Leveraging BERT and GPT for Enhanced Fake News Detection and Analysis

This study presents an advanced hybrid verification system for fake news detection that integrates state-of-the-art natural language processing models with external verification mechanisms. Our approach implements two independent parallel processes: (1) internal classification analysis using pre-trained BERT and GPT models, and (2) external similarity verification using the Tavily API and Ollama LLM for cross-referencing with verified news sources. The proposed system addresses key limitations of existing approaches by implementing a comprehensive hybrid verification methodology that combines the contextual understanding capabilities of transformer models with external fact-checking mechanisms. Trained and tested on a comprehensive dataset of 72,134 real and fake news articles, the system achieves exceptional performance with 99.78% accuracy. The hybrid verification architecture enables real-time news monitoring and provides robust performance across various contexts through continuous adaptation to emerging threats. This methodology offers a dynamic, scalable solution that empowers organizations and individuals to effectively combat disinformation while maintaining high precision in classification tasks.