Enhancing Rumor Detection in Social Media Using Dynamic Propagation Structures
Social media, such as Facebook and Twitter, has become one of the most important channels for information dissemination. However, these social media platforms are often misused to spread rumors, which has brought about severe social problems, and consequently, there are urgent needs for automatic rumor detection techniques. Existing work on rumor detection concentrates more on the utilization of textual features, but diffusion structure itself can provide critical propagating information in identifying rumors. Previous works which have considered structural information, only utilize limited propagation structures. Moreover, few related research has considered the dynamic evolution of diffusion structures. To address these issues, in this paper, we propose a Neural Model using Dynamic Propagation Structures (NM-DPS) for rumor detection in social media. Firstly, we propose a partition approach to model the dynamic evolution of propagation structure and then use temporal attention based neural model to learn a representation for the dynamic structure. Finally, we fuse the structure representation and content features into a unified framework for effective rumor detection. Experimental results on two real-world social media datasets demonstrate the salience of dynamic propagation structure information and the effectiveness of our proposed method in capturing the dynamic structure.