📰 NewsLens – Fake News Detector

A full‑stack system that analyzes news articles to estimate whether content is likely real or fake using an NLP pipeline and machine learning.

📚 Table of Contents

• Overview
• Features
• Tech Stack
• Architecture & Flow
• Project Structure
• Requirements
• Setup & Run
• Configuration
• API Reference
• Data Model
• Model & Preprocessing
• Retraining Workflow
• Troubleshooting
• Security & Production Notes

---

Overview

Input

• News title & body text or a source URL to scrape.

Processing

• Clean & lemmatize text
• Vectorize with TF‑IDF
• Classify using a trained ML model

Output

• label: real | fake
• confidence: probability score

All predictions are stored in MongoDB for history and analytics.

Features

• 🎨 Responsive React UI with a polished dark theme

• 🔗 URL scraping using Axios + Cheerio

• ⚡ Real‑time classification (Node → Python subprocess)

• 🗂 Persistent prediction history (MongoDB)

• 📊 Analytics dashboard:

  • Total predictions
  • Real/Fake ratio
  • Average confidence
  • Confidence histogram
  • Top source domains

• 🚀 One‑command setup to install and run everything

• 🔁 Retraining endpoint to rebuild the model from CSV

---

Tech Stack

Frontend

• React
• Vite
• Tailwind CSS
• Framer Motion
• lucide‑react icons

Backend

• Node.js
• Express
• Axios
• Cheerio
• MongoDB (native driver)

Machine Learning

• Python
• scikit‑learn
• NLTK
• joblib

---

Architecture & Flow

---

## Project Structure

NewsLens/ ├─ client/ # React frontend │ ├─ pages/ │ │ ├─ Input (/) │ │ ├─ Result (/result) │ │ └─ Analytics (/analytics) │ ├─ server/ # Express backend │ ├─ index.js │ └─ routes/ │ ├─ predict.js │ ├─ history.js │ ├─ analytics.js │ ├─ retrain.js │ └─ health.js │ ├─ model/ # Python ML pipeline │ ├─ src/ │ │ ├─ preprocess.py │ │ ├─ train.py │ │ └─ predict.py │ ├─ saved_models/ │ │ ├─ model.pkl │ │ ├─ vectorizer.pkl │ │ └─ metrics.json │ └─ requirements.txt │ └─ README.md

---

## Requirements

* **Node.js** ≥ 18
* **npm**
* **Python** ≥ 3.9
* **pip**
* **MongoDB** (local or remote)

---

## Setup & Run

From the project root:

```bash
npm start

This command will:

• Install frontend & backend dependencies

• Install Python requirements from model/requirements.txt

• Start:

  • Backend → http://localhost:3001
  • Frontend → http://localhost:5173

---

Configuration

Environment variables (optional):

MONGO_URI=mongodb://127.0.0.1:27017/NewsLens
MONGO_DB=NewsLens
WELFAKE_CSV=WELFake_Dataset.csv

---

API Reference

POST /predict

Body

{
  "title": "optional",
  "text": "optional",
  "url": "optional"
}

Response

{
  "status": "ok",
  "label": "real | fake",
  "confidence": 0.93
}

---

GET /history?limit=50&skip=0

{
  "status": "ok",
  "items": [ { "input": {}, "result": {}, "ts": 0, "domain": "" } ]
}

---

GET /analytics

{
  "status": "ok",
  "totals": {},
  "time": {},
  "top_domains": [],
  "confidence_hist": []
}

---

POST /retrain

{
  "status": "ok",
  "best_model": "LogisticRegression",
  "metrics": {}
}

---

GET /health

{ "status": "ok" }

---

Data Model

history collection

{
  "input": { "title": "...", "url": "...", "text": "..." },
  "result": { "label": "real|fake", "confidence": 0.93 },
  "ts": 1736450000000,
  "domain": "example.com"
}

analytics document

{
  "_id": "global",
  "total": 120,
  "real": 78,
  "fake": 42,
  "sumConfidence": 98.5,
  "updatedAt": "2026-01-09T16:05:00Z"
}

---

Model & Preprocessing

• Label mapping: 0 → real, 1 → fake

• Text preprocessing:

  • Lowercasing
  • HTML & URL removal
  • Non‑alphabetic filtering
  • Stop‑word removal
  • Lemmatization (WordNet)

• Vectorization:

  • TF‑IDF (uni‑grams & bi‑grams)
  • max_features = 50000
  • min_df = 2
  • max_df = 0.95

• Models evaluated:

  • Multinomial Naive Bayes
  • Logistic Regression

• Model selection: Best F1 score wins

Artifacts are saved in model/saved_models/ and loaded at inference time.

---

Retraining Workflow

• Call POST /retrain or run:

python model/src/train.py

• Ensure WELFAKE_CSV points to the dataset
• New artifacts overwrite old ones
• All future predictions use the updated model

---

Troubleshooting

• pip install errors:

  python -m pip install -r model/requirements.txt

• MongoDB connection issues:

  • Verify MongoDB is running
  • Check MONGO_URI

• NLTK WordNet missing:

  nltk.download('wordnet')

• CORS issues:

  • CORS is enabled by default in server/index.js

---

Security & Production Notes

• Enable HTTPS
• Add authentication & rate limiting
• Redact or hash sensitive text before storage
• Use managed MongoDB (Atlas) in production
• Store secrets using environment variables

---

🚀 NewsLens — turning misinformation into insight.