Rezvan MohammadiBaghmolaei

In the digital age, the widespread dissemination of fake news on social networks has posed serious challenges to public trust as well as social, political, and economic decision-making. This research addresses the automatic detection of fake news by focusing on sentiment analysis and the reduction of emotional noise in news content and user reactions. Despite recent advances in applying sentiment analysis in this field, several challenges remain such as model robustness against emotional noise and the effective utilization of adversarial training and reinforcement learning. The primary aim of this study is to design and implement an innovative hybrid model that improves the accuracy and generalizability of fake news detection by emphasizing the emotional aspects of news content and their influence on users, while leveraging deep learning algorithms. In this study, the Fakeddit, Snopes, and PHEME datasets were employed. The results demonstrated that the proposed model performed well across different datasets and exhibited acceptable robustness against emotional noise. These findings suggest that integrating sentiment analysis, deep learning, adversarial learning, and reinforcement learning can enhance the accuracy and robustness of fake news detection systems against sentiment noise and be effective for real-world applications such as social network monitoring.