Adversary-Aware Rumor Detection

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 1371–1382 August 1–6, 2021. ©2021 Association for Computational Linguistics 1371 Adversary-Aware Rumor Detection Yun-Zhu Song,1 Yi-Syuan Chen,1 Yi-Ting Chang,1 Shao-Yu Weng,2 Hong-Han Shuai,1 1National Yang Ming Chiao Tung University, Taiwan 2National Tsing Hua University, Taiwan {yunzhusong.eed07g, yschen.eed09g, joshchang0111.eed06}@nctu.edu.tw nthu106071077@gapp.nthu.edu.tw, hhshuai@nctu.edu.tw Abstract While social media becomes a primary source of news now, it also becomes more challeng- ing for people to distinguish the rumors and non-rumors, which attracts malicious manip- ulation and may lead to public health harm or economic loss. Consequently, many ru- mor detection models have been proposed to automatically detect the rumors based on the contents and propagation path. However, most previous works are not aware of mali- cious attacks, e.g., framing. Therefore, we propose a novel rumor detection framework, Adversary-Aware Rumor Detection including Weighted-Edge Transformer-Graph Network and Position-aware Adversarial Response Gen- erator, to improve the vulnerability of detec- tion models. To the best of our knowledge, this is the first work that can generate the ad- versarial response with the consideration of the response position. Experimental results show that our model achieves the state-of-the- art on various rumor detection tasks by the pro- posed Weighted-Edge Transformer-Graph Net- work and can maintain the performance under the adversarial response attack after the adver- sarial learning by Position-aware Adversarial Response Generator.1 1 Introduction With the popularity and accessibility of social me- dia, social media becomes the primary source for obtaining information.2 Compared with traditional news, posts on social media are usually with shorter lengths and faster transmission speed, which also increases the difficulty of message verification. As such, social media are increasingly targeted for ma- nipulation, leading to tremendous economic losses, 1The codes are released as a public download at https: //github.com/yunzhusong/AARD. 2https://pewrsr.ch/3nzYpQd Figure 1: Position test for an adversarial response. The reply position can make an influence on the detection model. and even deaths.3 Take the COVID-19 pandemic as an example, a newly published study shows that at least some 800 people died because of a rumor about drinking highly-concentrated alcohol can dis- infect their bodies (Islam et al., 2020). Therefore, fighting against misinformation in social networks gains a great deal of attention and becomes essen- tial and inevitable. In this paper, we study the problem of rumor detection on social media, where a rumor is de- fined as an unverified and instrumentally relevant information statement in circulation (DiFonzo and Bordia, 2007; Zubiaga et al., 2018). Given a con- versation thread, including a source post and re- lated responses, the rumor detection task aims to determine whether the source post is a rumor or not. Previous works of rumor detection can be cat- egorized into three classes according to the data usage. Content-based approaches only use the tex- tual information of the source posts and the user responses (Ma et al., 2016, 2019), while graph- 3https://s3.amazonaws.com/ media.mediapost.com/uploads/ EconomicCostOfFakeNews.pdf 1372 based methods (Ma et al., 2017, 2018; Bian et al., 2020) consider the message propagation paths or model the propagation …

Content truncated to 500 words