Personalized Differential Privacy in the Shuffle Model

Personalized local differential privacy is a privacy protection mechanism that aims to safeguard the privacy of data by using personalized approaches, while also providing practical data analysis results. It offers more flexible and precise privacy protection capabilities compared to traditional local differential privacy. By employing distinct privacy protection strategies for different users, it can better meet users’ privacy requirements while minimizing the impact on data. However, existing mechanisms for personalized local differential privacy suffer from issues such as low query accuracy and poor data utility. These issues need to be addressed to improve the effectiveness and practicality of personalized local differential privacy. In this work, we have proposed a framework of personalized differential privacy in the shuffle model. This framework introduces individualized perturbation to the data locally and then reshuffles the records in the dataset, disrupting the original order of the data and breaking the correlations between data points. This approach aims to achieve a higher level of privacy protection. We have validated the practicality and superiority of this framework on four different types of real-world datasets.