Research On Detecting Community In Complex Network Based On Similarity

Complex network is the abstract representation of a series of system, like biological system, social system, network, World Wide Web and so on. The community structure of complex network indicates the polymerization tendency of nodes in the network. So the community structure is a very important property to complex network, and detecting community in complex network means a lot when used in reality, such as collect pages with same topic, find function unit in metabolic network and so on. In addition, recently a lot research results show that the attribute of the whole network is very different from that of community. Therefore, we would lose a lot of meaningful properties if we ignore the study of community structure. Excellent detecting community algorithms should have high accuracy and low computational complexity. In the past few years, researchers have proposed a lot of algorithm to detect community, but only a few of them could achieve the above two points perfectly. Nowadays the scale of network is more complicated, and people pay more attention to the speed of the algorithm.In view of the limitation of traditional modularity, this paper introduces the concept of similarity, which has good scalability and low computational complexity. Based on the similarity, this paper proposed three algorithms: cluster algorithm, label algorithm and multi label algorithm. The former two is suitable for the non-overlapping community structure. Cluster algorithm proposes a new modularity based on node similarity to avoid the limit of traditional modularity. Label algorithm uses a method based on label similarity to avoid the random selection in the traditional label algorithm to solve the instability problem. The last one is extended from label algorithm by allowing node with multi labels instead of only one, which make the algorithm suitable for detecting overlapping community structure.We apply them in the real network and computer simulation network, and compare them with other community detection algorithms. The result shows that our algorithm has lower computational complexity and higher accuracy.