基于三阶路径自适应度惩罚的链路预测方法

Link Prediction Method Based on Three-Hop Path Adaptive Degree Penalization

链路预测目标是根据已知网络结构来预测缺失链接。现存大部分基于相似度方法仅考虑二阶路径而忽略三阶路径与网络拓扑特征关联程度,这将导致预测准确度下降而不适用于稀疏网络。针对以上不足,提出基于三阶路径自适应度惩罚(THADP)的链路预测算法改善稀疏网络预测精度。首先,统一泛化基于三阶路径相似度方法包括CN-L3、AA-L3和RA-L3,构建THADP框架保持节点所有三阶路径信息;其次,该框架与节点平均最短路径融合有利于捕获整个网络节点信息增强THADP鲁棒性;最后,在8个真实网络上,采用AUC和F1评价所提指标和基准方法性能,实验结果表明所提指标AUC和F1值最高分别提升了35.5%和21.5%。

The goal of link prediction is to predict missing links according to the known network structure.Most of the existing methods based similarity only focus on the 2-hop path while ignore the association between the three-hop path and the topology,which leads to the reduction of prediction accuracy and is not suitable for sparse networks.In view of this shortcomings,a link prediction algorithm based on three-hop path adaptive degree penalty(THADP)has been proposed to improve the prediction accuracy of sparse networks.Firstly,the methods based on three-hop path similarity including CN-L3,AA-L3 and RA-L3 are generally unified,which are used to construct THADP framework to maintain all three-hop path information of nodes.Secondly,the framework and the average shortest path of nodes are fused,which is conducive to capture the node information of the whole network and enhance the robustness of THADP.Finally,the AUC and F1 are used to evaluate the performances of the proposed metrics and the baseline method on eight real networks,and the experimental results show that the AUC and F1 of the proposed metrics are improved by 35.5%and 21.5%,respectively.