Exploring the Impact of Similarity Measures on Implicit Collaborative Filtering in Point of Interest Recommender Systems

In recent years, Location-Based Social Networks (LBSNs) such as Foursquare, Gowalla, and GeoLife have enabled individuals to share their visited locations with their peers. As LBSNs continue to collect increasing amounts of user checkin data, the challenge of identifying suitable services that align with individual user preferences has intensified. Point of Interest (POI) recommendation systems address this issue by offering personalized suggestions and services tailored to LBSN users. Among the various techniques employed in recommender systems, Collaborative Filtering (CF) is widely used to predict users' next potential POIs. CF employs various similarity measures, including Cosine, Jaccard, Pearson correlation, and Euclidean distance, among others. This study focuses on evaluating the impact of different similarity measures on the performance of POI recommendation systems based on CF. Our experimental analysis, conducted on three real-world dataset from Gowalla, Foursquare, and Yelp, demonstrates that the Jaccard similarity measure is particularly effective for user-based CF POI recommendation systems. The findings indicate that the choice of similarity measure significantly influences the performance of the recommendation system.