Nearby places: on location-based pruning for point-of-interest recommendation

Abstract

Point-of-interest (POI) recommendation in location-based social networks has been widely researched as a mechanism to help users discover new venues to visit in a city. Despite recent efforts, the currently available test collections trivially favor geography-aware POI recommenders by overlooking the fact that users will most often prefer nearby POIs anyway. As a result, it remains unclear what role other contextual factors play in a more realistic scenario where geography is constrained by definition given the user's location. To close this gap, we introduce a large-scale test collection for context-aware nearby POI recommendation, which includes several geography-constrained test cases, enriched with temporal and weather contexts in two large cities in the Americas. Also, we describe our proposed evaluation methodology for the nearby POI recommendation task and present a breakdown analysis of the state-of-the-art POI recommenders results as reference results for the new collection.