TCENR: A Hybrid Neural Recommender for Location Based Social Networks

Abstract

Point-Of-Interests (POI) recommendation, an important application of location-based social networks (LSBN), has been extensively researched in recent years. This sub-field of recommender systems (RS) poses unique challenges due to high data sparsity and its relative complexity. An emerging technique is the use of deep neural networks to improve the performance of collaborative filtering (CF) based models. Recent works have successfully integrated such networks with external data, such as social networks, locations, categories and written reviews. In this paper, we propose a new method, Textual and Contextual Embedding-based Neural Recommender (TCENR). The suggested algorithm combines two types of neural networks to model the user-POI interactions based on implicit ratings, social networks, geographical locations and natural language reviews. Experiments on the Yelp dataset show that the proposed model is able to learn the complex interaction and enables improved recommendation performance.