Detecting rumor on microblogging platforms via a hybrid stance attention mechanism

Abstract

Microblogging platforms are important social media in the Internet age. Considering the amount of users on microblogging platforms, the rumor spreading on microblogging platform could have a negative effect on individuals, groups and the whole society. Hence, automatic rumor detection is an important research issue. Stance information contains crucial features for rumor detection, because users discussing rumors tend to express more querying and denying stances. Moreover, different user stances have different importance. Motivated by this inspiration, in this paper, we propose a Rumor Detection Model with a Hybrid Stance Attention Mechanism (RDM-HSAM). The RDM-HSAM consists of four modules. The first module is a stance module, in which the tweet-level stance representation is constructed. The second module is the attention module in which a hybrid attention mechanism is used to construct the event-level stance representation of a microblogging event. The hybrid attention mechanism is consisted of two attention mechanisms, i.e. content attention mechanism and user attention mechanism which are applied at the stance information and user profile respectively. The third module is a rumor module which captures the content features and temporal features of a microblogging event. The fourth module is an integrate module in which event-level stance representations and rumor representations are concatenated together to detect rumors. Experiments on a real-world dataset from Weibo platform demonstrate that our proposed model RDM-HSAM improves the performance of rumor detection in terms of both efficiency and accuracy compared to other methods, and the accuracy of our model achieves 94.9%.