一种基于分类器的社交网络去匿名方法  

Classifier-based De-anonymization Method for Social Networks

作　　者：胡光武[1] 张平安[1] 马江涛 

机构地区：[1]深圳信息职业技术学院计算机学院,广东深圳518172 [2]解放军信息工程大学数学工程与先进计算国家重点实验室,河南郑州450001 [3]郑州轻工业学院计算机与通信工程学院,河南郑州450002

出　　处：《深圳信息职业技术学院学报》2017年第3期6-11,共6页Journal of Shenzhen Institute of Information Technology

基　　金：广东省自然基金资助项目(2015A030310492);深圳市基础研究项目(JCYJ20160301152145171);河南省科技厅科技攻关项目(162102410076;162102310578);河南省高等学校重点科研项目(16A520062;17A520064)

摘　　要：为保护社交网络用户隐私,验证社交网络提供商对社交数据进行匿名保护的有效性,本文提出了一种基于随机森林分类器的社交网络去匿名方案。首先,方法将社交网络的去匿名问题转化为辅助网络与匿名网络之间的节点匹配问题,然后把网络结构的特征(如节点度中心性、中介中心性、亲近中心性、特征向量中心性)作为节点特征向量训练分类器。最后,根据训练后分类器的判定结果,方法实现两个网络节点间的匹配,完成了去匿名测试过程。利用真实的学术社交网络数据进行了方案评估,结果表明本文提出的方法优于已有方案,在0.5%假阳性率的情况下仍能实现81%的社交网络节点去匿名化效果。In order to protect social network users＇ privacy and test the vulnerability of the anonymization methods that provided by social networks＇ provider, in this paper, we propose a random forest classifier-based automatic de-anonymize method for social networks. First, we convert the de-anonymization problem among social networks into two nodes mapping problem between two graphs. Then, we utilize social network structure features, such as node＇ s degree, hetweenness centrality, closeness centrality, eigenvector centrality, to train binary classifier. Then, we predict whether two nodes belong to the same kind of nodes through classifier. Therefore, we can map the node automatically through trained binary classifier to perform de-anonymization process. Finally, the experiments that conducted on real academic social networks＇ datasets demonstrate that the effective of our proposed method, which outperformances existing methods and it can de-anonymizc 81% of network nodes in the worst scenario of 0.5% false positive ratio.

关 键 词：社交网络 去匿名 节点匹配 社交网络结构 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]