Community Detection Algorithm With Node Information Based On DANMF Attributes

In real life,complex networks are ubiquitous.For these networks,the task of community detection is very important,which is conducive to discovering the inter-nal structure of the community.This paper proposes a deep class of self-encoding non-negative with node structure information and feature information Matrix fac-torization algorithm-DANMFSA algorithm.This algorithm mainly uses Gaussian kernel function to calculate similarity matrix with node feature information,and performs non-negative matrix decomposition of similarity matrix.The decomposi-tion process includes not only decoder and encoder,but also deep NMF.Deep NMF can learn the mapping relationship between the similarity matrix and the commu-nity membership matrix hierarchically,and the hidden information of the similarity matrix as well.In order to verify the effectiveness of the DANMFSA algorithm,this paper first divides lawyer network,world trade network,Cora,Citeseer dataset into two types according to whether the node feature matrix is a sparse matrix or not,and then conducts four types of experiments from these two types of data sets.In the first type of experiment,the DANMFSA algorithm is used for community detection on the two data sets.For comparison,another 6 community detection algorithms are also used for community detection on these two data sets.The evaluation criterion used for different algorithms is NMI.The DANMFSA algorithm performs better than other community detection algorithms.The second type of experiment mainly chooses different regularization parameters ?,and explores the impact of ? on the community detection results in the DANMFSA algorithm,so as to select the best?.The third type of experiment proposes an update method for the node feature weights ? for two data sets.The experimental results show that the update of? can improve the performance of the DANMFSA algorithm.The community detection effect of the data set I.The fourth type of experiment mainly uses the t-SNE tool to visualize the data set in the different layer matrices in the DANMFSA algorithm in two dimensions,thereby it can explain the necessity of deep NMF.