POI Recommendation Method of Neural Matrix Factorization Integrating Auxiliary Attribute Information

Point-of-interest (POI) recommendation is the prevalent personalized service in location-based social networks (LBSNs). A single use of matrix factorization (MF) or deep neural cannot effectively capture complex structure user–POI interactions. In addition, to alleviate data-sparsity problem, current methods primarily introduce auxiliary information users and POIs. Auxiliary often judged be equally valued, which will dissipate some valuable information. Hence, we propose a novel POI method fusing attribute based on factorization, integrating convolutional network attention mechanism (NueMF-CAA). First, k-means term frequency–inverse document frequency (TF-IDF) algorithms are used mine POIs from user check-in data problem. an applied learn expression distinguish importance information, respectively. Then, feature vectors concatenated with their respective latent obtain complete Meanwhile, generalized (GMF) multilayer perceptron (MLP) linear nonlinear interactions between POIs, respectively, last hidden layer connected integrate two parts implicit feedback problem make results more interpretable. Experiments real-world datasets, Foursquare dataset Weibo dataset, demonstrate that proposed significantly improves evaluation metrics—hit ratio (HR) normalized discounted cumulative gain (NDCG).