A Personalized Geographic-Based Diffusion Model for Location Recommendations in LBSN

Abstract

Location Based Social Networks (LBSN) have emerged with the purpose of allowing users to share their visited locations with their friends. Foursquare, for instance, is a popular LBSN where users endorse and share tips about visited locations. In order to improve the experience of LBSN users, simple recommender services, typically based on geographical proximity, are usually provided. The state-of-the-art location recommenders in LBSN are based on linear combinations of collaborative filtering, geo and social-aware recommenders, which implies fine tuning and running three (or more) separate algorithms for each recommendation request. In this paper, we present a new location recommender that integrates collaborative filtering and geographic information into one single diffusion-based recommendation model. The idea is to learn a personalized ranking of locations for a target user considering the locations visited by similar users, the distances between visited and non visited locations and the regions he prefers to visit. We conduct experiments on real data from two different LBSN, namely, Go Walla and Foursquare, and show that our approach outperforms the state-of-art in most of the cities evaluated.