🛡️ FakeNewsDetector — System NLP

AI-powered fake news detection with credibility scoring, truth probability, timeline evolution & region-based source comparison.
Flask + TF-IDF + RandomForest + heuristic pattern engine + Plotly visualizations

Demo • Features • Screenshots • Quick Start • API • How it Works

---

📸 Screenshots

Home & Input

Home With News Input Text

Credibility Analysis

Gauges: Credibility / Truth Probability

Timeline & Metrics

24h credibility evolution (Plotly, date-axis, spline)

Detailed Metrics

Sentiment, Bias, Fact-check, Manipulative, Source Reliability

Source Comparison

Region-based 3-article generation (US/UK/EU/Asia)

AI Insights

Pattern matches, impossibilities, confidence & conclusion

Demo WorkFlow:

Paste → Analyze → Gauges + Timeline + Sources in < 1s

---

✨ Features

Category	What it does
🎨 Modern UI	Dark & darker mode, responsive grid, Plotly gauges + timeline, animated cards, loading overlay
🧠 Hybrid AI Engine	TF-IDF (5000, 1-3 gram) + RandomForest (100 trees, balanced) + 140+ regex fake patterns + 50+ scientific impossibilities + credibility indicators
📊 Credibility Score	0–100 → truth_probability = credibility . Boost for policy/economic official sources, penalty for clickbait/conspiracy/extraordinary claims
⏱️ Credibility Timeline	24h realistic drift simulation (starts noisy → converges to final score), date-axis, dynamic Y-range, no duplicate labels
🔍 Detailed Metrics	sentiment (-100→100), bias (0→100), fact_check_confidence, manipulative_language, source_reliability, linguistic_complexity
🌍 Source Comparison	Auto region detection (US/UK/EU/Asia keywords) → 3 generated articles with bias-titles (center-left/center/center-right) + summaries
🤖 AI Insights	Lists matched patterns, impossible claims, credibility indicators, model confidence & human-readable conclusion
🔌 REST API	POST /analyze, POST /api/v1/analyze, GET /api/v1/sources + CORS

---

🏗️ Architecture

[ User Text ] → [ Flask app.py ] → [ FakeNewsDetector (models/detector.py) ]
                                    ├─ FakeNewsAI (fake_news_ai.py) → TF-IDF + RF → credibility + patterns
                                    ├─ CredibilityMetrics (utils/metrics.py) → sentiment/bias/manipulative
                                    └─ NewsSourceComparison (utils/news_sources.py) → region + 3 articles
                                              ↓
                              [ JSON ] → [ Plotly.js gauges + timeline + cards (static/js/script.js) ]

---

🧰 Tech Stack

Layer	Tools
Backend	Python 3.8+, Flask 3.x, Flask-CORS, scikit-learn, NLTK, NumPy, Pandas, Joblib
NLP	TF-IDF Vectorizer, WordNet Lemmatizer, stopwords, punkt, Regex heuristics
Frontend	HTML5, CSS3 (dark theme, grid), Vanilla JS, Plotly 2.14, Font Awesome 6
Models	models/fake_news_model.joblib (RF), models/tfidf_vectorizer.joblib

---

📁 Project Structure

FakeNewsDetector System NLP/
├── app.py                 # Flask routes: /, /analyze, /api/v1/*
├── requirements.txt
├── models/
│   ├── detector.py        # Orchestrator: score fusion + timeline + sources
│   ├── fake_news_ai.py    # RF + TF-IDF + patterns/impossibilities
│   ├── fake_news_model.joblib
│   └── tfidf_vectorizer.joblib
├── utils/
│   ├── metrics.py         # Sentiment/bias/manipulative/source-reliability
│   └── news_sources.py    # Trusted sources by region + article generator
├── templates/
│   └── index.html         # Dark/blue dashboard
├── static/
│   ├── css/style.css      # 1200+ lines, responsive, card animations
│   └── js/script.js       # Gauges, timeline (date-axis, purge), metrics, sources
└── Images/
    
        ├── hero.png
        ├── home.png
        ├── analysis.png
        ├── timeline.png
        ├── metrics.png
        ├── sources.png
        ├── insights.png
        ├── architecture.png
        └── demo.gif

---

🚀 Quick Start

1. Clone

git clone https://github.com/Sword4234/FakeNewsDetector-System-NLP.git
cd FakeNewsDetector-System-NLP

2. Create venv & Install

python -m venv venv
# Windows
venv\Scripts\activate
# macOS/Linux
source venv/bin/activate

pip install flask flask-cors nltk scikit-learn numpy pandas joblib
# or
pip install -r requirements.txt

3. NLTK data (auto-download on first run, or manual)

python -c "import nltk; nltk.download('punkt'); nltk.download('punkt_tab'); nltk.download('stopwords'); nltk.download('wordnet')"

If behind proxy: set NLTK_ALLOW_PROXIED_URLOPEN=1 (Windows) / export NLTK_ALLOW_PROXIED_URLOPEN=1

4. Run

python app.py
# → * Running on http://127.0.0.1:5000
# → Debug + watchdog auto-reload

Open http://127.0.0.1:5000 → paste news → Analyze. Try Try Sample News / Try Fake News Sample buttons.

