Protecting Location Semantics under Local Differential Privacy

With the widespread proliferation of mobile devices, location-based services (LBS) are gaining increasing popularity due to the emergency of convenient context-aware services. The collection and analysis of locations in LBS potentially lead to privacy risks due to the nature of locations. Several existing works try to protect location privacy under local differential privacy (LDP), which is a state-of-the-art privacy preservation model. However, these works still pose a risk to user privacy because untrusted data collectors can exploit location semantics to infer the user’s true location. In this paper, we propose a novel approach LSLDP to protect location semantics under LDP. In LSLDP, we first propose a novel LDP method SRLS that carefully combines physical distance and semantic distance during the perturbation process. Additionally, we design a location release method based on the Expectation Maximization (EM) reconstruction method to rectify estimation errors stemming from SRLS. Comprehensive experiments on real-world datasets substantiate the effectiveness of LSLDP.