BIC: Twitter Bot Detection with Text-Graph Interaction and Semantic Consistency

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 10326–10340 July 9-14, 2023 ©2023 Association for Computational Linguistics BIC: Twitter Bot Detection with Text-Graph Interaction and Semantic Consistency Zhenyu Lei1∗ Herun Wan1∗Wenqian Zhang1 Shangbin Feng2 Zilong Chen3 Jundong Li4 Qinghua Zheng1 Minnan Luo1† Xi’an Jiaotong University1, University of Washington2 Tsinghua University3, University of Virginia4 {Fischer, wanherun}@stu.xjtu.edu.cn Abstract Twitter bots are automatic programs operated by malicious actors to manipulate public opin- ion and spread misinformation. Research ef- forts have been made to automatically iden- tify bots based on texts and networks on so- cial media. Existing methods only leverage texts or networks alone, and while few works explored the shallow combination of the two modalities, we hypothesize that the interaction and information exchange between texts and graphs could be crucial for holistically evaluat- ing bot activities on social media. In addition, according to a recent survey (Cresci, 2020), Twitter bots are constantly evolving while ad- vanced bots steal genuine users’ tweets and di- lute their malicious content to evade detection. This results in greater inconsistency across the timeline of novel Twitter bots, which warrants more attention. In light of these challenges, we propose BIC, a Twitter Bot detection frame- work with text-graph Interaction and semantic Consistency. In particular, BIC utilizes a text- graph focused approach to facilitate the two communication styles of social media that may trade useful data throughout the training cy- cle. In addition, given the stealing behavior of novel Twitter bots, BIC proposes to model se- mantic consistency in tweets based on attention weights while using it to augment the decision process. Extensive experiments demonstrate that BIC consistently outperforms state-of-the- art baselines on two widely adopted datasets. Further analyses reveal that text-graph interac- tions and modeling semantic consistency are essential improvements and help combat bot evolution. 1 Introduction Twitter bots are controlled by automated pro- grams and manipulated to pursue malicious goals ∗ These authors contributed equally to this work. † Corresponding author: Minnan Luo, School of Com- puter Science and Technology, Xi’an Jiaotong University, Xi’an 710049, China. Figure 1: (a) Different types of strategies that combine modalities. Previous methods adopt text modality or graph modality alone, or just shallowly combine them. There is a need for an interactive method that interacts and exchanges information across the modalities. (b) Genuine users and Twitter bots have different patterns of semantic consistency. Tweets in red are abnormal and these example tweets show semantic inconsistency. such as advocating for extremism and producing spam (Dickerson et al., 2014; Berger and Morgan, 2015). Bots are also involved in spreading misin- formation during the pandemic (Shi et al., 2020). Since Twitter bots pose a threat to online society, many efforts have been devoted to detecting bots. The majority of the existing approaches are text- based and graph-based. The text-based methods analyze the content to detect Twitter bots by natu- ral language processing techniques. Kudugunta and Ferrara (2018) adopted recurrent neural networks to extract textual information. Guo …

Content truncated to 500 words