fact-check1-v2-calibrated

This is the "Brain 2" model from the Credo AI project, a powerful binary (FAKE/REAL) news classifier.

This model is a fine-tuned version of microsoft/deberta-v3-large. It was initially trained on a large, diverse corpus of over 50,000 news articles to become a "Fake News Specialist."

Version 2: The Calibration Update

This v2 version addresses a key limitation of the original model: a bias towards classifying simple, factual statements as FAKE. The original model (Arko007/fact-check1-v1) was so specialized in detecting the patterns of sensationalist news that it became overly suspicious of clean, encyclopedic facts.

This model has undergone a surgical fine-tuning (calibration) on a small, balanced dataset of:

• 5,000 pristine scientific facts from the allenai/scifact dataset.
• 5,000 high-quality fake news examples from the mrisdal/fake-news dataset.

This calibration has significantly improved the model's real-world accuracy and corrected its bias, allowing it to correctly identify both complex fake news and simple factual statements.

Intended Use

This model is intended to be used as a fast and powerful first-line filter for English-language news headlines and short texts.

from transformers import pipeline

classifier = pipeline("text-classification", model="Arko007/fact-check1-v2-calibrated")

# Example 1: Simple Fact
text1 = "The sun rises in the east."
print(classifier(text1))
# Expected output: [{'label': 'REAL', 'score': ...}]

# Example 2: Fake News
text2 = "BREAKING: Scientists confirm lizards are secretly running the government."
print(classifier(text2))
# Expected output: [{'label': 'FAKE', 'score': ...}]

Limitations

The model's training data was primarily focused on political news. While the calibration has broadened its understanding, it may still perform with lower accuracy on highly specialized domains like finance or medicine.