关键词: 局部差分隐私, 随机响应, 链接攻击, 隐私保护

Abstract: The traditional interactive differential privacy model perturbs the data query results, which cannot satisfy the users?? diverse needs for data. In order to effectively use the data and meet the privacy protection requirements, this paper uses the idea of local differential privacy to realize the link attack protection for the data set on the basis of random response. Firstly, this paper studies how to choose the random conversion matrix P better according to the distribution of the original data, to better realize the link privacy protection and to achieve the balance between data utility and privacy protection, so as to avoid the identity disclosure and attribute disclosure. Secondly, for sensitive attribute, quasi-identifier (QI) attributes and their combination between them, this paper discusses the corresponding privacy protection and data utility maximization methods, and gives a data perturbation algorithm. Finally, on the basis of the mean and variance of the known data distributions, experiments are used to verify the KL-divergence and the chi-square between the original and the disturbance data. The experimental results show that the randomization in this paper can lead to less utility loss.

Key words: local differential privacy, randomized response, linking attack, privacy protection