DUCK: Rumour Detection on Social Media by Modelling User and Comment Propagation Networks

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4939 - 4949 July 10-15, 2022 ©2022 Association for Computational Linguistics DUCK: Rumour Detection on Social Media by Modelling User and Comment Propagation Networks Lin Tian RMIT University, Australia s3795533@student.rmit.edu.au Xiuzhen Zhang RMIT University, Australia xiuzhen.zhang@rmit.edu.au Jey Han Lau The University of Melbourne, Australia jeyhan.lau@gmail.com Abstract Social media rumours, a form of misinforma- tion, can mislead the public and cause signif- icant economic and social disruption. Moti- vated by the observation that the user network — which captures who engages with a story — and the comment network — which captures how they react to it — provide complementary signals for rumour detection. In this paper, we propose DUCK (rumour detection with user and comment networks) for rumour detection on social media. We study how to leverage transformers and graph attention networks to jointly model the contents and the structure of social media conversations, as well as the network of users who engage in these conver- sations. Over four widely used benchmark rumour datasets in English and Chinese, we show that DUCK produces superior perfor- mance for detecting rumours, creating a new state-of-the-art. Source code for DUCK is available at: https://github.com/l tian678/DUCK-code. 1 Introduction Social media platforms bring easy access to a wealth of information. On the flip side, social media has also accelerated the spread of misin- formation (Starbird et al., 2014; Jin et al., 2017). Rumours, a form of misinformation typically de- fined as stories or statements with unverified truth value (Allport and Postman, 1947), can mislead the public and cause significant economic and social disruption. Since the seminal work on prediction of infor- mation credibility on social media by Castillo et al. (2011), automatic rumour detection on social me- dia has attracted significant research, which aims to detect rumour stories (in contrast to news arti- cles by credible news sources) or determine their veracity — true, false or unverified. Although the task is related to fake news detection, the use of social media for propagation means that various social context features can be leveraged for detec- tion. This is a contrast to FEVER-style fake news detection (Thorne et al., 2018) which relies mainly on a source of world knowledge (e.g. Wikipedia) to fact-check stories. Early studies of rumour detection focus on su- pervised learning algorithms incorporating features manually engineered from post contents, user pro- files as well as information propagation patterns (Castillo et al., 2011; Liu et al., 2015; Kwon et al., 2013; Ma et al., 2015; Rath et al., 2017). Re- cent neural approaches typically explore fusing different feature representations for rumour detec- tion. Sequence processing models such as BERT are used to encode the contents of social media conversations, e.g. source posts and the stream of comments (Kochkina et al., 2017; Tian et al., 2020), while graph models have been experimented to model the structure of social media conversa- tions (Bian et al., 2020; Ma …

Content truncated to 500 words