CLSPRec: Contrastive Learning of Long and Short-term Preferences for Next POI Recommendation

Abstract

Next point-of-interest (POI) recommendation optimizes user travel experiences and enhances platform revenues by providing users with potentially appealing next location choices. In recent research, scholars have successfully mined users' general tastes and varying interests by modeling long-term and short-term check-in sequences. However, conventional methods for long and short-term modeling predominantly employ distinct encoders to process long and short-term interaction data independently, with disparities in encoders and data limiting the ultimate performance of these models. Instead, we propose a shared trajectory encoder and a novel Contrastive learning of Long and Short-term Preferences for next POI Recommendation (CLSPRec) model to better utilize the preference similarity among the same users and distinguish different users' travel preferences for more accurate next POI prediction. CLSPRec adopts a masking strategy in long-term sequences to enhance model robustness and further strengthens user representation through short-term sequences. Extensive experiments on three real-world datasets validate the superiority of our model. Our code is publicly available at https://github.com/Wonderdch/CLSPRec.