Research on Privacy Protection Technology for Data Mining
Zhiming Zhu, Xingzhen Wang · 2023
More and more sensitive data is being collected and processed by third parties for data mining and analysis, in order to solve the problem of how to achieve a certain balance between privacy protection and data analysis accuracy, in this thesis, a personalized local differential privacy protection method based on feature weights is proposed. The influence weight of feature attributes on data analysis is taken as the core basis of the whole perturbation method. The gradient response algorithm K-RR [1] is mainly used to realize random disturbance of data. However, in order to ensure the accuracy of subsequent data analysis as much as possible, the consideration of the influence of feature attributes on weight size, feature attribute dimension, and privacy budget allocation is introduced. In addition, the personalized privacy requirements of different data owners are fully considered. By setting the total privacy budget defined by data owners and constructing the sensitive attribute set defined by data owners, personalized local differential privacy protection is realized based on the influence weight of attributes. In addition, different privacy budget loss calculation methods are constructed for different attributes based on the influence weight of characteristic attributes to ensure data privacy.