Histogram Publishing Algorithm Based on Sampling Sorting and Greedy Clustering

Abstract

The data produced by differential privacy histogram publishing algorithm based on grouping has low usability due to large approximation error and Laplace error. To solve this problem, a histogram publishing algorithm based on roulette sampling sort and greedy partition is proposed. Our algorithm combines the exponential mechanism with the roulette sampling sorting method, arranges the similar histogram bins together with a larger probability by the utility function and the restriction on the number of sampled entity. The greedy clustering algorithm is used to partition the sorted histogram bins into groups, and the error among histogram bins in each group is reduced by optimizing the lower bound error of the grouping. Extensive experimental results show that the proposed algorithm can effectively improve the usability of published data under the premise of satisfying differential privacy.