Modeling and Detecting Network Communities with the Fusion of Node Attributes

As a fundamental structure in real-world networks, communities can be reflected by abundant node attributes with the fusion of graph topology. In attribute-aware community detection, probabilistic generative models (PGMs) have become the mainstream fusion method due to their principled characterization and interpretation. Here, we propose a novel PGM without imposing any distributional assumptions on attributes, which is superior to existing PGMs that require attributes to be categorical or Gaussian distributed. Based on the famous block model of graph structure, our model fuses the attribute by describing its effect on node popularity using an additional term. To characterize the effect quantitatively, we analyze the detectability of communities for the proposed model and then establish the requirements of the attribute-popularity term, which leads to a new scheme for the model selection problem in attribute-aware community detection. With the model determined, an efficient algorithm is developed to estimate the parameters and to infer the communities. The proposed method is validated from two aspects. First, the effectiveness of our algorithm is theoretically guaranteed by the detectability condition, whose correctness is verified by numerical experiments on artificial graphs. Second, extensive experiments show that our method outperforms the competing approaches on a variety of real-world networks.