Link prediction based on symmetric nonnegative matrix factorization combining structure and clustering
Most of the existing link prediction algorithms only consider node clustering or link clustering and ignore the internal correlation between network structure and clustering,which leads to a decrease in prediction accuracy.In view of the above shortcomings,we propose a link prediction framework based on symmetric non-negative matrix fac-torization(SNMF)to fuse structure and cluster information capture network to maintain network local,global and node and link clustering.Firstly,the fusion node and edge clustering coefficient(NEC)captures the degree of node neighborhood association,and then undirected and weighted three local similarity methods Common Neighbor(CN),Resource Allocation(RA)and Adamic-Adar(AA))and clustering while maintaining structure and clustering;sec-ondly,the adjacency matrix is mapped to a low-dimensional latent space,and the above information is fused using graph regularization to propose three link prediction models:SNMF-NEC-CN,SNMF-NEC-AA and SNMF-NEC-RA;Furthermore,the proposed model parameters are learnt by iteratively updating the rules to obtain the optimal predic-tion probability matrix.Comparing with the existing representative methods on six networks,the experimental results show that the AUC and F1 values of the proposed model are improved by 22%and 11.4%,respectively.
link predictionsymmetric nonnegative matrix factorizationlocal structurenode and edge clustering