Information Diffusion Prediction via Dynamic Graph Neural Networks

Information diffusion prediction studies how information items spread in the social network. Previous works either concentrate on leveraging a single diffusion sequence once a time or using the social relationships among users. However, diffusion sequence as a simple chain structure cannot fully capture complex interactions between users. And the diffusion paths of all messages can construct a dynamic diffusion graph. In this paper, we propose a new dynamic structural-temporal graph neural network(DySTGNN) to jointly utilize the structural pattern of the social network and the temporal feature underlying the diffusion graph. What's more, the experimental results have shown that DySTGNN significantly outperforms many recent competitive works on three real-world datasets, which validate the effectiveness of our methods.