Research On Community Detection Methods For Attributed Networks Based On Non-negative Matrix Factorization

With the rapid development of internet technology,all kinds of online social platforms are coming.Due to hundreds of millions of users,these online social platforms produce more and more rich data in link and content,such as document link network,user interaction network,etc.These networks are usually modeled as attributed networks,with nodes representing objects,edges representing relationships between objects,and attributes representing nodes properties.Especially,the identification of attributed networks communities is of great significance for people to understand and use the semantic function of data.At present,community detection methods are emerging and have achieved good results.PSSNMF based on non-negative matrix factorization and node popularity is a high precision community detection method.It mainly uses the topology information of the networks and takes the prior information and the heterogeneous of node degree into account.However,this method does not consider the possibility of interaction between communities,there may be interaction between communities in the real networks.PSSNMF also does not consider the attribute information of nodes,so it cannot effectively recognize the semantic communities of attributed networks.In response to the above problems,the PSSNMF algorithm is improved in this paper.By considering the four factors of topology information,attribute information,prior information between nodes and heterogeneity of node degree,a new semi-supervised attributed network community detection model PSSNMTFC based on non-negative matrix tri-factorization is proposed.Then,based on the multiplicative updating rule,an efficient optimization algorithm of this model is proposed.Finally,the correctness and convergence of the algorithm are proved,and the time complexity analysis is given.Experiments on artificial and real networks show that the performance of this method is better than some representative algorithms,PSSNMF algorithm included.Finally,an effective and feasible extension scheme is proposed for the problems of this model to make the model more robust and generalization.