Research On Community Detection Algorithm Based On Structure Feature Of Complex Networks

With the rapid development of information technology, Internet has been gradually integrated into people's daily life. Therefore, a large number of complex networks are emerged.Community structure is one of the most important features of complex networks, and the research on community structure of complex networks can help researchers better understand their features and functions.This paper introduces the basic concepts of complex networks in details and describes the network model. Besides, it describes several common algorithms for community detection. On this basis, the author studies the topological structures of different types of complex networks,and explores the hidden information. Finally, the author proposes two community detection algorithms.(1) A community detection algorithm based on the node similarity, named AR-Cluster, is proposed. AR-Cluster can detect community structure of a complex network with node attributes. Through the analysis and research of IGC-CSM algorithm, we find that the IGC-CSM algorithm only considers the three connected relations in the topology. IGC-CSM does not analyze the close relationship between the nodes in the network topology, when it calculates the structural similarity between nodes. In order to detect communities with densely topological structures and similar attributes, I propose two new concepts: Attracting degree and Recommending degree. AR-Cluster redefines the structural similarity between nodes, and then uses collaborative similarity measure based on structure and attribute features to calculate the similarity between nodes. Finally, nodes in the attribute network are partitioned under K-Medoids framework. The experimental results show that AR-Cluster can effectively detect the community structure in the network with attribute features.(2) A community detection algorithm based on the node importance, named NI-DF, is proposed. NI-DF algorithm consists of two steps: hard partition and overlapping point detection.In the first step, the algorithm applies a method, named NI, which gets the hard partition of network. In order to get these partitions, I propose the node importance based on the network structural feature, and then use the central node expansion method of TROC algorithm. In the second step, the algorithm applies a method, named DF, which gets the overlapping points. In order to detect the overlapping points, I propose the difference function. The experimental results show that NI-DF algorithm can better detect communities and overlapping points.At present, AR-Cluster algorithm only studies the community detection with the same number of attributes, and it does not consider the asymmetry of attributes. NI-DF algorithm detects overlapping points too rough, and it needs to refine the method.