS UNIQUEMENT PAR LE CONTENU, sans présentation.`;
|
|
|
|
case 'texte':
|
|
if (element.instructions) {
|
|
return `${styleContext}
|
|
MISSION: ${element.instructions}
|
|
RÉPONDS UNIQUEMENT PAR LE CONTENU, sans présentation.`;
|
|
}
|
|
return `${styleContext}
|
|
MISSION: Rédige un paragraphe de 150 mots sur ${csvData.mc0}.
|
|
RÉPONDS UNIQUEMENT PAR LE CONTENU, sans présentation.`;
|
|
|
|
case 'faq_question':
|
|
if (element.instructions) {
|
|
return `${styleContext}
|
|
MISSION: ${element.instructions}
|
|
CONTEXTE: ${csvData.mc0} - ${csvData.t0}
|
|
STYLE: Question ${csvData.personality?.style} de ${csvData.personality?.nom}
|
|
CONSIGNES:
|
|
- Vraie question que se poserait un client intéressé par ${csvData.mc0}
|
|
- Commence par "Comment", "Quel", "Pourquoi", "Où", "Quand" ou "Est-ce que"
|
|
- Maximum 15 mots, pratique et concrète
|
|
- Vocabulaire: ${csvData.personality?.vocabulairePref || 'accessible'}
|
|
RÉPONDS UNIQUEMENT PAR LA QUESTION, sans guillemets ni introduction.`;
|
|
}
|
|
return `${styleContext}
|
|
MISSION: Génère une vraie question FAQ client sur ${csvData.mc0}.
|
|
CONSIGNES:
|
|
- Question pratique et concrète qu'un client se poserait
|
|
- Commence par "Comment", "Quel", "Pourquoi", "Combien", "Où" ou "Est-ce que"
|
|
- Maximum 15 mots, style ${csvData.personality?.style}
|
|
- Vocabulaire: ${csvData.personality?.vocabulairePref || 'accessible'}
|
|
RÉPONDS UNIQUEMENT PAR LA QUESTION, sans guillemets ni introduction.`;
|
|
|
|
case 'faq_reponse':
|
|
if (element.instructions) {
|
|
return `${styleContext}
|
|
MISSION: ${element.instructions}
|
|
CONTEXTE: ${csvData.mc0} - ${csvData.t0}
|
|
STYLE: Réponse ${csvData.personality?.style} de ${csvData.personality?.nom}
|
|
CONSIGNES:
|
|
- Réponse utile et rassurante
|
|
- 50-80 mots, ton ${csvData.personality?.style}
|
|
- Vocabulaire: ${csvData.personality?.vocabulairePref}
|
|
- Connecteurs: ${csvData.personality?.connecteursPref}
|
|
RÉPONDS UNIQUEMENT PAR LA RÉPONSE, sans introduction.`;
|
|
}
|
|
return `${styleContext}
|
|
MISSION: Réponds à une question client sur ${csvData.mc0}.
|
|
CONSIGNES:
|
|
- Réponse utile, claire et rassurante
|
|
- 50-80 mots, ton ${csvData.personality?.style} de ${csvData.personality?.nom}
|
|
- Vocabulaire: ${csvData.personality?.vocabulairePref || 'professionnel'}
|
|
- Connecteurs: ${csvData.personality?.connecteursPref || 'par ailleurs'}
|
|
RÉPONDS UNIQUEMENT PAR LA RÉPONSE, sans introduction.`;
|
|
|
|
default:
|
|
return `${styleContext}
|
|
MISSION: Génère du contenu pertinent pour ${csvData.mc0}.
|
|
RÉPONDS UNIQUEMENT PAR LE CONTENU, sans présentation.`;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* NOUVELLE FONCTION : Extraction batch TOUS les termes techniques
|
|
*/
|
|
async function extractAllTechnicalTermsBatch(baseContents, csvData, aiProvider) {
|
|
const contentEntries = Object.keys(baseContents);
|
|
|
|
const batchAnalysisPrompt = `MISSION: Analyser ces ${contentEntries.length} contenus et identifier leurs termes techniques.
