youtube-fact-checker-roberta

YouTube Fact Checker - Claim Classification Model

A fine-tuned RoBERTa model for identifying factual claims in YouTube transcripts.

This model is part of the YouTube Fact Checker project, where transcript sentences are first classified and only factual or disputed claims are forwarded to an LLM for verification.

---

Model Overview

Architecture: RoBERTa

Task: Text Classification

Framework: Hugging Face Transformers

Language: English

The model classifies each sentence into one of three categories:

---

Example

Input

Water boils at 100°C at sea level.

Output

FACTUAL_CLAIM

---

Input

The Earth is flat.

Output

DISPUTED_CLAIM

---

Input

I really enjoyed this video.

Output

NOT_A_CLAIM

---

Intended Use

This model is designed as the first stage of an AI-powered fact-checking pipeline.

Pipeline:

YouTube Video
      │
      ▼
Transcript Extraction
      │
      ▼
Sentence Segmentation
      │
      ▼
RoBERTa Claim Classifier
      │
      ▼
LLM Fact Verification
      │
      ▼
HTML / PDF Report

The classifier helps reduce unnecessary LLM calls by filtering out sentences that are not factual claims.

---

Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="AyushRawat1718/youtube-fact-checker-roberta"
)

result = classifier("The moon landing was faked.")

print(result)

---

Training

The model was fine-tuned for claim classification using a custom labeled dataset.

The objective was to distinguish between:

• Non-claims
• Verifiable factual claims
• Disputed claims

---

Limitations

• Designed primarily for English text.
• Performs best on transcript-style sentences.
• Does not perform factual verification.
• Final verification should be performed using an external LLM.

---

Related Project

This model is used in the YouTube Fact Checker project.

Features include:

• YouTube transcript extraction
• Sentence segmentation using spaCy
• Claim classification using this RoBERTa model
• AI-powered fact verification (Gemini + Groq)
• HTML/PDF report generation
• FastAPI backend
• React frontend (upcoming)

---

Author

Ayush Rawat

B.Tech Computer Science Engineering

AI • NLP • Full Stack Development

GitHub: https://github.com/AyushRawat1718

---

⭐ If you find this model useful, consider starring the GitHub repository.