POI Recommender System with Various Effects

Abstract

Seoul National University More and more Location-based Social Networks(LBSNs) becomes significantly popular due to generalization of smartphones and tablets. With the high accessibility for LBSNs dataset, Point-of-Interest (POI) recommendation gets lots of attention not only in academic area but also in industry recently. As normal recommendation (move and product), it has data sparsity problem. In addition, as opposed to normal recommendation, it has to deal with physical distance. Hence, make the problem harder. To address the aforementioned problems, many researchers try to utilize assorted effects such as users preference, temporal effect, geographical effect and social effect. In this paper, we proposed UFPCF-G which utilize Users preference, different types of Friends preference, POI similarity and Users activity based Collaborative Filtering with Global Influence. We also explore extended meaning of friends. Unlike other research, we do not regard social relationship as the solely friends. In addition, we propose structure based and activity based global influence. We execute extensive experiment on two real LBSNs dataset and our experiment results outperform other baseline methods in terms of precision and recall.