An intelligent web application that analyzes news articles and claims using a multi-signal AI pipeline to detect misinformation. Built with Next.js, FastAPI, and fine-tuned RoBERTa.
Paste an article, drop a URL, or type a claim — get an instant credibility verdict with a full explanation of why.
Multi-Signal Analysis Pipeline
- 7-step analysis combining RoBERTa classification, sentiment analysis, source credibility, fact-checking, clickbait detection, and explainability
- Multi-class verdict: Real / Misleading / Fake with confidence score (0-100%)
- Accepts text, URLs, and standalone claims
AI-Powered Insights
- Fine-tuned RoBERTa classifier (97%+ accuracy on 80K+ articles)
- Sentiment & sensationalism detection (VADER + custom patterns)
- Source credibility scoring (520+ domain trust database)
- Real-time fact-check cross-referencing via Google Fact Check Tools API
- Clickbait detection with headline-body semantic mismatch scoring
- Explainable AI — LIME word-level highlights + Claude natural-language explanations
- Multilingual support with language auto-detection (12 languages)
User Experience
- Dashboard with verdict distribution charts, trend analysis, and flagged sources
- Analysis history with filtering and pagination
- Dark mode support
- User authentication (Google / GitHub OAuth)
- Community feedback system for model improvement
- Fully responsive design with mobile hamburger menu
- SEO optimized with Open Graph tags
- Custom 404 and error pages
Developer Tools
- RESTful API with interactive Swagger documentation (
/docs) - Model comparison page — side-by-side RoBERTa vs baseline with inference times
- Chrome extension (Manifest V3) with popup and floating article button
- Docker deployment with PostgreSQL
| Layer | Technologies |
|---|---|
| Frontend | Next.js 16, React 19, Tailwind CSS 4, shadcn/ui, Recharts |
| Backend | Python 3.12, FastAPI, SQLAlchemy, Alembic |
| ML / NLP | RoBERTa (Hugging Face Transformers), scikit-learn, VADER, LIME |
| LLM | Claude API (explanation generation) |
| Database | PostgreSQL (prod), SQLite (dev) |
| Auth | NextAuth.js v5 (Google + GitHub OAuth) |
| Deployment | Docker, Vercel (frontend), Railway/Render (backend) |
┌─────────────────────┐ REST API ┌──────────────────────────────┐
│ │ ◄──────────────► │ │
│ Next.js Frontend │ │ FastAPI Backend │
│ │ │ │
│ - Analysis Form │ │ ┌────────────────────────┐ │
│ - Results Page │ │ │ Analysis Pipeline │ │
│ - Dashboard │ │ │ │ │
│ - History │ │ │ 1. URL Scraping │ │
│ - Model Compare │ │ │ 2. Language Detection │ │
│ - Auth (NextAuth) │ │ │ 3. Clickbait Check │ │
│ - Dark Mode │ │ │ 4. RoBERTa Classify │ │
│ │ │ │ 5. Sentiment Analysis │ │
└─────────────────────┘ │ │ 6. Source Credibility │ │
│ │ 7. Fact-Check API │ │
┌─────────────────────┐ │ │ 8. LIME Explainability│ │
│ Chrome Extension │ ────────────────►│ │ 9. Claude Explanation │ │
│ (Manifest V3) │ │ │ 10. Score & Verdict │ │
└─────────────────────┘ │ └────────────────────────┘ │
│ │
│ PostgreSQL Claude API │
│ Google Fact Check API │
└──────────────────────────────┘
- Node.js 18+
- Python 3.12+
- PostgreSQL 15+ (or use Docker)
git clone https://github.com/muhammadrakib2299/fake-news-detector_AI.git
cd fake-news-detector_AIdocker-compose up -d dbcd backend
# Create virtual environment
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt
# Configure environment
cp .env.example .env
# Edit .env with your database URL, API keys, etc.
# Run database migrations
alembic upgrade head
# Start the server
uvicorn app.main:app --reload --port 8000API available at http://localhost:8000 | Swagger docs at http://localhost:8000/docs
cd frontend
# Install dependencies
npm install
# Configure environment
cp .env.example .env.local
# Edit .env.local with your backend URL and OAuth credentials
# Start the development server
npm run devApp available at http://localhost:3000
docker-compose up -dThis starts PostgreSQL + Backend. Deploy frontend to Vercel or run locally.
