基于动态ε的社会网络差分隐私保护  

Social Network Differential Privacy Protection Based on Dynamic ε

作　　者：刘振鹏 王烁 张彬 孙静薇 LIU Zhenpeng;WANG Shuo;ZHANG Bin;SUN Jingwei(School of Cyber Security and Computer,Hebei University,Baoding 071002,China;Center for Information Technology,Hebei University,Baoding 071002,China;School of Electronic Information Engineering,Hebei University,Baoding 071002,China)

机构地区：[1]河北大学网络空间安全与计算机学院,河北保定071002 [2]河北大学信息技术中心,河北保定071002 [3]河北大学电子信息工程学院,河北保定071002

出　　处：《郑州大学学报（理学版）》2019年第4期56-62,共7页Journal of Zhengzhou University:Natural Science Edition

基　　金：河北省创新能力提升计划项目(179676278D,17455309D);教育部“云数融合科教创新”基金项目(2017A20004);河北省自然科学基金项目(F2019201427)

摘　　要：针对权重社会网络发布算法中使用全局统一的ε值而导致隐私保护不均衡的问题,提出一种动态ε社会网络差分隐私保护方法.使用马尔可夫聚类(Markov cluster algorithm,MCL)和Chameleon混合聚类把社会网络图划分成若干个簇,根据每个簇中边的权重信息使用ε函数f(x)来确定簇的ε值,对带有大权重边的簇添加较多的服从拉普拉斯分布的噪声,该方法满足ε-差分隐私模型.实验表明,动态ε使每个簇能添加合适的噪声,比全局统一设定ε值有更好的数据效用性.该方法有效解决了权重社会网络中的隐私保护不均衡问题.Aiming at the problem of uneven privacy protection caused by global uniformεvalue in weighted social network publishing algorithm,a dynamicεsocial network differential privacy protection method was proposed.The Markov cluster algorithm(MCL)and Chameleon hybrid clustering were used to divide the social network graph into several clusters.Theεfunction was determined by using theεfunction f(x)according to the weight information of the edges in each cluster.Adding more Laplace-dependent noise to clusters with large weights,this method satisfied theε-differential privacy model.Experiments showed that dynamicεenabled each cluster to add appropriate noise,which had better data utility than globally settingεvalue.This method effectively solved the problem of unbalanced privacy protection in weighted social networks.

关 键 词：动态ε 差分隐私 社会网络 MCL聚类 CHAMELEON算法 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]