|
|
|
|
CONTEXTE: ${csvData.mc0} - Secteur: signalétique/impression
|
|
|
|
CONTENUS À ANALYSER:
|
|
|
|
${contentEntries.map((tag, i) => `[${i + 1}] TAG: ${tag}
|
|
CONTENU: "${baseContents[tag]}"`).join('\n\n')}
|
|
|
|
CONSIGNES:
|
|
- Identifie UNIQUEMENT les vrais termes techniques métier/industrie
|
|
- Évite mots génériques (qualité, service, pratique, personnalisé, etc.)
|
|
- Focus: matériaux, procédés, normes, dimensions, technologies
|
|
- Si aucun terme technique → "AUCUN"
|
|
|
|
EXEMPLES VALIDES: dibond, impression UV, fraisage CNC, épaisseur 3mm, aluminium brossé
|
|
EXEMPLES INVALIDES: durable, pratique, personnalisé, moderne, esthétique
|
|
|
|
FORMAT RÉPONSE EXACT:
|
|
[1] dibond, impression UV, 3mm OU AUCUN
|
|
[2] aluminium, fraisage CNC OU AUCUN
|
|
[3] AUCUN
|
|
etc... (${contentEntries.length} lignes total)`;
|
|
|
|
try {
|
|
const analysisResponse = await callLLM(aiProvider, batchAnalysisPrompt, {
|
|
temperature: 0.3,
|
|
maxTokens: 2000
|
|
}, csvData.personality);
|
|
|
|
return parseAllTechnicalTermsResponse(analysisResponse, baseContents, contentEntries);
|
|
|
|
} catch (error) {
|
|
logSh(`❌ FATAL: Extraction termes techniques batch échouée: ${error.message}`, 'ERROR');
|
|
throw new Error(`FATAL: Analyse termes techniques impossible - arrêt du workflow: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* NOUVELLE FONCTION : Enhancement batch TOUS les éléments
|
|
*/
|
|
async function enhanceAllElementsTechnicalBatch(elementsNeedingEnhancement, csvData, aiProvider) {
|
|
if (elementsNeedingEnhancement.length === 0) return {};
|
|
|
|
const batchEnhancementPrompt = `MISSION: Améliore UNIQUEMENT la précision technique de ces ${elementsNeedingEnhancement.length} contenus.
|
|
|
|
CONTEXTE: Article SEO pour site e-commerce de signalétique
|
|
PERSONNALITÉ: ${csvData.personality?.nom} (${csvData.personality?.style} web professionnel)
|
|
SUJET: ${csvData.mc0} - Secteur: Signalétique/impression
|
|
VOCABULAIRE PRÉFÉRÉ: ${csvData.personality?.vocabulairePref}
|
|
|
|
CONTENUS + TERMES À AMÉLIORER:
|
|
|
|
${elementsNeedingEnhancement.map((item, i) => `[${i + 1}] TAG: ${item.tag}
|
|
CONTENU ACTUEL: "${item.content}"
|
|
TERMES TECHNIQUES À INTÉGRER: ${item.technicalTerms.join(', ')}`).join('\n\n')}
|
|
|
|
CONSIGNES STRICTES:
|
|
- Améliore UNIQUEMENT la précision technique, garde le style ${csvData.personality?.nom}
|
|
- GARDE la même longueur, structure et ton
|
|
- Intègre naturellement les termes techniques listés
|
|
- NE CHANGE PAS le fond du message ni le style personnel
|
|
- Utilise un vocabulaire expert mais accessible
|
|
- ÉVITE les répétitions excessives
|
|
- RESPECTE le niveau technique: ${csvData.personality?.niveauTechnique}
|
|
- Termes techniques secteur: dibond, aluminium, impression UV, fraisage, épaisseur, PMMA
|
|
|
|
FORMAT RÉPONSE:
|
|
[1] Contenu avec amélioration technique selon ${csvData.personality?.nom}
|
|
[2] Contenu avec amélioration technique selon ${csvData.personality?.nom}
|
|
etc... (${elementsNeedingEnhancement.length} éléments total)`;
|
|
|
|
try {
|
|
const enhanced = await callLLM(aiProvider, batchEnhancementPrompt, {
|
