一种面向直方图发布的均衡差分隐私保护方法  

Balanced Correlation Differential Privacy Protection Method for Histogram Publishing

作　　者：杨旭东 高岭[1,2] 王海 郭红波[1] 郑杰[1] YANG Xu-Dong;GAO Ling;WANG Hai;GUO Hong-Bo;ZHENG Jie(Department of Information Science and Technology,Northwest University,Xi’an 710127;Department of Computer Science,Xi’an Polytechnic University,Xi’an 710048)

机构地区：[1]西北大学信息科学与技术学院,西安710127 [2]西安工程大学计算机科学学院,西安710048

出　　处：《计算机学报》2020年第8期1414-1432,共19页Chinese Journal of Computers

基　　金：国家重点研发计划(2019YFC1521400);国家自然科学基金(61672426,61572401)资助。

摘　　要：作为一种常用的数据发布方法,直方图数据发布因其直观便捷的特点得到了广泛关注.直方图数据发布在带来方便的同时也面临着隐私泄露的风险.当前基于差分隐私的保护方法虽然提高了一定发布安全性,但仍然存在以下问题:(1)现有的差分隐私保护方法往往忽略直方图发布数据之间的关联特性;(2)同时,现有的方法缺乏有效评估直方图间接隐私泄露风险的方法;(3)现有的方法难以实现全面均衡的直方图隐私保护.本文针对上述问题,通过引入关联隐私泄露评估量化机制,设计了一种面向直方图数据发布的均衡差分隐私保护方法.首先,结合作用域的马尔科夫模型定义了直方图关联隐私;然后,基于隐私泄露损失因素,提出一种多指标决策的隐私泄露损失评估方法;最后,借鉴Nash博弈与Stackberg博弈思想,设计一种均衡差分隐私保护直方图发布方法.通过在两个不同数据集上的实验,验证了本文所提均衡隐私保护方法的有效性与鲁棒性,并证明了本文所提出的均衡差分隐私保护方法均衡地保护了直接与间接隐私泄露,且优于AHP、GS直方图隐私保护发布方法.The histogram data as its feature of concise,clear and intuitive form of data expression is widely used in statistics,analysis,and other fields.The sanitary administrative organs use histogram data to issue epidemic warnings and the family-planning bureaucracy use histogram data to display and analyze population distribution.As a commonly used data publishing method,histogram data publishing has received extensive attention due to its intuitive and convenient features.Although histogram data publishing increases the efficiency of users’acquisition of data knowledge,privacy disclosure in published data is still the main factor hindering its development.To cope with this challenge,histogram data publish under privacy protection has attracted academic attention.Differential privacy as a strictly provable approach to privacy protection has gained more attention than other approaches.However,the existing differential privacy protection methods often ignore the association characteristics between the histogram published data and there are three main problems:(1)The effect of data association privacy in histogram sequence on privacy protection;(2)It is difficult to give full consideration to the direct and indirect privacy of data;(3)The difficulty of equalizing the direct and indirect privacy of histogram data.Aiming at the protection privacy in publishing histogram data,we design a balanced differential privacy protection method for histogram data publishing by introducing the associated privacy leakage assessment and quantization mechanism.First,we define the histogram spatial correlation and privacy according to the scope-based Markov mode.The Markov transfer probability between spatial adjacent histograms is used to represent the correlation privacy between histograms,and it is shown that correlation privacy can cause the leakage of comprehensive privacy.Second,we propose a comprehensive quantitative evaluation method for multi-factor privacy leakage,which by combining various loss factors caused by privacy leaka

关 键 词：直方图发布 差分隐私 关联隐私 马尔科夫 NASH均衡 博弈论 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]