Topic-Sensitive Location Recommendation with Spatial Awareness

The popularity of location-based social networks (LBSNs) has enabled us to better understand human behavior and preferences. The location recommendation problem is to provide personalized places of interest. Unlike traditional recommendation, the detailed information of user historical records is traced in LBSNs. Spatial pattern of user behavior and textual information associated with locations can contribute to a more precise recommendation system. In light of this challenge, we propose a topic-sensitive recommendation model with spatial awareness by exploiting both textual and spatial information. Specifically, we first implement latent Dirichlet allocation (LDA) model to learn the user preference on different topics by mining the latent textual information of locations. Then, a topic-sensitive probabilistic model is proposed to infer user expertise on each topic. Based on the estimated expertise, we combine opinions from other users to recommend locations for a target user. Finally we further enhance the recommendation quality through incorporating geographical influence. Experimental results on real-world LBSN datasets show that our proposed methods outperform the baseline techniques.