Community-based Dynamic Graph Learning for Popularity Prediction

Shuo Ji

Abstract: Popularity prediction, which aims to forecast how many users will interact with a target item or online content in the future, can help online shopping and social media platforms identify popular items or digital content. Many existing studies investigate how multi-faceted factors such as item features, user preferences, and social influence affect user–item interactions, but relatively little attention has been paid to the evolutionary dynamics of these factors at the individual or group level. In this light, this paper develops a community-based dynamic graph learning method for popularity prediction. Specifically, we propose a dynamic graph learning framework that maintains time-evolving representations for item and user entities and updates them according to newly observed user–item interactions. We further design a community detection module to capture evolving community structures and identify influential nodes. More importantly, our framework incorporates community-level message passing to balance local and global information propagation during learning. Finally, the popularity of target items or online content is predicted based on the learned representations. Experiments on three real-world datasets demonstrate that the proposed method consistently outperforms baseline approaches, while effectively modeling the evolution of user preferences and community structures over time.