Delay-tolerant Privacy-preserving Continuous Histogram Publishing Method
Yunfeng Zou, Chao Shan · 2022
With the emergence of privacy protection in data analysis, continuous publishing of data stream privacy protection statistical histograms has received widespread attention. Existing researches mainly concentrates on high-real-time publishing scenarios. If the proposed methods are applied to low-real-time publishing scenarios, they cannot effectively identify the stable part of the data stream, and cannot balance grouping errors and noise errors, resulting in pool availability. To solve the problem of continuous publishing of statistical histograms of low real-time data streams, under the premise of satisfying differential privacy constraints, a low-delay tolerant histogram continuous publication method 1-delay HCP (Histogram Continuous Publication Delayed by 1 Time Unit) and a high-delay tolerant histogram continuous publication method w-delay HCP (Histogram Continuous Publication Delayed by w Time Unit) were proposed. By estimating the histogram of the newly added data stream at each time, for the low-delay tolerant scene and the high-delay tolerant scene, based on the cached data from the time to be released to the latest time, two adaptive grouping methods for bucket count streams are proposed to group the bucket counts to be released; the original bucket count were replaced by the disturbed group mean, effectively balance the grouping error and the Laplace error, and reduce the release histogram error. Based on the real data set, the baseline method that directly adds noise to the histogram at each time, the RG method that uses retrospective grouping, the 1-delay HCP and w-delay HCP methods are compared and analyzed. The results show that under the same privacy budget constraint, the error of histogram released by 1-delay HCP is lower than Baseline and RG; w-delay HCP is better than 1-delay HCP in scenarios where higher delays can be tolerated.