Multi-Stage Network Representation Learning For Exploring Heterogeneous Edges

Network representation learning attempts to maximize the retention of the original information of the network with a low-dimensional dense matrix.Such matrix is viewed as feature,and then perform subsequent network analysis tasks.The complex relationships between nodes are abstracted as heterogenous edges,thus heterogeneous edges represent different semantic information.Existing network representation learning methods are designed for homogenous edges,and fail to model heterogenous edges,.How to mine complex information from heterogeneous edge is still a challenge worth exploring.To explore the heterogeneous edges for network representation learning,by considering each relationship as a view that depicts a specific type of proximity between nodes,a Multi-stage Non-negative Matrix Factorization(MNMF)model is proposed,committed to utilizing abundant information in multiple views to learn robust network representations.In fact,most existing network embedding methods are closely related to implicitly factorizing the complex proximity matrix.However,the approximation error is usually quite large,since a single low-rank matrix is insufficient to capture the original information.Through a multi-stage matrix factorization process motivated by gradient boosting,the MNMF model achieves lower approximation error by progressively decomposing the residual matrix of the previous stage for each view.Meanwhile,the multistage structure of MNMF gives the feasibility of designing two kinds of Non-negative Matrix Factorization(NMF)manners to preserve consensus information and unique information better.In detail,the united NMF aims to preserve the consensus information between different views,and the independent NMF aims to preserve unique information of each view.By aggregating the representations of consensus information as well as unique information,the final node representation can be obtained.Experimental results show that MNMF model is better than the existing baseline methods for heterogeneous edge network representation learning in downstream tasks of multiple real-world datasets.For example,it achieves more than 2% gains in terms of f1 scores in the node classification task;It achieves around 1.5% gains in terms of mutual information score in the node clustering task.The network visualization task also shows that the community structure of nodes is more apparent.At the same time,the experiment also proves the rationality and effectiveness of multi-stage matrix factorization.