Early detection of rumors based on propagation prediction in social media

Social media, particularly microblogging platforms, have been more and more essential for rapid information sharing and public discussion. However, they are also ideal platforms for spreading rumors, often leading to serious consequences (e.g., social panic and chaos). Since the rumors spread extremely rapidly on social media, automatically debunking the rumors in the early stage is of great importance to keep social media a healthy environment. Existing studies for early rumor detection mainly learn clues from the contents of rumor candidates and user profiles, however, difficult to use the rumor propagation patterns adequately, which has been proven crucial for rumor detection, as the propagation structure containing the abnormal pattern has not yet been formed in the early stage. To fill this gap, we propose a Propagation Prediction based Early Rumor Detection (PreERD) model, which first predicts the propagation structure of a post and then combines it with the content and user profile features to identify rumors. Specifically, inspired by the idea of link prediction in complex networks, we construct the user propagation network, and apply a GAN-based model to predict the propagation link of posts, and then a GNN-based detection model with a hierarchical attention mechanism is designed to identify rumors, which aggregates the information of neighbours in predicted propagation network of the target post. Extensive experiments on 3 real-world datasets show that our PreERD model can achieve effective early rumour detection and significantly outperform multiple competitive rumour detection models.