iceman2434/xlm-roberta-base-fake-news-detection-tl

Tagalog Fake News Detection Model

Overview

This project implements a fake news detection model for Tagalog/Filipino using the XLM-RoBERTa base model with an accuracy of 95.46%.

Dataset

• Total Size: 18,522 samples
• Composition: 50/50 split of real and fake news
• Languages: Filipino, English

Dataset Split

• Train Set: ~12,968 samples
• Validation Set: ~2,784 samples
• Test Set: ~2,770 samples

Performance Metrics (on Evaluation Set)

• Accuracy: 95.46%
• F1 Score: 95.40%
• Precision: 95.40%
• Recall: 95.40%

Data Sources

The model was trained on a combined dataset from two primary sources:

1. Fake News Filipino Dataset

   • 3,206 rows used

2. Philippine Fake News Corpus

   • 15,312 rows used out of 22,458 available