Research Of Community Detecting Based On The Link Similarity

In many fields，many real networks can be abstract into complex network，such asexperts cooperation network，e-mail network，WWW, information network, railwaynetwork, electronic network and so on. With the rapid development of the complexnetwork research in various fields, the structure and features of the network have beenwidely known and application. Community structure is also one of the most importantstructure characteristics. It will be better to understand the role and behavior of body.Detecting the existence of the community structure is also very important for globalnetwork. With the development of network community structure research, a variety ofclassic algorithms appear. All kinds of algorithms have their own characteristics andpractical, but they also show their limitations at the same time. Many need to knowthe global information of network, or others are only applicable to the small networks.Therefore, in the face of the specific network problems, the researchers tend to lookfor the accurate algorithms with low time complexity. Since the21st century, alongwith the expansion of the information，researchers pay attention to the emergence ofoverlapping communities. The limitations of the past community partition algorithmsare further expanded. The focus of the researchers begins to gradually shift from thenodes to the edges. The detecting algorithms of complex network community alsoneed to be developed.This article collects and analysis the concept and characteristics of complexnetwork, and compares the advantages and disadvantages of classic communitypartition algorithms. First of all, it starts from the node similarity of the network, andproposes an improved community partition algorithm. And then the paper verifies theeffectiveness and accuracy of the algorithm through the Karate Club network andDolphins network. By the thought of node similarity, edge similarity is proposed; itcan solve the problem of multiple nodes in the overlapping communities. The papercombine the edge similarity and extension modules, then the end point can be sure forthe detecting algorithm. At last, the rationality of the algorithm is confirmed throughexamples.