Enhancing fake news detection with transformer-based deep learning: A multidisciplinary approach

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Information Science

Background:

• Fake news dissemination challenges digital information integrity and public trust.
• Automated detection mechanisms are crucial for combating misinformation.
• Existing methods may not fully capture the nuances of fabricated content.

Purpose of the Study:

• To propose a robust fake news detection framework using a transformer-based architecture.
• To enhance the Bidirectional Encoder Representations from Transformers (BERT) model with progressive training.
• To improve the automated identification of linguistic indicators differentiating real news from fake news.

Main Methods:

• Developed a fake news detection framework utilizing a transformer-based architecture.
• Applied and enhanced the Bidirectional Encoder Representations from Transformers (BERT) model.
• Implemented a progressive training methodology for incremental learning.
• Evaluated the framework on the large-scale WELFake dataset (72,134 articles).

Main Results:

• The enhanced BERT framework achieved high performance metrics: 95.3% accuracy, 0.953 F1-score, 0.952 precision, and 0.954 recall.
• The model demonstrated superior performance compared to traditional machine learning classifiers.
• The approach significantly outperformed other standard transformer-based implementations.
• The progressive training effectively captured complex contextual dependencies in text.

Conclusions:

• The enhanced BERT framework presents a powerful and scalable solution for fake news detection.
• The progressive training methodology enhances the model's ability to discern subtle linguistic differences.
• This approach offers a significant advancement in the automated fight against digital misinformation.
• The findings highlight the potential of advanced transformer models in maintaining information integrity.