|
temperature: 0.4,
|
|
maxTokens: 5000 // Plus large pour batch total
|
|
}, csvData.personality);
|
|
|
|
return parseTechnicalEnhancementBatchResponse(enhanced, elementsNeedingEnhancement);
|
|
|
|
} catch (error) {
|
|
logSh(`❌ FATAL: Enhancement technique batch échoué: ${error.message}`, 'ERROR');
|
|
throw new Error(`FATAL: Enhancement technique batch impossible - arrêt du workflow: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Parser réponse extraction termes
|
|
*/
|
|
function parseAllTechnicalTermsResponse(response, baseContents, contentEntries) {
|
|
const results = [];
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\[\d+\]|$)/gs;
|
|
let match;
|
|
const parsedItems = {};
|
|
|
|
// Parser la réponse
|
|
while ((match = regex.exec(response)) !== null) {
|
|
const index = parseInt(match[1]) - 1; // Convertir en 0-indexé
|
|
const termsText = match[2].trim();
|
|
parsedItems[index] = termsText;
|
|
}
|
|
|
|
// Mapper aux éléments
|
|
contentEntries.forEach((tag, index) => {
|
|
const termsText = parsedItems[index] || 'AUCUN';
|
|
const hasTerms = !termsText.toUpperCase().includes('AUCUN');
|
|
|
|
const technicalTerms = hasTerms ?
|
|
termsText.split(',').map(t => t.trim()).filter(t => t.length > 0) :
|
|
[];
|
|
|
|
results.push({
|
|
tag: tag,
|
|
content: baseContents[tag],
|
|
technicalTerms: technicalTerms,
|
|
needsEnhancement: hasTerms && technicalTerms.length > 0
|
|
});
|
|
|
|
logSh(`🔍 [${tag}]: ${hasTerms ? technicalTerms.join(', ') : 'pas de termes techniques'}`, 'DEBUG');
|
|
});
|
|
|
|
const enhancementCount = results.filter(r => r.needsEnhancement).length;
|
|
logSh(`📊 Analyse terminée: ${enhancementCount}/${contentEntries.length} éléments ont besoin d'enhancement`, 'INFO');
|
|
|
|
return results;
|
|
}
|
|
|
|
/**
|
|
* Parser réponse enhancement technique
|
|
*/
|
|
function parseTechnicalEnhancementBatchResponse(response, elementsNeedingEnhancement) {
|
|
const results = {};
|
|
const regex = /\[(\d+)\]\s*([^[]*?)(?=\[\d+\]|$)/gs;
|
|
let match;
|
|
let index = 0;
|
|
|
|
while ((match = regex.exec(response)) && index < elementsNeedingEnhancement.length) {
|
|
let content = match[2].trim();
|
|
const element = elementsNeedingEnhancement[index];
|
|
|
|
// NOUVEAU: Appliquer le nettoyage XML
|
|
content = cleanXMLTagsFromContent(content);
|
|
|
|
if (content && content.length > 10) {
|
|
results[element.tag] = content;
|
|
logSh(`✅ Enhanced [${element.tag}]: "${content.substring(0, 100)}..."`, 'DEBUG');
|
|
} else {
|
|
// Fallback si contenu invalide après nettoyage
|
|
results[element.tag] = element.content;
|
|
logSh(`⚠️ Fallback [${element.tag}]: contenu invalide après nettoyage`, 'WARNING');
|
|
}
|
|
|
|
index++;
|
|
}
|
|
|
|
// Vérifier si on a bien tout parsé
|
|
if (Object.keys(results).length < elementsNeedingEnhancement.length) {
|
|
logSh(`⚠️ Parsing partiel: ${Object.keys(results).length}/${elementsNeedingEnhancement.length}`, 'WARNING');
|
|
|
|
// Compléter avec contenu original pour les manquants
|
|
elementsNeedingEnhancement.forEach(element => {
|
|
if (!results[element.tag]) {
|
|
results[element.tag] = element.content;
|
|
}
|
|
