Sentiment-Aware Fake News Detection on Social Media with Hypergraph Attention Networks

The rapid development of social media makes it easy for people to acquire information while also providing a platform for publishing and spreading fake news. Fake news brings plenty of explicit and implicit risks to social stability, making fake news detection an issue that deserves attention. Recent methods based on graph neural networks (GNN) achieve impressive results in fake news detection, but their performance is still limited in practice due to the absence of high-order relations between nodes. In this paper, we propose a Sentiment-Aware Hypergraph Attention Network (SA-HyperGAT) for fake news detection. SA-HyperGAT can better leverage different kinds of information from news contents and user comments with hypergraphs, which can capture higher-order dependency between words and sentences compared with general graphs. Specifically, we first construct two hypergraphs with distinct types of nodes and hyperedges to utilize structural information of news contents and sentimental information of user comments. Then we adopt a hypergraph attention network with a dual attention mechanism to learn the composed representations of two hypergraphs for the final prediction. Our proposed SA-HyperGAT outperforms competitive baselines on two real-world datasets. Extensive experimental results prove the effectiveness of each component in SA-HyperGAT.