Spatial-Aware Deep Recommender System

Location recommendation systems are becoming more and more present in our life. In this paper, we propose a novel recommendation model, S-DEEPREC, that utilizes neural networks to jointly learn the locations preferences of users and incorporate the geographical constraints in their preferences. S-DEEPREC utilizes a feed forward neural network to learn the latent factors of users and locations and the interaction between them. Further, we utilize a feed forward neural network with a softmax layer to incorporate geographical constraints in latent factors of locations. We evaluate S-DEEPREC on one of the largest check-in datasets and compare it with existing factorization based recommendation models. The results show that S-DEEPREC performs 10 times better than state-of-the-art recommendation model.