A deep meta-level spatio-categorical POI recommender system

With the support of a wide range of location-based applications, location-based social networks have attracted the attention of academia, industry, and users by introducing several services. Moreover, point-of-interest (POI) recommendation becomes an indispensable task in these networks. Several studies on POI recommender systems have recently been introduced. These algorithms provide more opportunities to improve recommendation performance by gaining a better understanding of user behavior. However, most existing works focus on recent check-ins without effectively exploiting challenging information that may reflect users’ preferences. These details include reviews, location, and additional venue information. Users usually care about the physical distance, reviews left by other visitors, and main categories of the recommended POIs. In this article, we address these issues by proposing a deep meta-level spatio-categorical POI recommender system (DML-SC) to capture and learn users’ behaviors by incorporating spatial and categorical information. Firstly, we use a recurrent neural network to detect users’ sentiment polarity based on reviews. We also integrated an attention mechanism to effectively extract the most relevant sequences. Then, we placed a convolutional neural network that aims to detect users’ preferences from heterogeneous inputs based on two influential factors, namely spatial and categorical. Experimental results on real datasets show the effectiveness of the proposed framework.