Fact Checking System: Information Credibility Verification

PhD Thesis Prototype - Dominique S. Loyer
Citation Key: loyerModelingHybridSystem2025

Note

New in v2.2 (Jan 29, 2026):

• GraphRAG: Contextual memory from Knowledge Graph.
• Interactive Graph: D3.js visualization with physics and details on click.
• Cloud Ready: Docker & Supabase integration.

---

📋 Overview

A neuro-symbolic AI system for verifying information credibility that combines:

• Symbolic AI: Rule-based reasoning with OWL ontologies (RDF/Turtle)
• Neural AI: Transformer models for sentiment analysis and NER
• IR Engine: BM25, TF-IDF, and PageRank estimation

The system provides explainable credibility scores (High/Medium/Low) with detailed factor breakdown.

---

🚀 Quick Start (v2.2 - January 2026)

Installation via PyPI (Recommended)

Option 1: Minimal Installation (Lightweight, ~100 MB)

Perfect for exploring the code, basic credibility checking without ML features:

pip install syscred

Option 2: With Machine Learning (Complete, ~2.5 GB)

Includes PyTorch, Transformers, and all ML models for full credibility analysis:

pip install syscred[ml]

Option 3: Full Installation (All features)

Includes ML, production tools, and development dependencies:

pip install syscred[all]

Alternative: Run on Kaggle/Colab

1. Click the Kaggle or Colab badge above
2. Enable GPU runtime
3. Run All cells

Alternative: Local Installation with Docker

# Clone the repository
git clone https://github.com/DominiqueLoyer/systemFactChecking.git
cd systemFactChecking/02_Code

# Run with Startup Script (Mac/Linux)
./start_syscred.sh
# Access at http://localhost:5001

Python API Usage

from syscred import CredibilityVerificationSystem

# Initialize
system = CredibilityVerificationSystem()

# Verify a URL
result = system.verify_information("https://www.lemonde.fr/article")
print(f"Score: {result['scoreCredibilite']} ({result['niveauCredibilite']})")

# Verify text directly
result = system.verify_information(
    "According to Harvard researchers, the new study shows significant results."
)

---

📡 REST API Endpoints

Endpoint	Method	Description
/api/verify	POST	Full credibility verification
/api/seo	POST	SEO analysis only (faster)
/api/ontology/stats	GET	Ontology statistics
/api/health	GET	Server health check

Example Request

curl -X POST http://localhost:5000/api/verify \
  -H "Content-Type: application/json" \
  -d '{"input_data": "https://www.bbc.com/news/article"}'

Example Response

{
  "scoreCredibilite": 0.78,
  "niveauCredibilite": "HIGH",
  "analysisDetails": {
    "sourceReputation": "High",
    "domainAge": 9125,
    "sentiment": {"label": "NEUTRAL", "score": 0.52},
    "entities": [{"word": "BBC", "entity_group": "ORG"}]
  }
}

---

📁 Project Structure

systemFactChecking/ ├── README.md # This file ├── 01_Presentations/ # Presentations (.pdf, .tex) ├── 02_Code/ # Source Code & Docker │ ├── syscred/ # ⭐ CORE ENGINE (v2.2) │ │ ├── graph_rag.py # [NEW] GraphRAG Module │ │ ├── verification_system.py │ │ ├── database.py # [NEW] Supabase Connector │ │ └── ... │ ├── start_syscred.sh # Startup Script │ ├── Dockerfile # Deployment Config │ └── requirements.txt ├── 03_Docs/ # Documentation (.pdf) └── 04_Bibliography/ # References (.bib, .pdf)

---

## 🔧 Configuration

Set environment variables or edit `02_Code/v2_syscred/config.py`:

```bash
# Optional: Google Fact Check API key
export SYSCRED_GOOGLE_API_KEY=your_key_here

# Server settings
export SYSCRED_PORT=5000
export SYSCRED_DEBUG=true
export SYSCRED_ENV=production  # or development, testing

---

📊 Credibility Scoring

The system uses weighted factors to calculate credibility:

Factor	Weight	Description
Source Reputation	25%	Known credible sources database
Domain Age	10%	WHOIS lookup for domain history
Sentiment Neutrality	15%	Extreme sentiment = lower score
Entity Presence	15%	Named entities (ORG, PER)
Text Coherence	15%	Vocabulary diversity
Fact Check	20%	Google Fact Check API results

Thresholds:

• HIGH: Score ≥ 0.7
• MEDIUM: 0.4 ≤ Score < 0.7
• LOW: Score < 0.4

---

📚 Documentation & Papers

• Modeling and Hybrid System for Verification of Sources Credibility (PDF)
• Ontology of a Verification System (PDF)
• Beamer Presentation - DIC9335 (PDF)

---

🏷️ Citation

@software{loyer2025syscred,
  author = {Loyer, Dominique S.},
  title = {SysCRED: Neuro-Symbolic System for Information Credibility Verification},
  year = {2026},
  publisher = {GitHub},
  url = {https://github.com/DominiqueLoyer/systemFactChecking}
}

---

📜 License

MIT License - See LICENSE for details.

---

🔄 Version History

Version	Date	Changes
v2.0	Jan 2026	Complete rewrite with modular architecture, Kaggle/Colab support, REST API
v1.0	Apr 2025	Initial prototype with basic credibility scoring