A Joint Weighted Nonnegative Matrix Factorization Model via Fusing Attribute Information for Link Prediction

Link prediction is a widely studied problem and receives considerable attention in data mining and machine learning fields. How to efficiently predict missing or hidden edges in the network is a problem that link prediction needs to solve. Traditional link prediction only focuses on the information of network topology and ignores some non-topological information, which makes the prediction performance of algorithm decline rapidly when encountering extremely sparse network. To compensate for this deficiency, this paper proposes a joint weighted nonnegative matrix factorization model for link prediction via incorporates attribute information. By designing a weighted matrix to process the attribute information of each node, both the structure and attribute information fused into the nonnegative matrix factorization framework can fully play a guiding role in the link prediction task, thus solving the problem of structure sparsity and improving the prediction performance of the algorithm. Extensive experiments on five attribute networks demonstrate that the proposed model has better prediction performance than the dozen benchmark methods and the state-of-the-art link prediction algorithms.