I Know Where You Want to Go Next: Contrastive Preference Modeling for Next Location Recommendation

Keywords: Point-of-Interest; location recommendation; sequential recommendation
Abstract Type: Paper Abstract

Abstract

Recommending next location is a highly valuable and common need in many location-based services such as destination prediction and route planning. State-of-the-art models learn mobility patterns merely in users’ historical check-in sequences while overlooking the significance of modeling user preferences. In this paper, we propose a novel Point-of-Interest Transformer (POIFormer for short) for end-to-end next location recommendation. POIFormer comprises three major modules: history encoder, query generator, and preference decoder. The history encoder is designed to model mobility patterns from historical check-in sequences, while the query generator is for explicitly modeling user preferences to generate user-specific intention queries. Then, the preference decoder makes prediction of the user's next location. Extensive comparisons with existing methods and ablation studies on four real-world datasets demonstrate the effectiveness and superiority of the proposed method under various settings.