Privacy Preserving Social Tie Discovery Based on Cloaked Human Trajectories

Discovering social connections of people has become a flourishing research topic considering the rich social information inferable from human trajectories. Existing social tie detection methods often require exact locations of users, which cause serious privacy concerns. Although cloaking is a common technique for location anonymization, it has rarely been applied in social tie detection due to the potential loss of significant location information. In this paper, we propose a semantic tree model for social tie detection, which supports different levels of privacy preserving and allows better understanding of location content of the cloaking regions. We propose a novel algorithm that can infer the social ties between users using only their cloaked trajectories without exposing their exact locations. We model the obscured regions generated by the cloaking algorithms in a semantic region tree and infer the similarity between two users based on their temporal and spatial relations in the tree. We evaluate our proposed approach using real trajectory dataset and show that our algorithm can identify social ties successfully with 15% higher accuracy compared with existing approach.