FactFlow

Your Personal AI Fact Checker — Built as a Multi-layered Browser Extension

🔍 In an age of misinformation, FactFlow empowers users to navigate online news with confidence.

FactFlow is an intelligent browser extension designed to analyze and validate news articles in real-time.
By combining the power of Natural Language Processing, source credibility checks, and AI-based cross-referencing,
FactFlow delivers a layered analysis to help you identify fake, misleading, or unverifiable content — directly as you browse.

Whether it's political headlines or trending stories, FactFlow helps you verify before you trust.

🔧 Key Features

• 🧠 3-Layered Verification System

  • Pattern-based ML model trained on LIAR dataset
  • Source credibility score using MBFC database
  • Factual cross-checking with real-time LLM support

• ⚡ One-Click Analysis

  • Scrapes and processes the current web page automatically

• 🟩 Credibility Verdict Bar

  • Displays clear verdicts like: Fake, Soft Fake, Likely Real, Uncertain

• 🌐 Chrome Extension UI

  • Minimalistic interface built with React + Tailwind + ShadCN
  • Circular animated progress loader and hover effects

• 📡 FastAPI Backend

  • Unified API that integrates model inference, source scoring, and LLM calls

⚙️ Built With

💻 Frontend

🧠 Backend

📊 Machine Learning

🧩 Tools & Integrations

🧪 Multi-Layered Verification Pipeline

FactFlow analyzes content using three distinct yet complementary layers:

1️⃣ Pattern-Based Detection (ML)

• Uses a fine-tuned RoBERTa-Large model trained on the LIAR dataset
• Analyzes language style, semantic patterns, exaggeration, and bias indicators

2️⃣ Source Credibility Check

• Looks up the article’s source in the Media Bias/Fact Check (MBFC) database
• Uses source credibility scores and bias ratings to assess trustworthiness

3️⃣ Factual Cross-Reference

• Utilizes the Gemini LLM API to verify key claims
• Checks if claims are supported or contradicted by factual sources across the web

✅ Final Verdicts like Fake, Soft Fake, or Likely Real are assigned by a custom decision engine that aggregates all three layers.

🖼️ Screenshots, Demo

       

🎥 The extension scans the page, runs all 3 verification layers in real-time, and displays a final verdict with animated feedback and progress tracking.

📊 Performance & Results

🔍 Pattern-Based Model

• Model: Fine-tuned BERT on LIAR Dataset
• Accuracy: 87.3%
• F1 Score: 0.88
• Data: 15k labeled political statements

✅ Verdict Mapping

The final credibility verdict is determined by a custom decision engine that synthesizes all three layers:

Layer	Signal	Outcome Example
Pattern-Based	FAKE	🟧 Soft Fake
Source Score < 20	Questionable or Satire	🟥 Fake
Cross-Reference	Contradicted key claims	🟥 Fake
All Layers Agree (Real)	Factual, Credible, Clean	🟩 Likely Real
Conflicting Layers	Mixed results or missing	🟨 Uncertain

🧠 These verdicts are dynamically computed using a hybrid rule-based and AI-supported decision engine.

📖 Academic Recognition

📝 FactFlow was presented at the
IEEE 16th International Conference on Computing, Communication and Networking Technologies (ICCCNT 2025)
📍 IIT Indore, India
📅 July 2025

🎓 The paper introduces FactFlow as a novel browser-based misinformation detection framework, combining stylistic pattern analysis, source credibility evaluation, and content-aware LLM verification.

• 🏅 Status: Accepted for publication in IEEE Xplore
• 📌 Title: FactFlow: A Multi-Layered Fake News Detection System Using Pattern-Based and Content-Aware Machine Learning
• 🔗 IEEE Conference Website

Full paper coming soon to IEEE Xplore Digital Library 📚