A Unified Propagation Forest-based Framework for Fake News Detection

Proceedings of the 29th International Conference on Computational Linguistics, pages 2769–2779 October 12–17, 2022. 2769 A Unified Propagation Forest-based Framework for Fake News Detection Lingwei Wei1,2 and Dou Hu1,2 and Yantong Lai1,2 and Wei Zhou1∗ and Songlin Hu1,2 1 Institute of Information Engineering, Chinese Academy of Sciences 2 School of Cyber Security, University of Chinese Academy of Sciences {weilingwei, hudou, laiyantong, zhouwei, husonglin}@iie.ac.cn Abstract Fake news’s quick propagation on social media brings severe social ramifications and economic damage. Previous fake news detection usually learn semantic and structural patterns within a single target propagation tree. However, they are usually limited in narrow signals since they do not consider latent information cross other propagation trees. Motivated by a common phenomenon that most fake news is published around a specific hot event/topic, this paper develops a new concept of propagation for- est to naturally combine propagation trees in a semantic-aware clustering. We propose a novel Unified Propagation Forest-based framework (UniPF) to fully explore latent correlations be- tween propagation trees to improve fake news detection. Besides, we design a root-induced training strategy, which encourages representa- tions of propagation trees to be closer to their prototypical root nodes. Extensive experiments on four benchmarks consistently suggest the effectiveness and scalability of UniPF. 1 Introduction Recently, social media platforms have facilitated information dissemination greatly. Nevertheless, they quicken the proliferation of fake news as well due to the lack of authoritative regulators (Zhou et al., 2021). Its extensive dissemination would trig- ger great panic in society and severely impair the public and individuals (Difonzo and Bordia, 2007; Jin et al., 2017; Jankowski et al., 2020). To keep social media a healthy environment, it is desirable and socially beneficial to detect fake news. Among previous works on automatic fake news detection, textual news material is utilized by al- most all extant studies on fake news detection (Ma et al., 2016; Ruchansky et al., 2017; Popat, 2017; Potthast et al., 2018; Zhou et al., 2020). However, because fake news is purposefully designed to de- ceive readers by imitating actual news, detecting * Corresponding author. and distinguishing them solely from news material is challenging (Afroz et al., 2012; Shu et al., 2020a). Hence, an increasing number of works (Ma et al., 2016; Bian et al., 2020; Hu et al., 2021; Song et al., 2021; Wei et al., 2021) have been devoted to learn- ing potential propagation patterns by investigating relationships among tweets for each news article. However, most approaches usually consider that propagation trees are independent and ignore la- tent correlations across different propagation trees, which are supportive for identifying fake news for two-fold reason. 1) Most fake news is usually pub- lished deliberately around a specific hot event / topic and then widely disseminated in reality (Frig- geri et al., 2014; Nourbakhsh et al., 2015). Hence, potential semantical connections may exist across them. 2) Some spreaders were social bots that are manipulated by a malicious group (Shu et al., 2020a). These deliberate and organized behaviors during propagation may lead to …

Content truncated to 500 words