Rich Information Driven Popularity Prediction on Weibo

Xinyang Shen, Zhen Wu, Jingya Zhou, Yujian Zhu, Weiqing Xu

Abstract: Popularity prediction is to predict the number of social network users involved in information diffusion. Recently, deep learning methods have become mainstream explorations for popularity prediction. However, most existing research is limited to making popularity predictions from temporal and spatial (user interaction) data. In addition to temporal and spatial data, there are also widespread types of data such as text (such as information content posted by users and user profiles), geographical location (such as user residence), discrete tags (such as user authentication) in social networks, and these types of rich data have not been fully used. To this end, we propose a novel framework, called Rich Information heterogeneous graph-based popularity Predictor (RIP). RIP refines the granularity of information diffusion across users on Weibo and constructs a rich information Weibo forwarding heterogeneous graph containing multiple types of data. Our experimental results on real-world Weibo datasets including Wb-SSC and Wb-MSC demonstrate significant gains for RIP over state-of-the-art information diffusion prediction models.