1673-159X

CN 51-1686/N

CHEN Guangfu, LIU Qi, LI Xiaofei. Link Prediction Based on Weighted Structure and Weighted Nonnegative Matrix Factorization[J]. Journal of Xihua University(Natural Science Edition), 2022, 41(4): 57 − 65. . DOI: 10.12198/j.issn.1673-159X.4441
Citation: CHEN Guangfu, LIU Qi, LI Xiaofei. Link Prediction Based on Weighted Structure and Weighted Nonnegative Matrix Factorization[J]. Journal of Xihua University(Natural Science Edition), 2022, 41(4): 57 − 65. . DOI: 10.12198/j.issn.1673-159X.4441

Link Prediction Based on Weighted Structure and Weighted Nonnegative Matrix Factorization

  • The goal of link prediction is to predict missing links and possible future links based on known network structure information. However, most existing link prediction algorithms only focus on undirected and unweighted networks and ignores the natural weights and network structure which leads to the decrease of prediction accuracy. To address this problem, we propose a link prediction model of weighted non-negative matrix factorization to preserve natural weights and local structure of weighted network at the same time. Firstly, the adjacency matrix decomposition of the weighted network is mapped to a low-dimensional latent space to preserve the natural link weights of the original network, and then three classical weighted common neighbors (WCN), weighted Adamic-Adar (WAA) and weighted resource allocation (WRA) are used as indicators and matrix is assigned to the non-negative matrix factorization model to maintain the local structure of the network, and the above two types of information are fused to propose three link prediction models based on the weighted non-negative matrix factorization (WNMF) framework, namely WNMF-WCN, WNMF-WAA and WNMF-WRA . Furthermore, the lagrange multiplication rule is enabled to learn the proposed three model parameters. Compared with existing link prediction methods on 6 real-world weighted networks, the experimental results show that the proposed model improves the PCC and Precision values by 22.8% and 23.5%, respectively.
  • loading

Catalog

    Turn off MathJax
    Article Contents

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return