});
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
// ============= EXPORTS =============
|
|
|
|
module.exports = {
|
|
generateWithBatchEnhancement,
|
|
generateAllContentBase,
|
|
enhanceAllTechnicalTerms,
|
|
enhanceAllTransitions,
|
|
enhanceAllPersonalityStyle,
|
|
collectAllElements,
|
|
groupElementsByType,
|
|
chunkArray,
|
|
createBatchBasePrompt,
|
|
parseBatchResponse,
|
|
cleanXMLTagsFromContent,
|
|
analyzeTransitionNeed,
|
|
getPersonalityStyleInstructions,
|
|
createPromptForElement,
|
|
sleep,
|
|
separateFAQPairsAndOthers,
|
|
generateFAQPairsRestored,
|
|
createBatchFAQPairsPrompt,
|
|
parseFAQPairsResponse,
|
|
cleanFAQInstructions,
|
|
extractAllTechnicalTermsBatch,
|
|
enhanceAllElementsTechnicalBatch,
|
|
parseAllTechnicalTermsResponse,
|
|
parseTechnicalEnhancementBatchResponse
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/trace-wrap.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// lib/trace-wrap.js
|
|
const { tracer } = require('./trace.js');
|
|
|
|
const traced = (name, fn, attrs) => (...args) =>
|
|
tracer.run(name, () => fn(...args), attrs);
|
|
|
|
module.exports = {
|
|
traced
|
|
};
|
|
|
|
/*
|
|
┌────────────────────────────────────────────────────────────────────┐
|
|
│ File: lib/Utils.js │
|
|
└────────────────────────────────────────────────────────────────────┘
|
|
*/
|
|
|
|
// ========================================
|
|
// FICHIER: utils.js - Conversion Node.js
|
|
// Description: Utilitaires génériques pour le workflow
|
|
// ========================================
|
|
|
|
// Import du système de logging (assumant que logSh est disponible globalement)
|
|
// const { logSh } = require('./logging'); // À décommenter si logSh est dans un module séparé
|
|
|
|
/**
|
|
* Créer une réponse de succès standardisée
|
|
* @param {Object} data - Données à retourner
|
|
* @returns {Object} Réponse formatée pour Express/HTTP
|
|
*/
|
|
function createSuccessResponse(data) {
|
|
return {
|
|
success: true,
|
|
data: data,
|
|
timestamp: new Date().toISOString()
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Créer une réponse d'erreur standardisée
|
|
* @param {string|Error} error - Message d'erreur ou objet Error
|
|
* @returns {Object} Réponse d'erreur formatée
|
|
*/
|
|
function createErrorResponse(error) {
|
|
const errorMessage = error instanceof Error ? error.message : error.toString();
|
|
|
|
return {
|
|
success: false,
|
|
error: errorMessage,
|
|
timestamp: new Date().toISOString(),
|
|
stack: process.env.NODE_ENV === 'development' && error instanceof Error ? error.stack : undefined
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Middleware Express pour envoyer des réponses standardisées
|
|
* Usage: res.success(data) ou res.error(error)
|
|
*/
|
|
function responseMiddleware(req, res, next) {
|
|
// Méthode pour réponse de succès
|
|
res.success = (data, statusCode = 200) => {
|
|
res.status(statusCode).json(createSuccessResponse(data));
|
|
};
|
|
|
|
// Méthode pour réponse d'erreur
|
|
res.error = (error, statusCode = 500) => {
|
|
res.status(statusCode).json(createErrorResponse(error));
|
|
};
|
|
|
|
next();
|
|
}
