Regionalized location obfuscation mechanism with personalized privacy levels

Abstract

Global Positioning Systems are now a standard module in mobile devices, and their ubiquity is fueling rapid growth of location-based services (LBSs). This poses the risk of location privacy disclosure. Effective location privacy preservation is foremost for various mobile applications. Recently two strong privacy notions, geo-indistinguishability and expected inference error, are proposed based on statistical quantification. They are complementary for limiting the leakage of location information. In this paper, we continue to study the differential privacy preservation of location obfuscation mechanism based on PIVE framework proposed by Yu, Liu and Pu on ISOC Network and Distributed System Security Symposium (NDSS) in 2017. Since PIVE fails to offer differential privacy guarantees on adaptive protection location set (PLS) as claimed, we develop DPIVE, a regionalized location obfuscation mechanism with two phases. In Phase I, we determine disjoint sets by partitioning all possible positions such that different locations in the same set share the common PLS. In Phase II, we construct a probability distribution matrix by exponential mechanism in which each row has its own sensitivity of utility (diameter of PLS).This approach utilizes the relationship between two privacy notions based on the user-defined inference error threshold and the prior knowledge about user's location. Moreover, we introduce PDPIVE, a personalized privacy framework, to achieve that each location has its own privacy level on two privacy control knobs, minimum inference error and differential privacy parameter. Experiments with two public datasets demonstrate that our mechanisms have the superior performance typically on skewed locations.

Geo-obfuscation Mechanisms for Spatial Crowdsourcing via Multi-Objective Evolutionary Optimization

“…Yu et al [16] formally study their relationship and combine them by adding personalized error lower bounds. Later, DPIVE mechanism is presented and solves the previous privacy theory problem of intersections among protection location sets [7]. Recently, Liu et al [17] propose an obfuscation mechanism with the Gamma distribution and use a game-theoretic approach to maximize two-users' utilities while preserving desired location privacy radius.…”

“…Secondly, most of existing privacy mechanisms ignore the balance between privacy protection and practical quality. Currently, there is no standard optimization of conflicting metrics in a global sense when users have personalized requirements of distortion privacy [5], [7]. This motivates us to combine differential location privacy and MOEA to achieve the optimal trade-off between service quality loss and average expected inference error globally in SC.…”

“…In the PART 1, based on the condition ( 13), we first execute QK-means algorithm from [7] on each cell X i to find the optimal number k i of disjointed PLSs included. Then we randomly select k i or k i −1 centers to make centralized clustering within each cell X i in ascending order of distance.…”

“…For each true location x ∈ Φ j , the utility (scoring) of output x ∈ Y j can be defined as u(x, x ) = −d(x, x ). The sensitivity of u for each PLS Φ j is the diameter of Φ j [7].…”

“…and (4) Small-scale location domain and task allocation. Current LDP protection mechanisms [6] mainly adopt the linear programming methods to improve the allocation efficiency that result in much higher computational complexity, while a previous scheme DPIVE [7] can only be applied separately in small-scale disjoint domains.…”