Fake news detection based on dual-channel graph convolutional attention network

Fake news detection has attracted significant attention since the spread of fake news on social media has affected the media’s credibility. Some existing fake news detection models only applied news content features as input. They treated the extracted news and user features as text but ignored the interaction between the various components of the news. Furthermore, the news dissemination information was not captured effectively. To address these problems, we propose a dual-channel graph convolutional attention network with dynamic feature fusion named DGCAN. News on social media does not exist in isolation, but is connected and interactive with each other. Therefore, first, we construct a heterogeneous graph to model the semantic information between source tweets and words and the spread structure information between source tweets and users, respectively. Second, we capture the semantic information and dissemination structure information of news at the same time with a designed dual-channel graph convolutional network. Third, we integrate the graph attention mechanism to learn the importance of tweet, user, and word nodes in heterogeneous graphs. Finally, we apply the attention mechanism between sub-channels to learn the semantic and spread features. Experiments on two real-world datasets demonstrate that the model proposed in this paper achieves state-of-the-art performance.