|
|
|
|
/**
|
|
* HELPER : Nettoyer les instructions FAQ
|
|
* Remplace les variables et nettoie le HTML
|
|
* @param {string} instructions - Instructions à nettoyer
|
|
* @param {Object} csvData - Données CSV pour remplacement variables
|
|
* @returns {string} Instructions nettoyées
|
|
*/
|
|
function cleanFAQInstructions(instructions, csvData) {
|
|
if (!instructions || !csvData) {
|
|
return instructions || '';
|
|
}
|
|
|
|
let clean = instructions.toString();
|
|
|
|
try {
|
|
// Remplacer variables simples
|
|
clean = clean.replace(/\{\{MC0\}\}/g, csvData.mc0 || '');
|
|
clean = clean.replace(/\{\{T0\}\}/g, csvData.t0 || '');
|
|
|
|
// Variables multiples si nécessaire
|
|
if (csvData.mcPlus1) {
|
|
const mcPlus1 = csvData.mcPlus1.split(',').map(s => s.trim());
|
|
|
|
for (let i = 1; i <= 6; i++) {
|
|
const mcValue = mcPlus1[i-1] || `[MC+1_${i} non défini]`;
|
|
clean = clean.replace(new RegExp(`\\{\\{MC\\+1_${i}\\}\\}`, 'g'), mcValue);
|
|
}
|
|
}
|
|
|
|
// Variables T+1 et L+1 si disponibles
|
|
if (csvData.tPlus1) {
|
|
const tPlus1 = csvData.tPlus1.split(',').map(s => s.trim());
|
|
for (let i = 1; i <= 6; i++) {
|
|
const tValue = tPlus1[i-1] || `[T+1_${i} non défini]`;
|
|
clean = clean.replace(new RegExp(`\\{\\{T\\+1_${i}\\}\\}`, 'g'), tValue);
|
|
}
|
|
}
|
|
|
|
if (csvData.lPlus1) {
|
|
const lPlus1 = csvData.lPlus1.split(',').map(s => s.trim());
|
|
for (let i = 1; i <= 6; i++) {
|
|
const lValue = lPlus1[i-1] || `[L+1_${i} non défini]`;
|
|
clean = clean.replace(new RegExp(`\\{\\{L\\+1_${i}\\}\\}`, 'g'), lValue);
|
|
}
|
|
}
|
|
|
|
// Nettoyer HTML
|
|
clean = clean.replace(/<\/?[^>]+>/g, '');
|
|
|
|
// Nettoyer espaces en trop
|
|
clean = clean.replace(/\s+/g, ' ').trim();
|
|
|
|
} catch (error) {
|
|
if (typeof logSh === 'function') {
|
|
logSh(`⚠️ Erreur nettoyage instructions FAQ: ${error.toString()}`, 'WARNING');
|
|
}
|
|
// Retourner au moins la version partiellement nettoyée
|
|
}
|
|
|
|
return clean;
|
|
}
|
|
|
|
/**
|
|
* Utilitaire pour attendre un délai (remplace Utilities.sleep de Google Apps Script)
|
|
* @param {number} ms - Millisecondes à attendre
|
|
* @returns {Promise} Promise qui se résout après le délai
|
|
*/
|
|
function sleep(ms) {
|
|
return new Promise(resolve => setTimeout(resolve, ms));
|
|
}
|
|
|
|
/**
|
|
* Utilitaire pour encoder en base64
|
|
* @param {string} text - Texte à encoder
|
|
* @returns {string} Texte encodé en base64
|
|
*/
|
|
function base64Encode(text) {
|
|
return Buffer.from(text, 'utf8').toString('base64');
|
|
}
|
|
|
|
/**
|
|
* Utilitaire pour décoder du base64
|
|
* @param {string} base64Text - Texte base64 à décoder
|
|
* @returns {string} Texte décodé
|
|
*/
|
|
function base64Decode(base64Text) {
|
|
return Buffer.from(base64Text, 'base64').toString('utf8');
|
|
}
|
|
|
|
/**
|
|
* Valider et nettoyer un slug/filename
|
|
* @param {string} slug - Slug à nettoyer
|
|
* @returns {string} Slug nettoyé
|
|
*/
|
|
function cleanSlug(slug) {
|
|
if (!slug) return '';
|
|
|
|
return slug
|
|
.toString()
|
|
.toLowerCase()
|
|
.replace(/[^a-z0-9\-_]/g, '-') // Remplacer caractères spéciaux par -
|
|
