Community detection in attributed networks based on heterogeneous vertex interactions

Community detection is attracting more attention on social network analysis. It is to cluster densely connected nodes into communities. In attributed networks where nodes have attributes, community detection should take both topology and attributes into account. Traditional community detection algorithms only focus on the topological structure. They do not take advantage of attributes so their performance is limited. Besides, most community detection algorithms for attributed networks are far from satisfactory because of accuracy and algorithm complexity. Moreover, most of the algorithms depend on users to specify the community number, which also impacts the performance. Based on a high-performance community detection algorithm named Attractor, we propose Hetero-Attractor which can detect communities in attributed networks. It expands the sociological model of Attractor and generates a heterogeneous network from the attributed network. Hetero-Attractor analyzes the new network based on the interactions between vertices. By these interactions, the topological information and attribute information not only play a role in the community detection but also interact with each other to reach a balanced result. It also develops a novel way to analyze the heterogeneous network. The experiments demonstrate that our algorithm performs better by utilizing the attribute information, and outperforms other methods both in terms of accuracy as well as scalability, with a maximum promotion of 60% in accuracy.