Enhancing points of interest recommendation by integrating users’ proximity into the calculation of their similarities

Abstract

In recent years, tourists have increasingly used location-based social networks (LBSNs) to share their travel experiences with friends. Within the context of smart tourism, collaborative filtering (CF) is widely recognized as one of the most commonly used methods for point-of-interest (POI) recommendation systems. This approach analyzes user similarities using measures such Jaccard, or cosine similarity to predict the probabilities of choosing POIs to visit. However, traditional similarity measures fail to account for the physical distances between users and the locations of POIs. To address these limitations, we propose a novel similarity measure called IPUMC (integrating proximity of users in modified cosine similarity). This measure builds on the cosine similarity approach while incorporating geographic proximity between users into the calculation. Experimental results conducted on the Foursquare dataset reveal that IPUMC improves precision by 8.14%, mean average precision (MAP) by 18.01%, and normalized discounted cumulative gain (NDCG) by 16.99% compared to traditional similarity measures, specifically Pearson correlation, Spearman correlation, Euclidean distance, cosine similarity, adjusted cosine and Jaccard.