基于平滑聚类的差分隐私数据自动脱敏仿真  

Simulation of Automatic Desensitization of Differential Privacy Data Based on Smooth Clustering

作　　者：赵高丽[1,2] 宋军平 ZHAO Gao-li;SONG Jun-ping(School of Information and Engineering Henan Institute of Science and Technology,Xinxiang Henan 453000,China;School of Information Engineering Wuhan University of Technology,Wuhan Hubei 430070,China)

机构地区：[1]河南科技学院,河南新乡453000 [2]武汉理工大学,湖北武汉430070

出　　处：《计算机仿真》2020年第8期100-104,共5页Computer Simulation

摘　　要：传统隐私数据保护方法忽略了采取最小准则调整隐私预判,导致保护隐私数据效率低、精准低等问题,提出基于平滑聚类的差分隐私数据自动脱敏方法。对数据采取平滑聚类计算,将不确定数据点重新还原固定信息,并划分到合适的簇。构建差分隐私模型,通过差分隐私机制引入噪声,将干扰后的数据重组后传送给用户,降低敏感数据外泄几率。使用基于BIRCH的差分隐私数据自动脱敏方法,采取最小准则调整隐私预判,对发布的CF树进行加噪。根据计算所得的节点搜索概率,对节点使用隐私预判分配。同时运用线性回归与迭代解决方法中存在的不一致约束性问题,实现数据自动脱敏。仿真结果表明,所提方法能高效地保护隐私数据,保护数据信息稳定性强,且实用性较高。Traditionally,the privacy data protection method ignores to adjust the privacy prediction by the minimum criterion,leading to low efficiency and low accuracy.Therefore,a method of automatic desensitization of differential privacy data based on smooth clustering was put forward.Firstly,we restored the fixed information of uncertain data points by the smooth clustering,and then placed them into appropriate clusters.Secondly,we constructed a differential privacy model,and introduced the noise through the differential privacy mechanism.After recombining the disturbed data,we transmitted it to users,and thus to reduce the leakage probability of sensitive data.Thirdly,the method of automatic desensitization of differential privacy data based on BIRCH was adopted.Moreover,we adjusted the privacy prediction by the minimum criterion,and added noise on the published CF tree.According to the node search probability,the nodes were assigned with privacy prediction.Meanwhile,we used linear regression and iteration to solve the inconsistent constraints,so as to realize the automatic desensitization for data.Simulation results show that the proposed method can effectively protect the privacy data,the stability and practicability of data information is higher.

关 键 词：平滑聚类 差分隐私 自动脱敏 最短路径 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]