The Research Of Community Detection Algorithm Based On Fuzzy Clustering

The community structure as an important feature of the complex network.After the complex network is divided into several communities,Through the study of several small communities, we can study the module, function and evolution of the whole system. In this way, we can reduce the complexity of studying the large complex system directly and further deepen our understanding of the organizational principle, topological structure and dynamics of complex systems.Among the existing community detection algorithms of complex network, most of them are non-overlapping community detection algorithm, which means the network is divided into several independent communities there is no overlapping parts between communities. But in some real network, there are some overlapping nodes with the properties of several communities. According to the overlapping nodes, we can infer the correlation degree between communities. For dynamic network, overlapping parts can also predict the next step in the change of the community. Therefore, the discovery of overlapping nodes has special significance and research value.Based on the study of the existing community detection algorithm, this paper puts forward a community detection algorithm based on fuzzy clustering, which can be applied to the community detection of both overlapping community network and non-overlapping community network at the same time.Community detection algorithm based on fuzzy clustering can be seen as a kind of heuristic algorithm, the heuristic rules are as follows: The node is used as a sample in the fuzzy clustering algorithm, after applying the fuzzy clustering algorithm, the nodes which belong to the same equivalence class should belong to the same community. In the traditional fuzzy clustering algorithm, the value of membership function reflects the degree of fuzzy relation ~R between samples or similarity degree of two samples. With the nodes used as the cluster sample, in order to apply the traditional fuzzy clustering algorithm to community detection, the specific work of the algorithm is as follows: 1. The definition of traditional shared neighbor number is improved, and the similarity degree of the nodes is represented by the improved definition and a membership function based on the number of the shared neighbors is proposed. 2. In some cases, there are overlapping nodes existing in the network and a few of them do not belong to any of the equivalence class, namely the candidate solution of the overlapping nodes. After obtaining the results by fuzzy clustering algorithm, a criterion was proposed to detect the overlapping nodes in the candidate solution. The time complexity of algorithm is22( log)nO n. After testing In the GN benchmark and several real networks, the detection results of the GN benchmark whose community structure is already known, conform to the actual situation. The algorithm is verified on the real network,comparing with the results of other algorithms, the modularity is higher and it can detect overlapping nodes in the network better.