A Progressive Framework for Role-Aware Rumor Resolution

Proceedings of the 29th International Conference on Computational Linguistics, pages 2748–2758 October 12–17, 2022. 2748 A Progressive Framework for Role-Aware Rumor Resolution Lei Chen1, Guanying Li2,Zhongyu Wei1,3∗, Yang Yang4, Baohua Zhou2, Qi Zhang5, Xuanjing Huang5 1 School of Data Science, Fudan University, China 2 School of Journalism, Fudan University, China 3 Research Institute of Intelligent and Complex Systems, Fudan University, China 4 College of Computer Science and Technology, Zhejiang University, China 5 School of Computer Science, Fudan University, China 1,3,5{chenl18,zywei,qi_zhang,xjhuang}@fudan.edu.cn 2zhoubaohua@yeah.net;4yangya@zju.edu.cn Abstract Existing works on rumor resolution have shown great potential in recognizing word appear- ance and user participation. However, they ignore the intrinsic propagation mechanisms of rumors and present poor adaptive ability when unprecedented news emerges. To exploit the fine-grained rumor diffusion patterns and generalize rumor resolution methods, we for- mulate a predecessor task to identify trigger- ing posts, and then exploit their characteristics to facilitate rumor verification. We design a tree-structured annotation interface and extend PHEME dataset with labels on the message level. Data analysis shows that triggers play a critical role in verifying rumors and present similar lingual patterns across irrelevant events. We propose a graph-based model considering the direction and interaction of information flow to implement role-aware rumor resolution. Experimental results demonstrate the effective- ness of our proposed model and progressive scheme. 1 Introduction With the expansion of the Internet, online informa- tion tends to spread quickly and widely including fake news, misinformation and rumors, the last of which is defined as circulating stories unverifiable or deliberately false (DiFonzo and Bordia, 2007). Especially in current situation with infectious epi- demics and intensive international relationships, re- searchers have witnessed more than 900% growth in the number of English fact-checks during the COVID-19 outbreak. (Brennen et al., 2020). Auto- matically verifying rumors has become an urgent need for individuals and society. Conventional methods for rumor resolution de- pend on exploiting the evolutionary characteristics of content and spreaders (Kwon et al., 2013; Ma et al., 2015). Benefiting from various attention ∗ Corresponding author Two police officers have been injured in a shooting in #Montrouge in southern #Paris there is no direct link with the #CharlieHebdo attack How the hell can you say there is no direct link, what proof do you have to prove this. An intended shooting spree for policeman. yeah BBC worded it decent enough Amplify Deny Clarify Figure 1: An illustration of rumor cascades and typical roles of messages helpful for rumor verification. mechanisms, there is a growing tendency to re- trieve evidential messages, indicative tokens and critical users to enhance interpretability (Ma et al., 2019; Lu and Li, 2020; Wu et al., 2020a). How- ever, capturing patterns from historical records faces great challenges while transferring to un- precedented events as rumors evolve quickly and recur infrequently. Besides, extracting features of malicious users skips over the immediate indication and allusive roles of the content itself. Recently, researchers are dedicated to investi- gating the intrinsic mechanism of rumor propa- gation rather than linguistic or rhetorical features. Vosoughi …

Content truncated to 500 words