DBPR: Dynamic Bidirectional Propagation Relationship Graph Convolution Network for Fake News Detection on Social Media

With the emergence of social networks and digital media, there has been an increasing proliferation of channels through which people receive information, potentially resulting in the widespread dissemination of fake news. Previous approaches to detecting fake news have primarily focused on mining textual features, while neglecting the dynamic spatiotemporal characteristics of news propagation. In this paper, we propose a novel Dynamic Bidirectional Propagation Relationship (DBPR) model that integrates sequential propagation graphs and backtracking trees in the temporal dimension, incorporating positional encoding based on the temporal order of news propagation to more effectively capture its dynamic nature. Extensive experiments conducted on two publicly available datasets demonstrate that our proposed approach achieves state-of-the-art performance, particularly by providing valuable insights for early detection of fake news.