Semantic-aware fake news detection with heterogeneous graph attention

Jing Chen, Gang Zhou, Mingjing Lan, Shiyu Wang, Shunhang Li, Jicang Lu

Abstract: Social media unintentionally helps spread fake news, which threatens public trust and social stability. Existing detection methods typically model the sequential or co-occurrence relationships between words in news texts to extract semantic features of fake news content. Although effective, they still struggle with perceiving the fine-grained semantics of news content. They often overlook deeper semantic dependencies and fail to consider the varying contributions of different types of words to semantic expression. To tackle these challenges, we present a semantic-aware fake news detection method with heterogeneous graph attention (SHGAT). First, we construct a heterogeneous textual graph to capture the internal semantic dependencies and external knowledge associations in news posts by integrating relationships between semantic elements such as entity words, pattern words, and concept descriptions. To further enhance the fine-grained semantic perception of post content, we design a heterogeneous graph-based encoder with a dual-level attention mechanism to learn post patterns and knowledge-enhanced entity semantics. Additionally, an adaptive feature aggregation module is introduced to automatically select and optimize feature combinations. Extensive experiments conducted on two benchmark datasets demonstrate that our model achieves state-of-the-art accuracy, outperforming existing methods in the literature by approximately 2.9%.

External IDs:dblp:journals/jiis/ChenZLWLL25