A Non-negative Matrix Factorization(NMF)-based method is proposed to solve the link prediction problem in dynamic graphs. The method learns latent features from the temporal and topological structure of a dynamic netw...A Non-negative Matrix Factorization(NMF)-based method is proposed to solve the link prediction problem in dynamic graphs. The method learns latent features from the temporal and topological structure of a dynamic network and can obtain higher prediction results. We present novel iterative rules to construct matrix factors that carry important network features and prove the convergence and correctness of these algorithms. Finally, we demonstrate how latent NMF features can express network dynamics efficiently rather than by static representation,thereby yielding better performance. The amalgamation of time and structural information makes the method achieve prediction results that are more accurate. Empirical results on real-world networks show that the proposed algorithm can achieve higher accuracy prediction results in dynamic networks in comparison to other algorithms.展开更多
基金supported in part by the National Natural Science Foundation of China(Nos.61379066,61702441,61379064,61472344,61402395,61702441,and 61602202)the Natural Science Foundation of Jiangsu Province(Nos.BK20130452,BK2012672,BK2012128,and BK20140492)+1 种基金the Natural Science Foundation of Education Department of Jiangsu Province(Nos.12KJB520019,13KJB520026,and 09KJB20013)Six-talent-peak Project in Jiangsu Province(No.2011-DZXX-032)
文摘A Non-negative Matrix Factorization(NMF)-based method is proposed to solve the link prediction problem in dynamic graphs. The method learns latent features from the temporal and topological structure of a dynamic network and can obtain higher prediction results. We present novel iterative rules to construct matrix factors that carry important network features and prove the convergence and correctness of these algorithms. Finally, we demonstrate how latent NMF features can express network dynamics efficiently rather than by static representation,thereby yielding better performance. The amalgamation of time and structural information makes the method achieve prediction results that are more accurate. Empirical results on real-world networks show that the proposed algorithm can achieve higher accuracy prediction results in dynamic networks in comparison to other algorithms.