Overlapping Community Detection Based On The Structure Information Of Social Networks

Nowadays,social data is growing.Efficient detection of the hidden community structure in social networks has important research significance and practical value.In social networks,community structure is the most basic and important topological feature.There are a variety of community mining related research today.But in many real networks,the real sense of non overlapping communities is not so common.Therefore,mining the overlapping structure in the network is of realistic meaning.At present,The researchers have proposed a lot of overlapping community detection methods for complex network communities and applied it to the analysis of real scene.However,there are many problems about community division methods worth to be noticed.For example,the accuracy and efficiency of the community partition algorithm need to be improved,especially for overlapping communities.This involves the concept of structure information,community boundaries and node ownership in the network simply considering the topology of the network,but it does not take into account the more information behind the network.In this thesis,we analyze the existing overlapping community partition algorithm.The network structure information,the definition of community boundary and the attribution of nodes are also innovated.The main tasks are as follows:1.In view of the problems that the concept of similarity in existing clustering is not applicable in social networks.we extend the concept of similarity by reading literature summary and combining the structure information of the network.A new concept of link strength is proposed to generalize the network structure information.It combine with the feature that the number of common friends in our friends often embodies the relationship of two people.By using this concept,the distance metric of nodes in the network is put forward.It is applied to the clustering method based on fast search and peak density finding.A community detection algorithm based on density distance is proposed.Finally,the comparison experiment shows that the algorithm we proposed has better quality of community division.2.The original method of determining community boundaries is based on the concept of internal tight and external sparsity of community.And it is determined by thenumber of edges.For complex social networks,such a way involves too little information.And it will lead to a decline in the accuracy of the partition.In addition to community boundaries,the assignment of nodes will also affect the final division accuracy.To solve this problem,this thesis proposes a new definition of community boundary and node attribution based on the strength of networks link.We design an overlapping community detection algorithm based on link strength and it is based on local expansion optimization strategy.The algorithm starts from the initial node and gradually optimizes until the community boundary is no longer enlarged.Finally,the experimental comparison shows that the algorithm achieves good results in community detection.