The Research Of Overlapping Community Detection Algorithms In Complex Networks

In complex networks, it contains a lot of community structure. According to the characteristics of community structure, the mining process of community structure is called community detection. For now a large number of community detection algorithm is putting forward, and these algorithms can find the non-overlapping community structure effectively.However, there are still some unsolved problems: for example in complex network, some nodes may have the characteristics of multiple communities at the same time, that is to say,they belong to several communities. This research of overlapping communities is closer to reality and has a strong practical research value and exploration significance.To solve the problem of existing algorithms, we researched overlapping community detection algorithm. The main contents and innovations are as follows:(1) We proposed a label propagation algorithm based on hierarchy connection degrees. In order to solve the problem that label propagation algorithm(LPA) has poor stability, we improved the LPA algorithm. The initial starting point with independent labels will be rearranged according to the size of the hierarchical connection degree at the initial time. Then,update the node label value in turn. Finally, the nodes with same label will form a community.Experiments confirmed that in terms of rationality and accuracy HCD algorithm is performed better than LPA algorithm. Although it sacrificed an average running time 0.0337 s, but the average accuracy improved 3.87%.(2) We proposed a overlapping community detection algorithm based on fuzzy logic. It has imported the fuzzy logic theory. First, transform the compact degree between nodes into corresponding membership degree. Then, most of closely connected nodes will be divided into the corresponding non-overlapping communities, and the remaining discrete nodes will be redistributed. Finally, the nodes that satisfies specific condition will be identified as overlapping nodes, while the rest of the nodes will be divided into the corresponding communities. The experiments confirm that compared with the classical LFM algorithm the new algorithm caneffectively find the community structure and the overlapping parts among communities and can observe the community division in different hierarchy. It shows that experimental division is accurate, the result is stable, showing a good performance.At the end of this paper, we summarized the full text. Furthermore, we put forward the algorithm improvement ideas found in the process of research and further research and exploration direction.