基于属性隐私的统计查询定价模型.

Current statistical query pricing models have not considered the problem that query results reveal sensitive attributes of datasets, making it difficult to incentivize sharing by compensating data providers accordingly. Therefore, this paper proposed a pricing model based on attribute privacy. Firstly, the model calculated query sensitivity based on the relaxed approximation Wasserstein mechanism (RAWM) proposed, improving efficiency by directly calculate the relaxed upper bound of output distribution pairs. Then, with bounding privacy loss, the model compensated data providers based on query sensitivity, noise variance and compensation parameters. Finally, by using cost-plus pricing on compensation, this paper designed several arbitrage-free pricing functions, which could be used in scenarios such as single compensation costs and multiple marginal costs. The experiment results show that the running time of calculating query sensitivity is reduced from linear complexity to constant complexity, with a utility cost of only 0.52% when data volume is 100 million. The pricing model provides fine-grained compensation to incentivize sharing. Pricing functions satisfy arbitrage freeness. [ABSTRACT FROM AUTHOR]