A Utility-Preserving and Scalable Technique for Protecting Location Data with Geo-Indistinguishability

Location-based apps provide users with personalized services tailored to their geographical position. This is highly-beneficial for mobile users, who are able to find points of interest close to their location, or connect with nearby friends. However, sharing location data with service providers also introduces privacy concerns. An adversary with access to fine-grained user locations can infer private details about individuals. Geo-indistinguishability (GeoInd) adapts the popular differential privacy (DP) model to make it suitable for protecting users’ location information. However, existing techniques that implement GeoInd have major drawbacks. Some solutions, such as the planar Laplace mechanism, significantly lower data utility by adding excessive noise. Other approaches, such as the optimal mechanism, achieve good utility, but only work for small sets of candidate locations due to the use of computationally-expensive linear programming. In most cases, locations are used to answer online queries, so a quick response time is essential. In this paper, we propose a technique that achieves GeoInd and scales to large datasets while preserving data utility. Our central idea is to use the composability property of GeoInd to create a multiple-step algorithm that can be used in conjunction with a spatial index. We preserve utility by applying accurate GeoInd mechanisms and we achieve scalability by pruning the solution search space with the help of the index when seeking high-utility outcomes. Our extensive performance evaluation on top of real location datasets from social media apps shows that the proposed technique outperforms significantly the benchmark in terms of utility and/or computational overhead.