摘要： 在具有模体特征的食物链网络、社交网络中，局部朴素贝叶斯（LNB）的链路预测方法通过准确区分每个共邻节点的贡献以提高链路预测的精确度，但忽略了每个共邻节点对所在路径的贡献不同以及网络模体结构对链接形成的作用。针对LNB链路预测方法存在的局限性问题，结合路径模体特征与朴素贝叶斯理论，提出基于模体的朴素贝叶斯链路预测方法。定义模体密度以量化路径结构上模体的聚集程度。考虑路径结构上模体密度对链接形成的影响，构建每条路径的角色贡献函数，以量化每条路径结构的模体特征对节点相似性的影响。在此基础上，根据朴素贝叶斯理论与角色贡献函数推导节点相似性指标。在Football、USAir、C.elegans、FWMW、FWEW和FWFW 6个真实网络上进行实验，结果表明，该方法能有效提高预测性能且具有较优的鲁棒性，其中在具有显著模体特征的FWMW、FWEW、FWFW网络上，相比现有相似性指标中较优的Katz指标，所提相似性指标的AUC值提升了2%~7%。

关键词: 复杂网络, 链路预测, 朴素贝叶斯, 相似性指标, 模体密度

Abstract: In food chain networks and social networks with motif features, Local Naive Bayes (LNB) link prediction method improves the accuracy of link prediction by accurately distinguishing the contribution of each common neighbor node, but neglects the different contributions of each common neighbor node to the path and the role of the model structure in the network on the link formation.To address the limitations of LNB link prediction methods, this study proposes a motif-based naive Bayes link prediction method by combining path-motif features and naive Bayes theory.The motif density is defined to quantify the degree of aggregation of motifs on the path structure.Considering the influence of the motif density on the path structure on the link formation, the role contribution function of each path is constructed to quantify the impact of the motif features of each path structure on the similarity of nodes.Then, the similarity index of nodes is derived according to the naive Bayes theory and the role contribution function.Experiments on Football, USAir, C.elegans, FWMW, FWEW and FWFW networks show that the proposed method can effectively improve the prediction performance and has better robustness.On the FWMW, FWEW, and FWFW networks with obvious motif features, the AUC value of the proposed similarity index increased by 2%~7% compared with the suboptimal Katz index among the existing similarity indexes.

Key words: complex network, link prediction, naive Bayes, similarity index, motif density

中图分类号: 

• TP393