A Gaussian Kernel Approach for Location Recommendations
Proceedings of the 1st KDMiLe - Symposium on Knowledge Discovery, Mining and Learning, ISSN 2318-1060, (2013)Abstract
The Location Based Social Networks (LBSN) are systems where the user can share the locations he/she has visited with his/her friends. These systems have become very popular because of the increasing number of users with smartphones equipped with GPS, which allow them to be connected to the internet wherever they are and geotag their current locations. In this context, the task to recommend new places for the user to visit is very important in order to improve the user experience while using the system. Several recommendation models have been proposed to solve this problem. In all these works, the final model is a combination of several specialized recommender algorithms, being the most recurrent ones based on social networks, collaborative filtering, and geographic information. Although geographic information is a key dimension in LBSNs, there is no work so far providing an empirical comparison between the existing recommendation algorithms solely based on geographic information. In this paper, besides providing this comparison, we propose a new geographic-aware recommendation algorithm based on a gaussian kernel to infer the regions more likely to be visited by the user. We conduct experiments on real world datasets of two popular LBSNs: Gowalla and Foursquare. Our experiments show that our approach outperforms the state-of-the-art geographic-aware recommenders proposed by the literature.