Re-identification of Vehicular Location-Based Metadata

Abstract

Amid the flourish of various data services, the privacy problems on metadata have received sufficient attention. Generally, the identity is the most sensitive attribute in metadata as identity is the key linking all attributes together. Thus, quite a few methods, such as dummy and k-anonymity, have been applied to actual applications to protect the identity . However, we still argue that the identity is very likely to be disclosed. In this paper, we study the re-identification problem in the seemingly privacy-preserving VLBS (Vehicular Location-Based Service). We find that the trajectories of vehicles are highly unique after studying 131 millions mobility traces of taxis. More specifically, the experiments demonstrate that only four spatio-temporal points are sufficient to uniquely re-identify the vehicle, achieving an accuracy of 95.35%. This indicates that there exists a high risk of re-identification in VLBS even identity has been protected by traditional methods.