---

🔌 API Reference

POST /analyze & POST /api/v1/analyze

Request:

{ "news_text": "NASA rover finds evidence of water on Mars, according to peer-reviewed research..." }

Response 200:

{
  "credibility_score": 81.9,
  "truth_probability": 81.9,
  "detailed_metrics": {
    "sentiment_score": -3.7,
    "bias_score": 4.9,
    "fact_check_confidence": 53.9,
    "manipulative_language": 1.7,
    "source_reliability": 50.0,
    "linguistic_complexity": 61.8
  },
  "timeline_data": [
    { "timestamp": "2026-08-31T20:47:14.508464", "score": 74.2 },
    { "timestamp": "2026-09-01T19:47:14.508464", "score": 82.6 }
  ],
  "source_comparison": {
    "region": "us",
    "articles": [
      { "source": "Reuters", "title": "Fact Check: ...", "summary": "...", "date": "August 30, 2026", "url": "https://www.reuters.com/article/...", "bias": "center" }
    ]
  },
  "ai_insights": {
    "pattern_matches": [],
    "scientific_impossibilities": [],
    "credibility_indicators": ["peer-reviewed", "research (published|conducted) (in|by)"],
    "model_confidence": 66,
    "conclusion": "This content appears to be credible with no suspicious patterns detected."
  }
}

Low credibility: source_comparison: [] → frontend shows warning “Source comparison not available for low credibility (<80%)”.

GET /api/v1/sources

curl http://127.0.0.1:5000/api/v1/sources

{
  "us": [{"name":"The New York Times","url":"https://www.nytimes.com","bias":"center-left"}, ...],
  "uk": [...], "eu": [...], "asia": [...]
}

cURL examples:

# Credible
curl -X POST http://127.0.0.1:5000/api/v1/analyze -H "Content-Type: application/json" -d "{\"news_text\":\"Government announces fiscal policy and trade agreement according to official ministry sources.\"}"

# Fake
curl -X POST http://127.0.0.1:5000/api/v1/analyze -H "Content-Type: application/json" -d "{\"news_text\":\"NASA confirms moon made of cheese miracle cure shocking truth\"}"

---

🧠 How it Works

1. FakeNewsAI.analyze_text(text) (models/fake_news_ai.py:28)

• Preprocess: word_tokenize → lower → remove stopwords → WordNetLemmatizer
• Vectorize TfidfVectorizer(max_features=5000, ngram_range=(1,3)) → RandomForest (100 trees, balanced) → fake_news_probability
• Regex checks: 140+ fake patterns (miracle cure, doctors hate, aliens landed, breaking.*died), 50+ impossibilities (moon made of cheese, human landed on sun), credibility indicators (peer-reviewed, official statement, ministry of)
• Penalties/boosts: known fakes → credibility 10, sun landing → ≤8, pattern → -15 each (cap 75), impossibility → -30 (cap 90, max 25), extraordinary claims without 5 indicators → cap 35
• Stats: word/sentence counts, linguistic_complexity

2. FakeNewsDetector.analyze_news (models/detector.py:14)

• Fuses credibility_score == truth_probability, boosts policy/economic news if ≥2 terms + no patterns → ≥85
• Generates timeline_data via generate_timeline_data(base) (24h drift toward base)
• Region via keyword counts → get_related_articles if credibility ≥80 else []
• Merges source_credibility, fact_density + CredibilityMetrics scores

3. CredibilityMetrics (utils/metrics.py:15)

• Sentiment via positive/negative lexicons, bias via opinion words + first-person pronouns, fact-check via numbers/quotes/citations/dates, manipulative via clickbait regex, source reliability via domain whitelist

---

✅ Results & Testing

Input	Credibility	Pattern	Sources
NASA Perseverance peer-reviewed...	81.9	[]	3 articles (us)
Moon made of cheese miracle cure	0	miracle cure, doctors hate	[] (warning)
Fiscal policy trade bilateral... official ministry	100	[]	3 articles

Timeline: 24 pts, ISO timestamps, range ≈10-19 for high cred (visible spline, not flat), dynamic Y [min-8, max+8], purge fix removes double-line.

---

🚢 Deployment

Production (no debug/watchdog):

# app.py:66
if __name__ == '__main__':
    app.run(host='0.0.0.0', port=5000, debug=False)

Use waitress/gunicorn:

pip install waitress
waitress-serve --host=0.0.0.0 --port=5000 app:app

---

🤝 Contributing

PRs welcome! Please:

1. Fork → git checkout -b feat/your-feature
2. Use black/flake8 for Python, keep script.js vanilla
3. Test both credible & fake samples
4. PR description: what + why + screenshot

---

📄 License

MIT © 2026 Sword4234 — see LICENSE. No paper/IPR/patent — free for open-source, commercial, academic use with attribution.

---

👤 Author

Sword4234 / YashD — GitHub

Built with ❤️ for media literacy. If this helps your research, please ⭐ star the repo!