Privacy-Preserving POI Recommendation Using Nonnegative Matrix Factorization

Abstract

Location based social networks (LBSNs) have become an essential part of life for many smartphone users. With the sheer volume of new information in LBSNs produced every day, people can easily feel overwhelmed when deciding which places to visit, e.g., restaurants, grocery stores, bars. Point-of-interest (POI) recommender systems are there to help people find their favorite places. To make recommendations, the system needs to learn users' preference, which usually requires their check-in data. This can potentially deter people from using the system because personal location and check-in data are considered as users' privacy and many do not feel comfortable sharing the information with other parties. In this paper, we propose a nonnegative matrix factorization (NMF) based privacy-preserving POI recommendation framework, in which the latent factors in NMF are learned on user group preference instead of individual user preference. Recommendations are made by personalizing the group preference on user's local devices. There are no location or check-in data collected from the users anywhere throughout the learning and recommendation processes.