Fake news has emerged as a significant societal issue, eroding public trust in the government and news media, while also inciting public panic and triggering social chaos. However, relying solely on the news itself for fake news detection is inadequate due to its deceptive nature. To complement and enhance fake news detection, it is necessary to incorporate social context information related to the news. This paper proposes a fake news detection framework based on graph convolutional networks and attention mechanisms. By representing the news and its social context information (such as participating users and comments) as a heterogeneous graph, we employ meta-paths to capture the relationships between news-comments and news-users respectively. Graph convolutional networks are utilized to effectively extract meaningful social context information, while attention mechanisms are employed to learn the importance of various types of social context information, thereby acquiring informative representations of news node features. Experiments conducted on publicly available datasets validate the effectiveness of the proposed framework.
The wide dissemination of false news is increasingly threatening both individuals and society. Aiming at the problem of insufficient interpretability in false news detection task, the model TCIN based on collaborative attention mechanism is proposed. The model integrates the advantages of the existing dEFEND model and GCAN model. It captures more detailed semantic features through three attention layers (sentence-level cooperative attention layer, word-level cooperative attention layer and sentence-comment cooperative attention layer). The experimental results show that, compared with the model before combination, the accuracy of this model is 9.7% higher on the data set of Weibo-20 and on Twitter15 data set the accuracy is increased by 7.43%.