Adaptive Overlapping Community Detection Algorithm Based on Mixing Parameter with the Trust Degree of Edge

The community structure in a network simplifies the analysis of the network topology, reveals the internal rules of the system, and provides strong support for information recommendation and information dissemination control. The overlapping community structure of the network is closer to real-life scenario, but its analysis is more difficult than the non-overlapping community. Therefore, to solve the overlapping community detection, based on the peak clustering, an adaptive overlapping community detection algorithm based on the mixing parameter with the trust degree of edge is proposed. In this study, the neighbor edge set of the network and the trust function between the edge and its neighbors are defined, and the total information of the edge is obtained through information transfer. Based on this concept, the concept of mixing parameters is introduced. Then, based on the k-means algorithm, clustering is performed using the mixed parameter, i.e., the edges in the network are divided into a core edge set and a non-core edge set, and each core edge acts as a clustering center. According to the distance from the non-core edge to the core edge, the non-core edges are divided into the community of the nearest cluster center. According to the relation between edges and nodes in the network, overlapping node discovery is achieved. Ultimately the overlapping communities are detected. The advantage of this algorithm is that each edge finds the structure of the community by independently completing information transfer. Moreover, compared to the traditional peak clustering algorithm, the proposed algorithm does not need to set parameters; therefore, adaptive detection of overlapping communities is achieved. To verify the feasibility of our algorithm, the complexity of the algorithm is analyzed. The two evaluation indices of the community detection, normalized mutual information and modularity, are used to experiment on the artificial dataset and the six real datasets respectively. In comparison to other algorithms, the experimental results show that the proposed algorithm is more feasible and effective. © 2019, Editorial Board of Journal of Tianjin University(Science and Technology). All right reserved.