BLR-MPC: A BERT-Based Location Recommendation Model with Micro-Profile Clustering

Online Social Networks (OSNs) and Location-Based Social Networks (LBSNs) have become essential platforms for users to share experiences, opinions, and check-ins. As a result, Location-Based Recommendation Systems (LBRSs) have emerged to suggest personalized Points of Interest (POIs). However, existing approaches face challenges such as data sparsity, cold-start issues, and the inability to incorporate contextual factors like local popularity and temporal dynamics. In this paper, we propose BERT-based Location Recommendation with Micro-Profile Clustering (BLR-MPC), a novel framework that integrates user micro-profile clustering with a BERT-based model to generate more accurate and personalized POI recommendations. BLR-MPC identifies regions of interest using DBSCAN clustering with micro-profiles, considering temporal context, and constructs semantic location embeddings using BERT, effectively capturing user preferences based on check-in sequences for active users. To address the cold-start problem, BLR-MPC utilizes a HITS-based model to recommend frequently visited locations by local experts when data is insufficient for new users. Experimental results on a real-world dataset demonstrate that BLR-MPC significantly outperforms existing methods in terms of precision and recall and effectively addresses key limitations of traditional LBRSs.