基于树状朴素贝叶斯模型的社会网络关系预测  

Tree augmented Nave Bayesians model for ties prediction in social networks

作　　者：伍杰华[1,2] 

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]华南理工大学信息科学与技术学院,广州510641

出　　处：《计算机应用》2013年第11期3134-3137,3200,共5页journal of Computer Applications

基　　金：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)

摘　　要：在社会网络关系预测研究领域,把基于拓扑结构信息的共邻节点属性作为相似性度量的预测模型应用比较广泛,但是该类算法具有较强的假设独立性,不能完全反映社会网络的"链接"结构。引入树状朴素贝叶斯(TAN)分类模型,采用信息熵度量节点对的角色,赋予共邻节点集合差异化的贡献权重进行社会关系预测,同时把模型推广到CN,AA和RA等3种基于相似度的链接预测算法中。对5个真实社会网络采用AUC和ROC曲线进行实验评价后证明,该模型能够在深入挖掘共邻节点对贡献及解决共邻节点角色独立性的基础上提高预测精确度,同时为该类模型的研究提供一种新的方案。In the research field of social ties prediction, taking common neighbors property as the similarity-based topological measure to carry the task of prediction has been widely used and better results have been achieved, which nevertheless has strong assuming independence and can not reflect the ＂ link＂ and related network structure. This paper proposed a new measure of link prediction by introducing a Tree Augmented Na＇fve Bayesians （TAN） classification model, which used information entropy measure to define the role of the node pair and gave differentiated neighbors set contribution to the task of social ties prediction, and then it was extended to Common Neighbor （CN）, Adamic-Adar （AA） and Resource Allocation （RA） similarity-based prediction algorithms. The experimental evaluation by Area Under ROC Curve （AUC） and Receiver Operating Characteristic （ROC） curve on five real social networks prove that the proposed model can mine the latent common neighbors＇ contribution and alleviate the independence hypothesis, which leads to enhance the accuracy of link prediction.

关 键 词：社会网络分析 关系预测 链接预测 共邻节点 贝叶斯模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]