Link Prediction Using Protected Location History

This paper investigates the link prediction problem in location-based social networking services (LBSNS) with protected location history. While former approaches mainly utilize the accurate locations, the relevant data we analyzed are modeled by a location privacy protection model called k-anonymous spatial-temporal cloaking model (KSTCM) which perturbs the location-related records on both temporal and spatial dimensions. We also propose a KSTCM-based co-located relationship model based on which various contextual features are extracted to help building the prediction model. Furthermore, we study the extent to which link between two users can be inferred from the co-located situation in space, time and degree of privacy protection. The experimental results show that our link prediction model can obtain a very high accuracy.