Semi-Supervised Community Discovery Based On Non-Negative Matrix Factorization

With the rapid development of modern communication technology,all kinds of complex and diverse network big data appear in front of our eyes.These complex networks are one of the hot spots in network analysis and research.And community discovery is an important issue in complex networks.It is important theoretical significance and application value to study it.Since the relationship found in measuring the community is non-negative,and non-negative matrix factorization(NMF)has good physical interpretation and versatility,this paper uses non-negative matrix factorization as a tool to solve community discovery problems.After reading the articles of many scholars,it is found that the community discovery through the non-negative matrix factorization method of topology information often has a relatively low accuracy rate.Considering that the pairwise constraints generated by the actual background information can help the community to detect,this article aims to use the background information(that is,a priori information)to achieve performance improvement,and analyze and explore according to existing methods.On the basis of other outstanding scholars,a novel pairwise constrained non-negative matrix factorization method(PCNMF)was proposed,which can effectively improve the results of community testing.By comparing the existing methods,it is found that PCNMF achieves better results in the undirected network structure;and the existing methods are rarely applied to the directed network situation,so the existing methods are extended to the directed network and found only PCNMF can maintain stability and show good results.