adityaG04/fake-news-bert-finetuned

🔬 Fake News Detector (Fine-Tuned RoBERTa)

A RoBERTa-based fake news classifier fine-tuned using transfer learning.

Model Details

Property	Value
Base Model	jy46604790/Fake-News-Bert-Detect
Architecture	RoBERTa-base (125M params)
Dataset	GonzaloA/fake_news (~40K articles)
Accuracy	98%
Labels	LABEL_0 = Fake, LABEL_1 = Real
Max Length	512 tokens

Training Details

• Method: Transfer learning with layer freezing
• Frozen Layers: Layers 0-5 (embeddings + first 6 encoder layers)
• Trainable Layers: Layers 6-11 + Classification head
• Learning Rate: 2e-5 with cosine scheduler
• Batch Size: 32 (effective)
• Epochs: 5 with early stopping (patience=2)
• Optimizer: AdamW with weight decay 0.01

Usage

from transformers import pipeline

classifier = pipeline("text-classification", model="adityaG04/fake-news-bert-finetuned")

# Test with a real news article
result = classifier("The Federal Reserve raised interest rates by 0.25 percentage points on Wednesday.")
print(result)
# [{'label': 'LABEL_1', 'score': 0.99}]  → REAL ✅

# Test with a fake article
result = classifier("BREAKING: Scientists confirm drinking bleach cures all diseases!")
print(result)
# [{'label': 'LABEL_0', 'score': 0.98}]  → FAKE 🔴