A Novel GAN based User Desensitization Data Generation Algorithm
Congcong Shi, Pengfei Yu, Xiuli Huang · Proceedings of the 2021 5th International Conference on Electronic Information Technology and Computer Engineering · 2021
In recent years, how to use user data rationally has become a hot topic of discussion. In order to solve the problem that existing user desensitised data generation methods lead to the unavailability of the user profile analysis function of applications, this paper proposes a generative adversarial network-based algorithm for generating user privacy desensitised data, which improves the usability of the data while ensuring data privacy. The algorithm is based on generative adversarial networks, using a differential autoencoder to extract user data features and migrate the features to the desensitised data; it also uses a multi-headed attention mechanism to intensively extract the main features of the user data; in order to ensure that the desensitised data can be used for the Android app's own user portrait analysis, its discriminator uses the app's own neural network learning model to adjust the output of the generator. Due to the stochastic nature of the differential auto-encoder learning process and the approximation process, the desensitised data are independent of the user's original data, thus protecting the user's private data. The desensitised data are guided by the discriminator with the characteristics of the original data, and the results of user profiling based on this desensitised data are approximately the same as the results of the original data analysis, which protects user privacy without compromising profiling.