Differentially private trajectory event streams publishing under data dependence constraints

Abstract

Differential Privacy (DP), as a de facto privacy-preserving paradigm, can be widely applied in numerous event streams publishing schemes to protect private events (i.e., trajectory points of users). However, most existing schemes assume these events are independent, which is a vulnerable assumption that can lead to unforeseeable privacy degradation. The privacy-preserving for event streams publishing under data dependence is still challenging because existing dependence quantify models cannot dynamically measure the degree of dependent relationships between events. In this paper, we investigate the privacy degradation issue of trajectory event streams publishing and propose (Θ,ϵ)-dependent privacy mechanism to mitigate this issue. Specifically, we introduce spatial dependence to model the dependent relationship between events in trajectory data and extend it to quantify the influence of multiple dependent events on the target event. We define the notion of max dissimilarity to formalize the privacy level of DP mechanisms in trajectory event streams publishing. The proposed notion of global dependent coefficient helps the private mechanism to provide the expected privacy level by calibrating the added noise. We further achieve ω-event privacy protection over trajectory event streams under data dependence constraints with (Θ,ϵ)-dependent privacy mechanism. Theoretical analysis proves our scheme provides the expected privacy guarantee under data dependence constraints. Extensive experiments validate our scheme is efficient and effective.