.replace(/-+/g, '-') // Éviter doubles tirets
|
|
.replace(/^-+|-+$/g, ''); // Enlever tirets début/fin
|
|
}
|
|
|
|
/**
|
|
* Compter les mots dans un texte
|
|
* @param {string} text - Texte à analyser
|
|
* @returns {number} Nombre de mots
|
|
*/
|
|
function countWords(text) {
|
|
if (!text || typeof text !== 'string') return 0;
|
|
|
|
return text
|
|
.trim()
|
|
.replace(/\s+/g, ' ') // Normaliser espaces
|
|
.split(' ')
|
|
.filter(word => word.length > 0)
|
|
.length;
|
|
}
|
|
|
|
/**
|
|
* Formater une durée en millisecondes en format lisible
|
|
* @param {number} ms - Durée en millisecondes
|
|
* @returns {string} Durée formatée (ex: "2.3s" ou "450ms")
|
|
*/
|
|
function formatDuration(ms) {
|
|
if (ms < 1000) {
|
|
return `${ms}ms`;
|
|
} else if (ms < 60000) {
|
|
return `${(ms / 1000).toFixed(1)}s`;
|
|
} else {
|
|
const minutes = Math.floor(ms / 60000);
|
|
const seconds = ((ms % 60000) / 1000).toFixed(1);
|
|
return `${minutes}m ${seconds}s`;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Utilitaire pour retry automatique d'une fonction
|
|
* @param {Function} fn - Fonction à exécuter avec retry
|
|
* @param {number} maxRetries - Nombre maximum de tentatives
|
|
* @param {number} delay - Délai entre tentatives (ms)
|
|
* @returns {Promise} Résultat de la fonction ou erreur finale
|
|
*/
|
|
async function withRetry(fn, maxRetries = 3, delay = 1000) {
|
|
let lastError;
|
|
|
|
for (let attempt = 1; attempt <= maxRetries; attempt++) {
|
|
try {
|
|
return await fn();
|
|
} catch (error) {
|
|
lastError = error;
|
|
|
|
if (typeof logSh === 'function') {
|
|
logSh(`⚠️ Tentative ${attempt}/${maxRetries} échouée: ${error.toString()}`, 'WARNING');
|
|
}
|
|
|
|
if (attempt < maxRetries) {
|
|
await sleep(delay * attempt); // Exponential backoff
|
|
}
|
|
}
|
|
}
|
|
|
|
throw lastError;
|
|
}
|
|
|
|
/**
|
|
* Validation basique d'email
|
|
* @param {string} email - Email à valider
|
|
* @returns {boolean} True si email valide
|
|
*/
|
|
function isValidEmail(email) {
|
|
const emailRegex = /^[^\s@]+@[^\s@]+\.[^\s@]+$/;
|
|
return emailRegex.test(email);
|
|
}
|
|
|
|
/**
|
|
* Générer un ID unique simple
|
|
* @returns {string} ID unique basé sur timestamp + random
|
|
*/
|
|
function generateId() {
|
|
return `${Date.now()}-${Math.random().toString(36).substr(2, 9)}`;
|
|
}
|
|
|
|
/**
|
|
* Truncate un texte à une longueur donnée
|
|
* @param {string} text - Texte à tronquer
|
|
* @param {number} maxLength - Longueur maximale
|
|
* @param {string} suffix - Suffixe à ajouter si tronqué (défaut: '...')
|
|
* @returns {string} Texte tronqué
|
|
*/
|
|
function truncate(text, maxLength, suffix = '...') {
|
|
if (!text || text.length <= maxLength) {
|
|
return text;
|
|
}
|
|
|
|
return text.substring(0, maxLength - suffix.length) + suffix;
|
|
}
|
|
|
|
// ============= EXPORTS =============
|
|
|
|
module.exports = {
|
|
createSuccessResponse,
|
|
createErrorResponse,
|
|
responseMiddleware,
|
|
cleanFAQInstructions,
|
|
sleep,
|
|
base64Encode,
|
|
base64Decode,
|
|
cleanSlug,
|
|
countWords,
|
|
formatDuration,
|
|
withRetry,
|
|
isValidEmail,
|
|
generateId,
|
|
truncate
|
|
};
|
|
|