Histogram Publishing Method Based on Differential Privacy

Abstract

Differential privacy does not care about the background knowledge of attackers, and can strongly protect the data information to be released. The distribution of differential privacy histograms based on groupings has drawn much attention from researchers and how to balance the approximation error caused by the group mean with the Laplace error caused by adding noise is the key. This paper proposes a method of APG (Affinity Propagation Clustering and Grouping algorithm) based on clustering grouping to distribute differential privacy histogram, which can effectively balance the approximate error with the Laplace error and improve the histogram posting accuracy. In this method, the index mechanism is used firstly to realize the sorting of the buckets. Then, through the clustering of AP algorithms, the sorted buckets are adaptively grouped to find the optimal grouping strategy and release the data. Experimental results on real data show that APG method is superior to GS, AHP and IHP methods in accuracy of distribution.