LGAN-DP: A novel differential private publication mechanism of trajectory data

Abstract

With the popularity of mobile Internet, big data of trajectory not only provides convenience for people, but also brings privacy leakage. In order to overcome the problems in the current works such as insufficient accuracy, insufficient efficiency and insufficient protection of trajectory data privacy in trajectory data publishing methods, a trajectory data differential privacy publishing mechanism (LGAN-DP) based on LSTM-GAN is proposed in this paper, which uses a long short-term memory network and a generative adversarial network to generate synthetic trajectories. We design a trajectory loss function to judge the trajectory similarity loss of the synthetic trajectories trained by the trajectory data deep learning model proposed in this paper. LGAN-DP processes the result set of trajectory data publishing by using differential privacy. The experimental results show that LGAN-DP can better guarantee the balance between the privacy and availability of trajectory data on the MI. The availability of trajectory data is improved by 40%∼50% as compared with the existing methods, with less time complexity.