计算机应用

• 人工智能与仿真 •    下一篇

一种基于拓扑连接紧密度的相似性链路预测算法

丁大钊1,黄开枝2,刘树新3   

  1. 1. 国家数字交换系统工程技术研究中心
    2. 国家数字交换系统工程技术研究中心,郑州 450002
    3. 国家数字交换系统工程中心
  • 收稿日期:2017-02-22 修回日期:2017-05-02 发布日期:2017-05-02 出版日期:2017-05-13
  • 通讯作者: 丁大钊

A link prediction method for complex network based on the closeness between nodes

  • Received:2017-02-22 Revised:2017-05-02 Online:2017-05-02 Published:2017-05-13

摘要: 链路预测可以预测复杂网络中未知、缺失的连接,具有重要实际应用价值。基于网络结构提出了许多相似性链路预测方法,而共同邻居在其中发挥着举足轻重的角色。当前,许多方法仅仅关注预测的准确度衡量指标,忽略了精确度衡量标准在实际应用中的重要作用,且没有考虑共同邻居与预测节点间紧密度对相似性刻画的影响。基于共同邻居节点周围拓扑结构,提出了一种基于拓扑连接紧密度的相似性链路预测算法。多个实际网络测试表明,相比CN, RA, AA, LP, Katz,等相似性方法,该方法具有较高的链路预测精确度。

Abstract: Link prediction can predict the missing links of complex network, which plays an important role in practical applications. Many similarity indices have been proposed based on topology structure for link prediction, and common neighbor play an irreplaceable role in these indices. Plenty of similarity indices pay too much attention on the standard metric AUC. However, they ignore the metric precision and closeness of common neighbors and endpoints under different topology structure. Considering the local topological information around common neighbors, a link prediction method based on closeness between nodes is proposed. Empirical study on ten real networks has shown that the index proposed can achieve higher prediction accuracy, compared with CN, RA, AA, LP, Katz.

中图分类号: