一种集成局部加权聚类系数的链接预测算法  

Algorithm of integrating local weighted clustering coefficients for link prediction

作　　者：伍杰华[1,2] 张小兰[1] 沈静[1] 周蓓[1] Wu Jiehua;Zhang Xiaolan;Shen Jing;Zhou Bei(Dept.of Computer Science ＆ Engineering,Guangdong Polytechnic of Industry ＆ Commerce,Guangzhou 510510,China;School of Computer Science ＆ Engineering,South China University of Technology,Guangzhou 510641,China)

机构地区：[1]广东工贸职业技术学院计算机工程系,广州510510 [2]华南理工大学计算机科学与工程学院,广州510641

出　　处：《计算机应用研究》2018年第12期3588-3592,3613,共6页Application Research of Computers

基　　金：广东省科技计划资助项目(2017ZC0348);广东高校重大科研项目与成果培育计划资助项目(2017GKTSCX009);广东省优秀青年教师培养计划项目(YQ2015177)

摘　　要：针对基于局部结构的加权链接预测算法仅仅利用了一级共邻节点的拓扑属性,无法反映共邻节点的邻居对潜在节点对的贡献以及度量共邻节点互连密集程度对预测结果的影响这一问题,从局部结构的密集层面来分析共邻节点对潜在节点对的影响,提出了一种集成加权聚类系数的相似度指标(WCCLP)。该指标能够有效地扩大局部共邻节点结构对预测性能的影响,同时也能轻易地拓展到加权局部朴素贝叶斯链接预测模型(WLNB)中。采用无监督学习的实验表明,WCCLP在多个真实数据集比现有的基准指标取得了更好的预测效果,拓展到WLNB的实验效果证明加权聚类系数的定义能够有效推广到其他模型当中。同时在有监督学习的链接预测场景中,由WCCLP构建的特征比现有的局部相似度算法构成的特征更具判别性。The weighted link prediction algorithm based on local traditional architecture only uses the topological attribute of the one-layer neighbor node. They are unable to reflect the contribution of common neighbors＇ neighbor node on the potential node pair,and calculate the influence of the degree of interconnection of neighbors. This paper analyzed the contribution of the neighboring nodes on the potential pairs from the dense level of the local structure,and proposed a similarity index called（ WCCLP） of the integrated weighted clustering coefficients,which could effectively enlarge the influence of the local structure on the prediction performance. Besides,WCCLP could easily be extended to the weighted local naive Bayesian link prediction model（ WLNB）. The experiments with unsupervised learning show that,compared with the existing local similarity algorithm,WCCLP achieves better and predictive results in many real data sets. The experimental effect of extending to WLNB also proves that weighted clustering coefficients can be effectively extended to other models. At the same time,under the classifier with supervised learning for link prediction,the feature construction with WCCLP are more discriminative than the features which derived from the existing local similarity algorithm.

关 键 词：加权网络 复杂网络 聚类系数 链接预测 加权聚类系数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]