Privacy-preserving point-of-interest recommendation based on geographical and social influence

We investigate a privacy-preserving problem for point-of-interest (POI) recommendation system for rapidly growing location-based social networks (LBSNs). The LBSN-based recommendation algorithms usually consider three factors: user similarity, social influence between friends and geographical influence in. The LBSN-based recommendation system first needs to collect relevant information of users and then provide them with potentially interesting contents. However, sensitive information of users may be leaked when the recommendation is provided. In this article, we focus on preventing user’s privacy from disclosure upon geographical location and friend relationship factors. We propose a geographical location privacy-preserving algorithm (GLP) that achieves 〈r,h〉-privacy and present a friend relationship privacy-preserving algorithm (FRP) through adding Laplacian distributed noise for fusing the user trusts. Subsequently, we integrate the GLP and FRP algorithms into a general recommendation system and build a privacy-preserving recommendation system. The novel system enjoys the privacy guarantee under the metric differential entropy through theoretical analysis. Experimental results demonstrate a good trade-off between privacy and accuracy of the proposed recommendation system.