ISSN 1000-1239 CN 11-1777/TP

Journal of Computer Research and Development ›› 2015, Vol. 52 ›› Issue (2): 423-436.doi: 10.7544/issn1000-1239.2015.20140221

Special Issue: 2015大数据管理

Previous Articles     Next Articles

Multiple Sources Fusion for Link Prediction via Low-Rank and Sparse Matrix Decomposition

Liu Ye1,Zhu Weiheng2,Pan Yan3, Yin Jian1   

  1. 1(School of Information Science and Technology, Sun Yat-sen University, Guangzhou 510006); 2(College of Information Science Technology, Jinan University, Guangzhou 510632); 3(School of Software, Sun Yat-sen University, Guangzhou 510006)
  • Online:2015-02-01

Abstract: In recent years, link prediction is a popular research field of link mining in social network and other complex networks. In the problem of link prediction, there usually exist multiple additional sources of information used to improve the performance of predicting the probability of the links in network. Among all the sources, the major source of all the information sources usually plays the most significant role on predicting. It is important to design a robust algorithm to make full use of all the sources and balance the major source and additional sources to get better link prediction result. Meanwhile, the traditional unsupervised algorithms based on topological calculation are mostly useful methods to calculate the scores for solving link prediction problem. In the approach of link prediction methods, the most important step is to construct a precise input seed matrix. Since many real-world network data may be noisy, which decreases the accuracy of most link prediction methods. In this paper, we propose a novel method with the multiple additional sources which take advantage of the leading information seed source matrix and others. And then, the seed source matrix is combined with other sources to construct a better matrix with lower noise and more precise structure than the seed matrix. The new matrix is used as the input matrix to traditional unsupervised topological algorithm. Experiment results show that the new proposed method can get better performance of the link prediction problem in different kinds of multiple sources real-world datasets.

Key words: low-rank constraint, matrix decomposition, multiple sources fusion, link prediction, machine learning

CLC Number: