Behavior sequence mining model based on local differential privacy

Abstract

Most of local differential privacy frameworks target statistics on certain privacy behaviors of users, but not behavior sequence. In this paper, we explore and propose a behavior sequence mining model that satisfies the local differential privacy requirement to settle the matter. We decompose their potential behavior sequence into multiple temporal pairs that are computed by the server to infer indirectly behavior sequence of users, shrinking the statistical sample space with adjacent temporal pairs to reduce statistical errors. The experiment takes an example, trajectories of users can be inferred by their location information, to demonstrate the effect our model achieved. It shows that the model can approximate users' trajectories under the requirement of local differential privacy.