GBERT: A hybrid deep learning model based on GPT-BERT for fake news detection

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Information Security

Background:

• The digital age facilitates global communication but also amplifies the spread of fake news.
• Fake news, defined as deliberately false information, poses significant risks to societal harmony, politics, economics, and public opinion.
• Bogus news detection is a critical research area for distinguishing authentic information from fabricated content.

Discussion:

• The proposed Generative Bidirectional Encoder Representations from Transformers (GBERT) framework integrates Generative Pre-trained Transformer (GPT) and Bidirectional Encoder Representations from Transformers (BERT) models.
• This hybrid approach capitalizes on BERT's deep contextual understanding and GPT's generative capabilities for robust text representation.
• The framework is designed to address the complex challenge of fake news classification in the digital landscape.

Key Insights:

• The GBERT framework achieved high performance metrics on two real-world benchmark corpora: 95.30% accuracy, 95.13% precision, 97.35% sensitivity, and a 96.23% F1 score.
• Fine-tuning both GPT and BERT models within the GBERT framework demonstrated significant effectiveness in fake news detection.
• Statistical tests confirm the framework's efficacy, highlighting its potential for real-world application.

Outlook:

• The GBERT framework presents a promising avenue for combating the global issue of fake news.
• Further research and development could lead to more sophisticated and widely deployable fake news detection systems.
• This approach has the potential to significantly contribute to a more trustworthy digital information ecosystem.