Social Bot-Aware Graph Neural Network for Early Rumor Detection

Proceedings of the 29th International Conference on Computational Linguistics, pages 6680–6690 October 12–17, 2022. 6680 Social Bot-Aware Graph Neural Network for Early Rumor Detection Zhen Huang1,∗ � , Zhilong Lv1,∗, Xiaoyun Han1,∗, Binyang Li2 �, Menglong Lu1, Dongsheng Li1 1National University of Defense Technology, China 2University of International Relations, China 1{huangzhen, lvzhilong, hxy_1021, lumenglong, dsli}@nudt.edu.cn 2byli@uir.edu.cn Abstract Early rumor detection is a key challenging task to prevent rumors from spreading widely. Soci- ological research shows that social bots’ behav- ior in the early stage has become the main rea- son for rumors’ wide spread. However, current models do not explicitly distinguish genuine users from social bots, and their failure in iden- tifying rumors timely. Therefore, this paper aims at early rumor detection by accounting for social bots’ behavior, and presents a Social Bot- Aware Graph Neural Network, named SBAG. SBAG firstly pre-trains a multi-layer percep- tion network to capture social bot features, and then constructs multiple graph neural networks by embedding the features to model the early propagation of posts, which is further used to detect rumors. Extensive experiments on three benchmark datasets show that SBAG achieves significant improvements against the baselines and also identifies rumors within 3 hours while maintaining more than 90% accuracy. 1 Introduction Rumor is defined as unverified information at the time of posting (Qazvinian et al., 2011; Zubiaga et al., 2018; Lu et al., 2022). Malicious rumors that are spread massively on social media have become a threat to mislead the public and cause social panic. There is a need to debunk rumors in the early stage so as to prevent rumors from the wide spread. Sociological research (Shao et al., 2018a) shows that there exist social bots during the rumor spread, and these bots are particularly active in the early stage of rumors, which will affect the real users through replies and mentions and accelerate the spread (Shao et al., 2018b; Beskow and Carley, 2018; Feng et al., 2022). Current models (Ma et al., 2016; Chen et al., 2018; Song et al., 2019; Zhou et al., 2019; Xia et al., 2020; Han et al., 2021) mainly focus on post *These authors contributed equally to this work. �Corresponding authors content or propagation sequence. These methods model posts as a chronological sequence, and ex- tract the textual feature through GRU, LSTM, and CNN for rumor detection. Other methods (Liu and Wu, 2018; Yuan et al., 2020) account for early ru- mor detection by modeling user characteristics or credibility with user propagation structures. How- ever, they do not explicitly distinguish genuine users from social bots, so the participation of social bots will lead to the failure of the features captured from both contents and propagation structures. To this end, this paper presents a model named Social Bot-Aware Graph Neural Network (SBAG) for early rumor detection. This model consists of two parts: Social bot Detection (SD) and Bot- Aware Graph Rumor Detection (BAG). The for- mer one is pre-trained based on a large sample of bot users and …

Content truncated to 500 words