A differential privacy based probabilistic mechanism for mobility datasets releasing

With the rapid popularization and development of the global positioning systems, location-based services (LBSs) are springing up to provide mobile internet users with door-to-door services. The users' privacy becomes one of the main concerns of such services, as location data reflects various sensitive information, such as home address, employment and even health conditions. Releasing the aggregated mobility datasets, i.e., the population of mobile users at different regions in the area, is one of the solutions in solving the privacy concerns that covers the individual users' information and accepted as a valid privacy preserving method in releasing mobility datasets. However, in a recent research, by exploiting the uniqueness and regularity of mobility data, individual trajectories can be recovered from the aggregated mobility datasets with accuracy about 73-91%. In this paper, we propose a novel differential privacy based probabilistic mechanism for mobility datasets releasing (DP-Mobi), in which the privacy preserved population distributions are generated and released to support LBSs. We employ a probabilistic structure count min sketch in the mechanism to count the number of users at different regions, and add noise drawn from Laplace distribution to perturb the sketches. Meanwhile, we prove the perturbed sketches satisfy differential privacy, so that the users are able to control the privacy level by tuning the parameters of Laplace distribution. Through evaluation, we show that comparing with another privacy preserving approach in resisting the attack model, our mechanism DP-Mobi achieves 8% more recovery error with the same utility loss.