Research Of Community Detection Algorithm Depending On Seed Expansion

Many complex systems in modern life can be abstracted as complex networks.With the growth of data volume in complex systems,more and more complex systems will be regarded as complex networks.Community is a typical feature of complex networks,and communities correspond to units in complex systems.Community detection aims at revealing community structure of complex networks.Successful community detection can assist us to reveal insight laws of complex networks,so community detection for complex networks has great practical significance.Researchers have proposed many community detection algorithms,which composed of modularity-based algorithm,LPA-based algorithm,hierarchical algorithm,seed expansion algorithm and density-based algorithm.We conduct a deep research in these community detection algorithms.In this paper,we focus on the improvement of seed expansion algorithms.There remains some problem to resolve: the result is not satisfying because too little seeds are selected;the selected seeds can not cover all communities;the time complexity is high;the overlapping nodes are not handled;the seeds of traditional seed expansion algorithm are not selected from the perspective of whole networks.In this paper,we propose a density influence based seed-centric community detection algorithm(DenISeC).DenISeC calculates an influence on whole network for every node if the node is removed from networks.The differences between sum of densities on networks and sum of densities on networks whose node is removed is taken as the influence of the node.The nodes with high influence are taken as seeds of communities.Seeds make up the initial communities.The remaining nodes are assigned into initial communities according to the similarity between the node and communities.To more effectively find communities,we further propose a density-based seed-centric community detection algorithm(DenSeC).Take the nodes with small degrees as boundary nodes of communities,DenSeC finds seeds by searching from boundary parts of communities to core parts of communities.These seeds make up initial communities,and then remaining nodes are assigned into a community according to the similarities between the nodes and communities.The proposed algorithms are tested on numerical datasets,and several algorithms are taken as comparison algorithms.Experiments have proven that DenSe C improves the accuracy of traditional seed expansion algorithms.The way of selecting seeds as well as time complexity are more reasonable than traditional methods in DenSeC.More reasonable seeds are selected and time complexity decreases in DenISeC.