Research On Overlapping Community Detection For Attributed Networks

Many systems in the real-world can be abstracted into complex networks.The community structure is an important feature of complex networks.The community structure which is widespread in various complex networks represents a collection of individuals with common characteristics.For example,users with common interests in social networks often form a same community.The accurate and efficient mining of the communities structure from networks can help us better understand the topology structure of the network and comprehensively understand the internal rules of complex systems.However,in the process of community detection,there are problems that the community center cannot be accurately selected and nodes in real-world always contain a large amount of attribute information.Besides,there is an overlapping characteristic between communities.The above problems led to poor quality in community detection.Therefore,how to design an efficient automatic community center search algorithm is very important.How to fuse the network topology structure and node attributes to generate community partition decisions is a critical issue that the community detection needs to solve.Aim at this problems,this article focus on the following research :（1）Real-world network nodes contain a large amount of attribute information.For this problem,a description and measurement method of the essential characteristics of the network community was proposed.The network topology structure and node attributes were fused to define the intensity degree and interval degree of network nodes,which were designed to describe the characteristics of community—tight internal connections and loose external connections respectively.In the process of community detection,because the combined effect of topology structure and node attributes in the formation of community was considered,the essential characteristics of the community can be more accurately expressed.（2）Community centers cannot be accurately selected during community detection.To solve this problem,a fast search algorithm which can automatically determine community centers was proposed.Firstly,community centers have the characteristics of “higher local density and farther away from higher density community centers”.Based on the characteristics of community centers,the potential community centers were searched out in order to any possible community center is not missed in the subsequent selection.Finally,according to the distance comparison method,the community centers were selected from the potential community centers.（3）Because of an overlapping characteristic between communities,an iterative calculating method for the membership of non-central nodes about each community was proposed to realize the overlapping communities division.While dividing communities,first assign the non-central nodes to the community with the highest degree of membership,and then identify the overlapping nodes in the network.（4）Compared and analyzed with several classic community detection algorithms on real networks,the results show that the proposed algorithm performs well in the aspects of EQ,precision rate,recall rate and F₁-measure than the algorithms of LINK,COPRA,and DPSCD（Density Peaks-based Clustering Method）.The proposed algorithm improves the accuracy of community partition and guarantees the stability of the algorithm.