TaHiD: Tackling Data Hiding in Fake News Detection with News Propagation Networks

Anonymous

Abstract: Fake news with detrimental societal effects has attracted extensive attention and research. Despite early success, state-of-the-art methods fall short of addressing the data hiding challenge, which can be divided into two aspects: disguise and disappearance. Disguise means that fake news publishers may disguise fake news as real ones by imitating the content. Thus, it is not enough to identify fake news only using news content, the different mediums that news propagates through should be taken into account. Disappearance means the related medium information may lose during the propagation of news due to relevant regulations or fake news publishers. It requires the model to capture the propagation features of the news and identify what matters in the propagating process. In this paper, we propose a novel graph-based and heterogeneous-aware fake news detection framework named TaHiD. TaHiD addresses the disguise challenge by encoding multiple mediums during propagation to obtain heterogeneous information. TaHiD aggregates mediums information and measures the influence of different propagation through a propagation transformer module, to handle the disappearance challenge. Experiments demonstrate that TaHiD outperforms state-of-the-art methods on benchmark datasets. Additional studies also show that TaHiD is capable of identifying fake news in the case of data hiding.

Paper Type: long

Research Area: NLP Applications