Incorporating Neural Point Process-Based Temporal Feature for Rumor Detection

Social network platforms have facilitated information exchange, but also made the spread of rumors more convenient and rapid. Rumors on the internet leave behind multiple pieces of information with each repost, with temporal information playing a crucial role. Existing studies have focused on extracting various features to discern the veracity of rumors, but the direct analysis of repost timing has been overlooked. In this paper, we present a comprehensive rumor detection method by incorporating temporal features derived from a neural point process model. Our approach investigates the divergences in temporal patterns of reposts between true and false rumors. Moreover, our proposed features can be easily integrated with existing rumor detection methods based on alternative features. We conduct experiments on two publicly available datasets to validate the effectiveness of temporal features. The results demonstrate our proposed model outperforms competing methods.