Early Rumor Detection with Prior Information on Social Media

With the rapid development of social media on the Internet, many would-be rogues use social media to spread rumors, and rumor detection is born out of this. Rumors on social media change in real time. The earlier we can discover the truth of the event, the more effective it is to curb rumors spreading. This paper studies automatic event-level rumor detection in social media, which is a series of posts that appear in chronological order after an event published. The difficulty of early rumor detection is the available information is limited. Therefore, we take the prior events as auxiliary information and use the fusion of prior events and current event to judge rumors. The model can learn representations of events in the early stage more accurately and realize early rumor detection. Our method can effectively achieve good performance with lack of information in the early stage of social media. Experiments on three benchmark datasets show the proposed method has better advantages.