Community Detection Algorithm Based On Node Clustering Coefficient And Edge Clustering Coefficient

In the recent years, many scholars began to concern with the study of complex networks which are ubiquitous in the real world. Building grahps and analyzing their topology structure and mathematical properties, scholars found that there were some approximate phenomenon and same characteristics in many complex, such as six degree of separation, the power-law distribution of the node degree, aggregation properties and so on. The nature of community structures is the most important and significant property. Community detection can explore the information hidden in the exterior data relationships, understand the internal structure and function of graph, improve the efficiency of the system, and forecast the development trends of network. So community detection has a high practical value.This paper has a more in-depth study with the community detection algorithm of complex network. Looking for the central node of the community can improve the efficiency of community detection, but the existing core node metrics have deficiencies. For instance, degree centrality has single standard, and it’s accuracy is limited. Closeness centrality and betweenness centrality have large amount of calculation and low efficiency. In the process of community detection, a lot of algorithms, using global information parameters such as edge betweenness, have higher computational complexity and affect the efficiency of community detection. Some algorithms that using local information parameters adopt the single standard, form the communities with smaller scale and have the lack of global optimization. In view of these, this paper adopts the idea of agglomerative method, proposes a more effective and reasonable central node metrics and detect communities based on local information parameters including of node clustering coefficient and edge clustering coefficient. The main work of this paper is as follows:(1) Considering the central node has high degree and links more closely with their adjacent nodes, this paper proposes the concept of clustering centrality as the standard of selecting community centers which are closer to the actual central node. While improves the accuracy of community detection, this method also can reduce the computational complexity, only using the adjacent node information.(2) This paper will gather node with the highest clustering centrality as the initial state of the first community and agglomerate community using node clustering coefficient and edge clustering coefficient which are two local information parameters. When every node in the graph is divided into a certain community, the initial community partition come into being. This kind of algorithm can avoid the problem of high computational complexity which using global information parameters such as edge betweenness. At the same time, comparing to the method only based on node clustering coefficient, this algorithm can avoid forming a large number of communities with smaller size.(3) This paper adjusts the initial community partition based on the difference between indegree and outdegree of communities in order to facilitate the global optimization. So that the final result is closer to actual community structure and this algorithm improves the accuracy of community detection.Finally, this paper implements the algorithm introduced above in C++language, and applies the method in some real network. The experimental results are the same with the actual situation, or they are very close to. These can indicate the reliability and accuracy of the algorithm introduced in this paper.