Geographic-categorical diversification in POI recommendations

Abstract

Nowadays, Recommender Systems (RSs) have been used to help users to discover relevant Points Of Interest (POI) in Location Based Social Network (LBSN), such as Yelp and FourSquare. Due to the main challenges of data sparsity and the geographic influence in this scenario, most of works about POI recommendations has only focused on improving the system’s accuracy. However, there is a consensus that just it is not enough to assess the practical effectiveness. In real scenarios, categorical and geographic diversities have been identified as key dimensions of recommendation utility. The few existing works are concentrated on just one of these concepts, singly. In this work, we propose a novel post-processing strategy to merge these concepts in order to improve the user interest in POIs. Our experimental results in Yelp datasets show that our strategy can improve users’ satisfaction, considering different RS and multiple diversification metrics. Our method is able to improve the diversity up to 120% without significant accuracy losses.