Discrete Ricci Flow: A powerful method for community detection in location-based social networks
Abstract
Community detection is crucial to understanding behavioral patterns in location-based social networks (LBSNs) where user locations, media, and check-ins are involved. This hierarchical structure enables the formation of user communities, where a community represents a group of users sharing similar interests. In addition, selecting an appropriate community for a recommendation scenario is crucial and challenging. To address these issues, in this article, we propose a novel method to link LBSNs to the Discrete Ricci Flow (DRF) community detection algorithm. Then we use the communities formed by the Ricci curvature of the network to provide recommendations in a user-based collaborative filtering (CF) recommender system. Our evaluation method considers spatial–temporal features and user relationships. The evaluation encompasses unsupervised and supervised learning methodologies, employing the modularity evaluation index and the CF recommender system. Comparative analysis against traditional community detection algorithms, including Leiden, Infomap, Walktrap, and Fast Greedy, reveals the superior performance of our proposed method, as it achieves an impressive 0.5075% and 0.8486% modularity scores for Gowalla and Brightkite respectively that indicates the efficacy of the method in capturing the inherent structure of the data. Furthermore, when integrated into the CF recommender system, the proposed method based on DRF demonstrates superior performance compared to other community detection methods for different data sets such as Gowalla and Brightkite. In particular, for Gowalla it improves the performance of the Point Of Interest (POI) recommendation system by an average of 10.92% and 8.02% in Recall@15 and Recall@20, respectively.
Published on: Computers and Electrical Engineering , Q1 , IF: 4.9
Hyperbolic Neural Networks Outperform Euclidean Models in Spatiotemporal Link Prediction for Location-Based Social Networks
Abstract
Link prediction is a fundamental task for understanding user interactions in location-based social networks (LBSNs), where check-ins and spatial media data exhibit hierarchical structures. This paper proposes a hyperbolic neural network model that integrates spatiotemporal user features with friendship relations and compares it to a classical multi-layer perceptron (MLP) in Euclidean space. Experiments on the Brightkite and Gowalla datasets show that the hyperbolic model achieves substantial gains over the MLP, with up to 13.1% higher F1-score and 23.8% higher Recall for class 1 on Brightkite, and 7.2% and 9.9% improvements on Gowalla. Moreover, the hyperbolic network demonstrates faster convergence to optimal configurations in largescale settings despite higher per iteration complexity. These results highlight the advantage of hyperbolic geometry for modeling hierarchical relationships in LBSNs.
Under review