Data desensitization mechanism of Android application based on differential privacy

Abstract

In recent years, the mining and analysis of user data by Android applications have posed the risk of privacy breaches. However excessive permission control can affect the usability of applications. A mechanism that balances security and usability is urgently needed. In this paper, a data desensitization mechanism for Android applications based on differential privacy techniques is proposed. The mechanism can address the privacy protection of data flows generated by the interaction between users and Android applications. In order to solve the problem of constraints on the basic functions of the application caused by privacy security technique, this paper introduces a differential privacy mechanism based on Gaussian process. The mechanism performs hyperparametric optimization methods that combine sparse approximations and classification results. Also, by specifying the global sensitivity of the differential privacy budget specific randomization algorithm, the mechanism selects parameters with a specific probability to obtain the most effective parameter combination. Experimental results show that the differential privacy technique based on Gaussian process further enhances the availability of Android application data while obtaining the same privacy protection effect compared with ordinary differential privacy mechanisms.