• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2015, Vol. 37 ›› Issue (01): 173-178.

• 论文 • 上一篇    下一篇

差分化节点特征对复杂网络链接预测的分类性能分析

伍杰华1,2,朱岸青1,3 ,蔡雪莲1,张小兰1   

  1. (1.广东工贸职业技术学院计算机工程系,广东 广州 510510;2.华南理工大学信息科学与技术学院,广东 广州 510641;
    3.暨南大学信息科学与技术学院,广东 广州 510632)
  • 收稿日期:2013-04-15 修回日期:2013-08-07 出版日期:2015-01-25 发布日期:2015-01-25
  • 基金资助:

    国家自然科学基金资助项目(61003045);广东省教育部产学研结合项目(2012B091100043)

Classification performance analysis on link prediction
via diverse node features in complex network   

WU Jiehua1,2,ZHU Anqing1,3 ,CAI Xuelian1,ZHANG Xiaolan1   

  1. (1.Department of Computer Science and Engineering,Guangdong College of Industry and Commerce,Guangzhou 510510;
    2.College of Information Science and Technology,South China University of Technology University,Guangzhou 510641;
    3.College of Information Science and Technology,Jinan University,Guangzhou 510632,China)
  • Received:2013-04-15 Revised:2013-08-07 Online:2015-01-25 Published:2015-01-25

摘要:

链接预测属于复杂网络分析的研究分支,它根据网络历史结构信息预测未来节点间会产生链接的可能性,从而挖掘网络的传播和演化方式。通过引入差分化节点的贡献权重并结合经典的节点和共邻节点网络拓扑结构特征,分别应用七类有监督学习分类模型对社交、生物、交通等不同领域的八个真实复杂网络数据集进行实验,并采用Precision和ROC曲线对实验结果进行分析与评价。实验表明,引入基于差分化节点的贡献特征能够在深入挖掘网络结构信息的基础上比其余特征有更优的预测精确度,同时差异化的分类模型和特征选择对链接预测性能有相异的影响。

关键词: 链接预测, 复杂网络, 特征选择, 分类, 共邻节点

Abstract:

Link prediction is a branch of complex network analysis. It predicts future linkage between nodes according to historical network structure information in order to reveal the network evolution and diffusion. This paper introduces a novel feature based on differentiated node contribution and provides seven supervised learning frameworks to carry the classification task via our introduced features and two categories of classic topological features which include node information and common neighbors’ property. The experimental results on eight realworld networks demonstrate interesting finding on the prediction influence by different classification models and topological diverse features via precision and ROC curve. In addition, we prove that our newly introduced features outperform the classic ones, which can mine the latent network information and enhance accuracy of link prediction.

Key words: link prediction;complex network;feature selection;classification;common neighbors