Abstract
Data trading has attracted increasing attention over the years as a cost-effective business paradigm, probably producing a tremendous amount of economic value. However, the study of query-based trading in the user data market is still in the initial stage. To design a practical user data trading mechanism, we have to consider three major challenges: privacy concern, compensation cost minimization and revenue maximization in a Bayesian environment. By jointly considering these challenges, we propose a profit-maximizing mechanism for user data trading with personalized differential privacy, called READ, which comprised two components, READ-COST for cost minimization and READ-REV for revenue maximization. Especially, READ adopts personalized differential privacy to satisfy each data owner’s diverse privacy preferences. READ-COST greedily selects the most cost-effective data owner to achieve the sub-optimal data query cost. Given this query cost, READ-REV calculates the maximum expected revenue in a Bayesian setting. Through rigorous theoretical analysis and real-data based experiments, we demonstrate that READ achieves all desired properties and approaches the optimal profit.
LightDP: towards automating differential privacy proofs
