VeriFi - Misinformation Classification Chrome Extension

Overview

VeriFi is a Chrome extension designed to detect misinformation and fact-check news articles on social media. It combines machine learning models and API-based verification techniques to provide users with a reliability score for news articles and posts, helping them make informed decisions about the authenticity of the information they consume.

How it Works

1. Tavily Analysis

• Debunked Keywords: Reduces score (-2 per keyword, -30 base penalty if found).
• Verified Content: Rewards score (+1 per supporting source, +15 base reward).

2. NewsAPI Scoring

• Credible Sources: +8 per article from Reuters/AP.
• Fact-Checking Websites: +12 per article from Snopes, FactCheck.org.
• Unverified Sources: -3 per unreliable article.

3. AI Model Scoring

• Predicts fake: Reduces score (-25 scaled by confidence level).
• Predicts credible: Increases score (+15 scaled by confidence level).

Confidence Score Calculation

1. Base Score: 50
2. Adjustments: Based on Tavily, NewsAPI, and AI model results.
3. Randomness: Small variances to avoid rigid thresholds.
4. Final Score: Clamped between 5-95, rounded to whole number.

Fake News Classification

• Threshold: 45 ±5 to avoid rigid boundaries.

Example Outcomes

Fake News

• Tavily: 5 debunking keywords (-40)
• AI Model: Predicts fake (90% confidence, -22.5)
• Final Score: 5 (Clamped) → Likely Fake

Credible News

• Tavily: 3 supporting sources (+18)
• NewsAPI: 2 credible sources (+16)
• AI Model: Predicts credible (80% confidence, +12)
• Final Score: 95 → Likely Credible

Borderline News

• Tavily: 1 supporting source (+16)
• NewsAPI: 1 unverified source (-3)
• AI Model: Predicts credible (50% confidence, +7.5)
• Final Score: 71 → Likely Credible

Fake-News-BERT Model

The Fake-News-BERT model is a deep learning model fine-tuned for misinformation detection. It was developed using the following approach:

• Dataset: The model was trained on a dataset of 72,134 news articles with 35,028 real and 37,106 fake news. Dataset contains four columns: Serial number (starting from 0); Title (about the text news heading); Text (about the news content); and Label (0 = fake and 1 = real).
• Preprocessing: Text cleaning, tokenization, and feature extraction were performed using NLP techniques.
• Model Training: A BERT-based transformer model was fine-tuned using supervised learning.
• Evaluation: The model was validated using accuracy, F1-score, and other performance metrics.

Features

• Real-time Fact-Checking: VeriFi analyzes news articles and social media posts in real time.
• Reliable Scoring System: The hybrid approach ensures accurate misinformation detection.
• Seamless Browser Integration: Works as a Chrome extension with a simple UI for quick insights.
• Secure & Efficient: Processes data securely without storing user information.

Installation

1. Clone the repository:

   git clone https://github.com/Dakshya52/verifi.git

2. Navigate to the directory:

   cd verifi/verifi

3. Install dependencies:

   npm install

4. Build the extension:

   npm run build

5. Load the extension in Chrome:
   • Open chrome://extensions/
   • Enable "Developer mode"
   • Click "Load unpacked"
   • Select the VeriFi/verifi/dist folder

Usage

• Once installed, you don't need to do anything further.
• Open X.com, and on the bottom left of every post, you will see a button.
• Click it to verify the credibility of the post—that's it!

Contributors

• Dakshya - GitHub | LinkedIn
• Dhruv - GitHub | LinkedIn

License

This project is open-source and available under the MIT License.