Catching All Pokémon: Virtual Reward Optimization With Tensor Voting Based Trajectory Privacy

Abstract

The widespread adoption of mobile devices with global positioning system enables location-based games (LBGs) to use real world maps, while locations and objectives in LBGs can make the progression, achievements, and virtual rewards feel more palpable and entertaining. However, allowing location sharing in LBGs gives dishonest parties opportunities to learn users’ trajectories, which compromises the users’ privacy. In this paper, we propose a novel scheme jointly maximizing LBG players’ virtual rewards while preserving their trajectory privacy. Briefly, we first introduce a quantitative machine learning-based approach to model trajectory inference attacks via tensor voting. Then, to thwart this attack, we propose a tensor voting-based $k$ -anonymous obfuscation strategy. Considering the trajectory privacy concerns and power constraint of hand-held mobile devices, we mathematically formulate the LBG players’ virtual reward maximization optimization into the mixed integer problem and develop the feasible solutions. Simulation results and analysis show that the proposed scheme can effectively preserve LBG players’ trajectory privacy against tensor voting based inference attacks while maximizing LBG players’ virtual rewards.