一种面向LDP的政府民意数据隐私保护方法

An LDP-oriented Privacy Protection Approach for Government Polls Data

为了保护政府民意调查中存在的隐私信息,提出了基于本地化差分隐私(LDP)的政务数据共享方法B-RAPPOR。上述方法在算法RAPPOR的基础上引入了数据分箱思想,通过等宽分箱将数据记录分入更小的数据域范围内,以克服当前LDP方法在数据域较大且数据量较少时统计误差大的问题。将B-RAPPOR与RAPPOR算法分别从频数估计、误差分析两方面进行了实验对比与分析。结果表明,经过B-RAPPOR隐私处理得到的数据相较于RAPPOR具有更好的数据效用性,解决了当前LDP方法在数据域较大时对数据量大小要求严格的问题。其次,以同样通过数据分箱思想改进的BCS、BRR为对照组,将B-RAPPOR算法与其分别从数据规模、隐私预算、数据域大小三方面进行了实验对比与分析。通过上述实验分析,界定了三种不同算法分别适用的场景。

To protect privacy information in government polls,B-RAPPOR,an improved LDP-based(Local Differential Privacy)approach is proposed.This approach combines data binning with the Randomized Aggregatable Privacy-preserving Ordinal Response(RAPPOR)algorithm.Equi-width binning is adopted to divide the data records into smaller data domains to overcome the problem of large statistical errors in the current LDP approaches when the data domain size is large and the data volume is small.B-RAPPOR is compared and analyzed with RAPPOR in terms of frequency estimation and error analysis.The experimental results show that the data obtained through the privacy processing of B-RAPPOR has better data utility than RAPPOR,which solves the problem of the current LDP approach that requires strict data size when the data domain is large.In addition,the BCS(Binning Count Sketch)and BRR(Binning Randomized Response)algorithms,which are also improved through the idea of data binning,are used as the control group,and the experimental comparison and analysis are carried out with the BRAPPOR(Binning RAPPOR)algorithm from the three aspects of data size,privacy budget,and data domain.Finally,through the above-mentioned experimental analysis,the applicable scenarios of three different algorithms are respectively given.