Link prediction by deep non-negative matrix factorization

作者：

Highlights：

• We propose a novel multi-layer network link prediction framework, namely FSSDNMF.

• FSSDNMF can exploit the observed links and topological information for hidden layer.

• We employ the ℓ2,1-norm to eliminate random noise.

• We provide theoretical and experimental analysis of the convergence of FSSDNMF.

• Experimental demonstrate that the FSSDNMF outperforms the state-of-the-art methods.

摘要

•We propose a novel multi-layer network link prediction framework, namely FSSDNMF.•FSSDNMF can exploit the observed links and topological information for hidden layer.•We employ the ℓ2,1-norm to eliminate random noise.•We provide theoretical and experimental analysis of the convergence of FSSDNMF.•Experimental demonstrate that the FSSDNMF outperforms the state-of-the-art methods.

论文关键词：Link prediction,Deep non-negative matrix factorization,Structural information,Sparsity-constrained

论文评审过程：Received 18 February 2021, Revised 11 June 2021, Accepted 26 September 2021, Available online 14 October 2021, Version of Record 23 October 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.115991