Diversifying Top-k Point-of-Interest Queries via Collective Social Reach

Abstract

By "checking into'' various points-of-interest (POIs), users create a rich source of location-based social network data that can be used in expressive spatio-social queries. This paper studies the use of popularity as a means to diversify results of top-k nearby POI queries. In contrast to previous work, we evaluate social diversity as a group-based, rather than individual POI, metric. Algorithmically, evaluating this set-based notion of diversity is challenging, yet we present several effective algorithms based on (integer) linear programming, a greedy framework, and r-tree distance browsing. Experiments show scalability and interactive response times for up to 100 million unique check-ins across 25000 POIs.