基于多质心异质图学习的社交网络用户建模

User modeling has attracted great attention in both academia and industry. Most of the existing approaches focus on incorporating the personal relationships in communities, while the users' generated content such as posts is not well studied. In this paper, through the analysis of the actual public opinion dissemination, we show that the research on user attributes plays an important role in the process of public opinion dissemination, and propose the screening method of user data. Meanwhile, we propose an approach to capture more diverse community characteristics via heterogeneous multi-centroid graph pooling for user modeling.Specifically, we first construct a heterogeneous graph where the nodes consist of both users and keywords and adopt a heterogeneous GCN on it. To facilitate the graph representation for user modeling, we then propose a multi-centroid graph pooling mechanism, which incorporates the affiliated group features with multiple centroids into representation learning. Extensive experiments on three benchmark datasets show the effectiveness of our proposed approach.