Community Detection On Complex Networks

With the coming of the big data era, the theory of complex network has been developed rapidly. Community structure, as an important property of complex networks, is of great significance to analyze the structure and function of complex networks. Furthermore, the majority of the real network is evolving over time, which makes the dynamic complex network community detection algorithm has a more important practical significance.In this paper, the community structure and dynamic features of complex network are introduced. And by the comparison of several popular community detection methods, the strong and weak points of current algorithms are summaries.Based on the strong demand on dynamic community detection, we propose a new community detection approach called RCD, which can automatically learn parameters from the network without requiring any redundant input. RCD is divided into two steps as finding initial community structure and tracking dynamic community structure. In the first step, RCD initializes the similarity values of the network and captures the initial community structure by comparing the similarity value with threshold value. In the second step, based on the previous community partition result, RCD redistricts communities according to the input dynamic information at each time-period. By adopting the offset of sigmoid function, RCD reduces dependency on the input cluster number.Therefore, RCD is insensitive to the man-made interference and the robustness is guaranteed. In addition, RCD is not restricted to the type of input networks, because it only depends on the topological structure of network rather than requiring labels or other information of networks.Thus, the application robustness is ensured.RCD are evaluated on both the synthetic and realistic network data.The experiment result shows that by introducing Sigmoid function, the error rate of misclassification and iterative times are decreased.