Mapping user interest into hyper-spherical space: A novel POI recommendation method. Issue 2 (March 2023)

• Record Type: Journal Article
• Authors: Gan, Mingxin; Ma, Yingxue
• Is Part Of: Information processing & management. Volume 60:Issue 2(2023)
• Publication Date: 2023-03
• Subjects: POI recommendation -- Deep learning -- User preference -- Check-in interest -- Interest model; Information storage and retrieval systems -- Periodicals; Information science -- Periodicals; Systèmes d'information -- Périodiques; Sciences de l'information -- Périodiques; Information science; Information storage and retrieval systems; Periodicals; 658.4038
• DOI: 10.1016/j.ipm.2022.103169
• Languages: English
• ISSNs: 0306-4573

Abstract

Highlights: A hyper-spherical interest model is proposed to improve recommendation quality. High-dimensional spatial model helps recommender well describing users using POIs. Graph representation learning helps recommender system better characterize user personality. A series of experiments verified the effectiveness and robustness of the model. Abstract: Point-of-interest (POI) recommendation helps users quickly filter out irrelevant POI by considering the spatio-temporal factor. In this paper, we address the problem of check-in preference modeling in POI recommendation, and propose a novel POI recommendation method that depicts user preference by constructing unique hypersphere interest model for each user. Different from existing works, we have done three innovative work. (1) We build a check-in graph and adopt DeepWalk algorithm to learn POI embedding, further aggregating them to obtain a hypersphere interest space with an interest center and interest radius. (2) We established a stacked neural network module by a bidirectional LSTM, a self-attention and a memory network, to grasp memory features contained in check-in histories. (3) We proposed a novel candidate POI filter method that updates ranking score by evaluating the Euclidean distance between the vectors of candidate POI and interest center. We evaluate the performance of our method on the four real-world check-in datasets constructed from Foursquare. The comparison between our method and six baselines demonstrates the outstanding performance on various measurements. Compared to the best baseline method, our method achieves about 50% performance improvement on NDCG. In terms of MRR, Precision and Recall, our method achieves about 37%, 21% and 9% performance improvement over the best baseline method. Further ablation experiments verified the importance and effectiveness of the hypersphere interest model, as removing this component caused significant performance degradation.