Journal of Frontiers of Computer Science and Technology ›› 2019, Vol. 13 ›› Issue (9): 1459-1470.DOI: 10.3778/j.issn.1673-9418.1809012

Previous Articles     Next Articles

Link Prediction Using Meta-Path Selection and Matrix Factorization Across Social Networks

WANG Yao, KOU Yue, SHEN Derong, NIE Tiezheng, YU Ge   

  1. School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China
  • Online:2019-09-01 Published:2019-09-06

元路径选择和矩阵分解的跨社交网络链路预测

王 瑶,寇月申德荣聂铁铮于戈   

  1. 东北大学 计算机科学与工程学院,沈阳 110819

Abstract: It is of great significance to predict links across networks based on information of different users on multi-source social networks, which is helpful for user recommendation, behavior analysis and preference recommendation. The traditional link prediction technologies only consider the local structural characteristics. Some networks??sizes are large, nodes are sparse, and there are a large number of isolated points, which may lead to difficulty in modeling and low computation efficiency. Based on this, this paper proposes a method for extracting meta-path and matrix decomposition to find hidden features using link prediction methods across networks. First, a network graph is contructed based on social relationship among users across social network; then, meta-paths are automatically extracted by using the node active degree and the edge active degree, and the meta-path information related to the target type objects is mapped to the low-dimensional space by using matrix factorization. Finally, the ensemble classification method is used to optimize the link prediction model. The experiments show that the link prediction method has higher accuracy.

Key words: multi-social networks, link prediction, meta-path, matrix factorization

摘要: 基于多源社交网络上的用户信息实现跨网络链路预测具有重要的意义,有助于进行用户推荐、行为分析、偏好推荐。传统的链路预测技术仅考虑社交网络上的局部结构特征,有些网络规模庞大、节点稀疏、存在大量孤立点,易导致建模困难、计算效率低等问题。基于此,提出了一种基于元路径选择和矩阵分解的跨社交网络链路预测方法。首先,根据跨社交网络中用户间的社会关系构建一个网络图;然后,利用元路径的节点活跃度和边的活跃度自动提取特征;接下来,利用矩阵分解将目标类型对象相关的元路径信息在低维空间上显示;最后,利用集成分类方法对链接模型进行优化。实验数据表明,提出的链路预测方法具有较高的准确性。

关键词: 多社交网络, 链路预测, 元路径, 矩阵分解