Community Detection Of Complex Network Based Self-organizing Neural Network

Complex network theory is a systematic theory created by people when theyunderstand the world, so complex network can be seen as an abstract world, the study ofcomplex network theory has been concerned about by the natural sciences and severalacademic researchers. The community structure is a ubiquitous structure characteristic ofcomplex networks, community discovery research for the development of complex networktheory has great significance, so the community structure of complex network discoveryalgorithm has been studied in various fields and sustained attention of researchers. But sofar the problem has not been the perfect solution, especially overlapping communitystructure discovery, research in this area has brought new challenges.This article starting from the Research background and significance of communitystructure discovery algorithm in complex networks, check out a lot of relevant literature,understanding the research Status in this field. Secondly, it introduces the associatedconcept and the historical process of complex network theory. Again, this article alsodiscussed the concept of community and community structure discovery algorithmsevaluation, and begin from several representatives algorithm, we discusses the currentprogress and the lack of the community discovery research. Finally, this paper alsodescribes the self-organizing neural network theory, starting from competitive learning tothe self-organizing maps model, highlighting the topology retention characteristics andprobability retention characteristics.For the insufficient of current overlapping community structure discovery algorithm,the paper taking into account the real constituent elements of community structure is not anode but the network edge, raised edges close group concept, and with the edges closegroup to construct edge vectors in order to more accurately express the true information ofthe network edge implied. In order to remove the useless boundary edges in network whenrun the community structure discovery algorithm, Based on the concept of a edges closegroup we proposed a new method to identify the boundary edges. On this basis, this paperraised a overlapping community structure discovery algorithm EVKM based on edgevectors idea, combined with the idea of k-means algorithm to discovery the overlappingcommunity structure in network. Then, this paper proposed a new approach to overlapping community detection based on self-organizing map and edge vector(called SOMEV).Finally, we designed several experiments to verify the effectiveness of the algorithmwe propose, first we comparing the effects of edge vectors normalized operating to theresults of algorithm, then we also test how the different values of the edge belonging impactthe community structure discovery accurate rate, and finally, we comparing our proposedalgorithm and some classical community discovery algorithms on artificial networks andreal networks. Through comparative experiments show that the overlapping communitystructure discovery algorithm based on edge vectors thought has obvious advantages thanthe classical algorithm in the task of overlapping community structure recognition. Inaddition, self-organizing neural network with topology retention characteristics andprobability retention characteristics also help to improve the accuracy of overlappingcommunity structure discovery algorithm.