Misinformation Detector – Real-Time AI Claim Verification

This project is a real-time, explainable system designed to detect misinformation in AI-generated text. It combines the power of LLMs (like GPT-3.5) with fact-checking APIs, knowledge graph queries, and semantic similarity models to classify claims as either Reliable or Misinformation, along with a confidence score and detailed breakdown of reasoning.

Live Demo: Watch the 5-minute project demo on YouTube

---

Features

• Claim-level misinformation detection
• LLM-based semantic similarity (Sentence-BERT)
• Stance detection using DistilBERT sentiment analysis
• Fall-back logic when claims are novel or unseen
• Real-time classification via Streamlit app
• Detailed explanation of predictions (stance, source quality, similarity)
• Preloaded reliable claim set + Phase 5 embeddings

---

Demo Screenshots

Reliable Claim	Misinformation

---

How It Works

1. Input Claim
   The user submits a natural-language factual claim.

2. Semantic Matching & Verification
   System checks similarity to known verified claims and searches knowledge bases.

3. Fallback Model
   If no match is found, the system estimates stance, semantic consistency, and source quality.

4. Classification
   Outputs a label: Reliable ✅ or Misinformation ❌ with confidence %.

5. Explainability
   Shows why the claim was classified that way (stance score, source quality, etc.).

---

Getting Started Locally

Installation

1. Clone the repository:

   git clone https://github.com/rishika7006/misinformation-detector.git
   cd misinformation-detector

2. Create a virtual environment and activate it:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Place the following model and data files in the root directory:

   • production_model_v2.pkl
   • phase5_claims.json
   • phase5_embeddings.npy

Run the App

streamlit run app.py

Then visit http://localhost:8501 in your browser.

---

File Structure

.
├── app.py                     # Streamlit app entry point
├── requirements.txt           # Python dependencies
├── production_model_v2.pkl    # Trained classifier model
├── phase5_claims.json         # Pre-verified claims with metadata
├── phase5_embeddings.npy      # Embeddings for phase 5 verified claims
├── screenshots/               # App screenshots for README
└── README.md

---

Tech Stack

• Streamlit – for interactive web app
• Transformers (HuggingFace) – for stance detection
• Sentence-BERT (SBERT) – for semantic similarity
• scikit-learn – for final classification
• Wikidata / DBPedia – for future knowledge graph integration

---

Acknowledgements

• OpenAI GPT-3.5 (for claim extraction & guidance)
• HuggingFace Transformers
• Streamlit for fast UI development
• FEVER, LIAR datasets for reference
• Course: CS 6375 – Machine Learning
  Instructor: Dr. Wei Yang
  Institution: The University of Texas at Dallas

---

License

This project is for educational purposes only. All data and model outputs are used in accordance with academic fair use.