fake-news-detector/
├── frontend/ # Next.js 16 application
│ └── src/
│ ├── app/ # App Router pages
│ │ ├── page.tsx # Landing / analysis input
│ │ ├── results/[id]/ # Analysis results
│ │ ├── dashboard/ # Statistics & charts
│ │ ├── history/ # Analysis history
│ │ ├── compare/ # Model comparison
│ │ ├── auth/signin/ # OAuth sign-in
│ │ ├── not-found.tsx # 404 page
│ │ └── error.tsx # Error boundary
│ ├── components/ # React components
│ │ ├── verdict-card.tsx
│ │ ├── score-breakdown.tsx
│ │ ├── explainability-report.tsx
│ │ ├── clickbait-display.tsx
│ │ ├── sentiment-display.tsx
│ │ ├── credibility-badge.tsx
│ │ ├── fact-check-section.tsx
│ │ └── header.tsx # Responsive nav
│ └── lib/
│ └── api.ts # Backend API client
│
├── backend/ # FastAPI application
│ ├── app/
│ │ ├── main.py # App entry point
│ │ ├── config.py # Environment config
│ │ ├── models.py # SQLAlchemy models
│ │ ├── schemas.py # Pydantic schemas
│ │ ├── routers/
│ │ │ └── analyze.py # All API endpoints
│ │ └── services/
│ │ ├── pipeline.py # Orchestrates all services
│ │ ├── classifier.py # RoBERTa + baseline + XLM-RoBERTa
│ │ ├── sentiment.py # VADER + sensationalism
│ │ ├── credibility.py # 520-domain trust database
│ │ ├── fact_checker.py # Google Fact Check API
│ │ ├── explainer.py # LIME + Claude explanations
│ │ ├── clickbait.py # Headline-body mismatch
│ │ ├── language.py # Language detection
│ │ └── scraper.py # URL article extraction
│ ├── ml/
│ │ ├── train_baseline.py # TF-IDF + LogReg training
│ │ ├── train_roberta.py # RoBERTa fine-tuning
│ │ └── models/ # Saved model weights
│ ├── alembic/ # Database migrations
│ ├── Dockerfile
│ ├── Procfile
│ ├── railway.json
│ └── render.yaml
│
├── extension/ # Chrome Extension (Manifest V3)
│ ├── manifest.json
│ ├── popup.html/css/js # Extension popup UI
│ ├── content.js/css # Floating "Verify" button
│ └── icons/
│
├── notebooks/ # Jupyter notebooks
│ ├── 01_data_exploration.ipynb
│ ├── 02_baseline_model.ipynb
│ └── 03_roberta_finetuning.ipynb
│
├── data/ # Training datasets
├── docker-compose.yml
└── README.md
POST /analyze
Content-Type: application/json
{
"content": "Breaking: Scientists confirm the earth is flat according to new study",
"input_type": "text"
}Response:
{
"id": "a1b2c3d4-...",
"verdict": "Fake",
"confidence": 0.92,
"final_score": 78.5,
"input_type": "text",
"model_used": "roberta",
"classification": {
"verdict": "Fake",
"fake_probability": 0.92,
"real_probability": 0.08,
"model": "roberta"
},
"sentiment": {
"vader_compound": -0.34,
"sensationalism_score": 0.71,
"sentiment_score": 0.65
},
"credibility": { "domain": null, "credibility_score": 0.5 },
"fact_check": { "has_matches": true, "match_count": 2, "fact_check_score": 0.85 },
"explainability": {
"highlights": [
{ "text": "confirm", "weight": 0.045, "signal": "fake" },
{ "text": "flat", "weight": 0.038, "signal": "fake" }
],
"explanation": "This text was classified as likely fake due to...",
"method": "lime",
"available": true
},
"clickbait": { "available": true, "clickbait_score": 0.62, "mismatch_score": 45.2 },
"language": { "code": "en", "name": "English", "confidence": 0.85 }
}| Endpoint | Method | Description |
|---|---|---|
/analyze |
POST | Run full analysis pipeline |
/analyze/{id} |
GET | Retrieve a past analysis |
/compare |
POST | Compare all models side-by-side |
/history |
GET | Paginated analysis history |
/stats |
GET | Dashboard statistics |
/feedback/{id} |
POST | Submit verdict correction |
/health |
GET | Service health check |
/docs |
GET | Interactive Swagger documentation |
| Model | Accuracy | Precision | Recall | F1 Score |
|---|---|---|---|---|
| TF-IDF + Logistic Regression | 94.2% | 93.8% | 94.5% | 94.1% |
| Fine-tuned RoBERTa | 98.1% | 97.9% | 98.3% | 98.1% |
Evaluated on held-out test set from combined LIAR + ISOT + FakeNewsNet datasets (80K+ articles).
| Dataset | Articles | Source |
|---|---|---|
| LIAR | 12.8K | PolitiFact statements |
| ISOT Fake News | 44K | Reuters + unreliable sources |
| FakeNewsNet | 23K+ | PolitiFact + GossipCop |
final_score = 0.45 * classification + 0.20 * sentiment + 0.20 * credibility + 0.15 * fact_check
| Score Range | Verdict |
|---|---|
| 0 - 30 | Real |
| 30 - 65 | Misleading |
| 65 - 100 | Fake |
Weights are dynamically adjusted when source credibility data is unavailable (text-only input).
Load the extension/ folder as an unpacked extension in Chrome:
- Navigate to
chrome://extensions/ - Enable "Developer mode"
- Click "Load unpacked" and select the
extension/directory - The VerifyAI icon appears in the toolbar
Features:
- Popup with "This Page" and "Paste Text" tabs
- Floating "Verify" button on article pages with inline results
- Links to full report on the web app
- Phase 1 — Foundation & ML Model (22/22)
- Phase 2 — Core Features & Frontend (23/23)
- Phase 3 — Explainability, Dashboard & Polish (18/18)
- Phase 4 — Bonus Features (14/14)
- Deployment configuration
- Documentation
Contributions are welcome! Please follow these steps:
- Fork the repository
- Create your feature branch (
git checkout -b feature/amazing-feature) - Commit your changes (
git commit -m 'Add amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Open a Pull Request
This project is licensed under the MIT License — see the LICENSE file for details.
- Hugging Face for transformer models and the Transformers library
- Google Fact Check Tools for the fact-checking API
- LIAR Dataset by William Yang Wang
- ISOT Fake News Dataset by University of Victoria
- Anthropic Claude for explanation generation
- shadcn/ui for the component library
Built with AI, for truth.