Next point of interest (POI) recommendation

Abstract

Next Point-of-Interest (POI) Recommendation systems nowadays often assume that the users' check-in records are accurate. However, the accuracy and certainty of a user's check-in history may not be guaranteed in a real-world application due to various reasons. In order to make POI Recommendation systems overcome this problem, we first processed and analyzed real-world data to investigate the key influencer of users' decisions. This report then proposes a novel model that utilizes the users' past spatial and temporal information to predict users' intentions and offer them suggestions on where to go. Experiments were conducted on 3 sets of real-world data. The performance of the model was also compared with other baseline models to demonstrate the advantages of this model.