🔍 French Influencer Monitor

A comprehensive Python web application that monitors French influencers by scraping public online data and classifying them based on controversies, positive actions, and overall trustworthiness.

🌟 Features

• Multi-Source Scraping: Parallel scraping from:

  • 📰 French news sites (Le Monde, Le Figaro, etc.)
  • 🎥 YouTube
  • 🐦 Twitter/X
  • 💬 Reddit
  • 🗨️ French forums

• AI-Powered Analysis:

  • French sentiment analysis using CamemBERT
  • Automatic classification: Drama, Good Action, or Neutral
  • Keyword-based detection for controversies and positive actions

• Trust Scoring System:

  • 0-100 trust score calculation
  • Weighted by recency (recent events matter more)
  • Based on drama/good action ratio and sentiment

• Interactive UI:

  • Real-time search and analysis
  • Visual dashboards with charts
  • Detailed mention breakdown
  • Source citations with links

🚀 Installation

Prerequisites

• Python 3.8+
• pip

Setup

1. Clone or navigate to the project directory:

cd "/Users/roane/roane/perso/hackathon blackbox"

2. Create a virtual environment:

python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Configure environment (optional):

cp .env.example .env
# Edit .env with your API keys if needed

🎯 Usage

Run the Streamlit App

streamlit run streamlit_app.py

The app will open in your browser at http://localhost:8501

How to Use

1. Enter a French influencer's name (e.g., "Squeezie", "Norman", "Cyprien")
2. Click "🔍 Analyser"
3. Wait for the parallel scraping and analysis to complete
4. View the results:
   • Trust score (0-100)
   • Controversy count
   • Positive action count
   • Detailed mentions with sources
   • Visual charts and breakdowns

Using Cache

Enable "Utiliser le cache" in the sidebar to use previously analyzed data instead of re-scraping.

📊 Architecture

├── streamlit_app.py          # Main Streamlit UI
├── orchestrator.py            # Parallel execution coordinator
├── analyzer.py                # Sentiment analysis (CamemBERT)
├── scorer.py                  # Trust score calculation
├── database.py                # SQLite database management
├── config.py                  # Configuration and keywords
├── scrapers/
│   ├── base_scraper.py       # Base scraper class
│   ├── news_scraper.py       # News sites scraper
│   ├── youtube_scraper.py    # YouTube scraper
│   ├── twitter_scraper.py    # Twitter/X scraper
│   ├── reddit_scraper.py     # Reddit scraper
│   └── forum_scraper.py      # French forums scraper
└── requirements.txt           # Python dependencies

🧠 How It Works

1. Parallel Scraping

All scrapers run concurrently using asyncio to maximize speed:

• Each scraper searches for the influencer's name
• Extracts relevant text excerpts and URLs
• Returns structured data

2. Sentiment Analysis

• Uses cmarkea/distilcamembert-base-sentiment model
• Analyzes French text for positive/negative sentiment
• Combines with keyword matching for classification

3. Classification

Content is classified as:

• Drama: Negative sentiment + controversy keywords (scandale, polémique, etc.)
• Good Action: Positive sentiment + charity keywords (don, charité, etc.)
• Neutral: Everything else

4. Trust Score Calculation

Base Score = 50
+ (good_actions × 10 × recency_weight)
- (dramas × 15 × recency_weight)
+ (avg_sentiment × 20)
Normalized to 0-100

Recency weight uses exponential decay: more recent mentions have higher impact.

🔧 Configuration

Edit config.py to customize:

• Sentiment model
• Drama/good action keywords
• News sources
• Scraping parameters

📝 Database

SQLite database stores:

• Influencers: Name, trust score, counts, timestamps
• Mentions: All scraped data with sentiment labels
• Analysis History: Historical trust scores

⚠️ Limitations

• Web scraping depends on site availability and structure
• Some platforms (Twitter) may require API keys for better results
• Sentiment analysis accuracy depends on text quality
• Rate limiting may slow down scraping

🛠️ Troubleshooting

Model Download Issues

If the sentiment model fails to download:

# Pre-download the model
python -c "from transformers import AutoTokenizer, AutoModelForSequenceClassification; AutoTokenizer.from_pretrained('cmarkea/distilcamembert-base-sentiment'); AutoModelForSequenceClassification.from_pretrained('cmarkea/distilcamembert-base-sentiment')"

Scraping Errors

• Check your internet connection
• Some sites may block scrapers - this is normal
• The app will continue with available data

📄 License

This project is for educational purposes (Hackathon).

🤝 Contributing

Built for the Blackbox Hackathon 2025.

🎓 Technologies Used

• Backend: Python, asyncio, aiohttp
• UI: Streamlit
• Scraping: BeautifulSoup, Newspaper3k
• AI: HuggingFace Transformers (CamemBERT)
• Database: SQLite, SQLAlchemy
• Visualization: Plotly

---

Made with ❤️ for the Hackathon Blackbox