Improving Accuracy of Interactive Queries in Personalized Differential Privacy

Privacy-preserving data publishing has been an important research field in the era of big data. Various privacy protection schemes have been proposed to balance privacy and utility. Personalized differential privacy (PDP) is especially noteworthy as it offers stronger privacy guarantees, taking into account the diverse privacy requirements of users. However, existing PDP mechanisms produce query results with low accuracy, which leads to poor data utility. Specifically, PDP is realized by sampling data and invoking on differential privacy, thus the problem of poor accuracy in differential privacy itself is brought into PDP. Interactive queries is a natural setting in differentially private publishing scenarios. Considering the need for interactive queries in PDP, we firstly establish a privacy budget allocation method by equalizing the common privacy budget and then recycling the remaining privacy budget, and then propose a combined mechanism to answer multiple queries. As accurately, for each round of query, there is a high probability that user’s data with high record sensitivity will not be sampled, thereby reducing the overall sensitivity. And a smaller noise is added according to the corresponding allocated privacy budget. Finally, the combined mechanism is proved to meet the privacy demands, and the experiment results executing on synthetic datasets and real datasets demonstrate that the combined mechanism under the privacy budget method can achieve better accuracy than the traditional mechanisms.