Research On Overlapping Community Detection Algorithm In Social Networks

With the developing of the web,social network has been a part of daily life,people share the life or work on the social network.And with the developing lager of the scale of the social network,people gradually become fire on the research of the social network.Community detection is a significant branch of the analysis of the social network,it detects the latent clusters through the careful study.Face up with the various characteristics of human,traditional disjoint community detection algorithms are not supposed to find the right latent cluster.In this paper,we focus on overlapping communities.In the overlapping communities,overlapping nodes are those who belong to several clusters.When one node belongs to several communities,we call it overlapping node.For every cluster,one node has different belonging preference,traditional overlapping community detection can not suggest this preference.In this paper,we bring the fuzzy membership-degree to the proposed algorithm FMA,fuzzy membership-degree is a great evaluation for the preference.FMA is an algorithm which is based on the membership-degree,and it includes two steps: one is the propagation process of fuzzy membership-degree,the other is detection again through the extended modularity.Different from traditional label propagation,the FMA algorithm put forward the Node-Attraction to biased propagate the fuzzy membership degree.After the propagation process of the fuzzy membership degree,in order to improve the modularity of communities,the FMA set the threshold for the membership-degree with the guidance of the modularity.And through the way of combining the communities,FMA improved the modularity.The FMA guarantees the modularity of the algorithm.Comparing to the traditional algorithm,FMA shows the preference of overlapping nodes and ensure the modularity of the communities.We take experiments on both the synthetic network and real social network.Simulation results demonstrate that the algorithm FMA can effectively detect the community.LCDA is an algorithm which is based on link partition.LCDA major relies on the link cluster,and then naturally constructs the overlapping communities.LCDA includes three steps: point-graph transmitting into link-graph,links cluster together,link-community transmit into point-community.In the process of the point-graph transmitting into link-graph,the paper proposed a method to compute the similarity of edges: getting a comprehensive parameter for the topology similarity and edge similarity.After the process of the point-graph transmitting into link-graph,we cluster links through the way of co-clustering.And final we transmit the link-community to point-community.Traditionalalgorithm of overlapping communities detection is easy to drop into the trouble of the local optimization,while LCDA keeps the global optimization.