Enhanced Fake News Detection Using BERT-Attention Model and Comparative Algorithm Analysis

In the contemporary digital era, the pervasive dissemination of misinformation via social media and various online platforms poses a significant challenge to social governance, leading to information distortion and public misguidance. To enhance the efficacy and precision of identifying news authenticity, this paper conducts a comparative analysis of different machine learning and deep learning algorithms for fake news detection, utilizing the WELFake dataset. The study evaluates Logistic Regression, Decision Tree, Random Forest, LSTM, and BERT models, and subsequently introduces an attention mechanism to construct an enhanced BERT-Attention model, aimed at further improving predictive accuracy. Empirical results demonstrate that the BERT-Attention model outperforms all other models, achieving an accuracy of 99.11% on the test set. The research highlights that pre-trained language model incorporating the attention mechanism would offer significant advantages in text semantic understanding and fake news detection task. This study contributes to the field by demonstrating the effectiveness of advanced models in improving the reliability of news dissemination in digital communication.