Abstract: Community search in attribute graph is a local community detection method, which is essentially based on the query nodes provided by users to return personalized subgraphs containing query nodes with similar attributes of attributes together with structural cohesion. This task helps users better understand how and why communities are formed. This paper proposes a random walk mechanism based on the structure-attribute interaction bipartite graph, which can effectively support the community search in the attribute graph. Specifically, this method first constructs the structure probability transfer matrix based on the network topology, then explores the bipartite graph formed by the interaction of structure and attribute to define the two-stage node-attribute-node probability transfer matrix, and effectively fuses it with the structure probability transfer matrix to obtain the probability transfer matrix of attribute graph. Finally, a random walk with restart method is designed, and the parallel conductance value based on fusion structure and attribute is used to accurately query the community. Experiments on real datasets and artificial data sets verify the effectiveness and efficiency of the proposed method. 

Key words: community search, random walk, attribute graph, conductance value, transition matrix