Next PoI Recommendation Based on Graph Convolutional Networks and Multiple Context-Awareness

Next Point-of-interest recommendation involves modeling user interactions with Point-of-interests (PoIs) to analyze user behavior patterns and suggest future scenarios. Data sparsity problems in PoI recommendations can significantly impact the performance of the recommendation model. This paper introduces the Graph Convolutional Network and Multiple Context-Aware PoI Recommendation model (GMCA). First, we present a weighted graph convolutional network that aims to capture the optimal representations of users and PoIs within the user-PoI interaction graph. Second, we employ a fine-grained approach to analyze user check-in records and cluster them into multiple user activity centers. Furthermore, we incorporate time, location, and social context information into the matrix decomposition process. Third, User activity centers are constructed by clustering user check-in records, and the geographical influence of PoI location on user behavioral patterns is explored using probabilistic factor decomposition. The evaluation of the GMCA model on the Yelp and Gowalla datasets shows a significant improvement in Precision@10 indicators. Specifically, there is a 13.85% increase in Precision@10 on the Yelp dataset and a 9.01% increase on the Gowalla dataset. The effectiveness of the GMCA model has been confirmed through numerous experiments conducted on two public datasets.

I. Introduction

As location-based social network (LBSN) applications like Yelp, Gowalla and Uber grow in popularity, users can share check-ins, photos and reviews on these platforms to connect and strengthen existing social relationships with friends. Indeed, PoI recommendation has attracted the attention of many researchers due to the increasing availability of location-based data and the growing popularity of location-based services [1], [2]. Many PoI recommendation models have been proposed in the literature [3], [4], [5] to help recommend locations to users based on their preferences, including categories such as restaurants, museums and entertainment venues.