Deep Learning For Recommender Systems With Graph-based Structure Information

As the speed of the information increases explosively on the Internet,effective and accurate recommender systems become one of the research hotspots.Introducing consumed items as users’ implicit feedback in matrix factorization(MF)method,SVD++ is one of the most effective collaborative filtering methods for personalized recommender systems.Though powerful,SVD++ has two limitations:(i).only user-side implicit feedback is utilized,whereas item-side implicit feedback,which can also enrich item representations,is not leveraged;(ii).in SVD++,the interacted items are equally weighted when combining the implicit feedback,which can not reflect user’s true preferences accurately.To tackle the above limitations,in this paper we analyze the structure information in user-item bipartite graph used by existing models and propose Graph-based collaborative filtering(GCF)model,Weighted Graph-based collaborative filtering(WGCF)model and Attentive Graph-based collaborative filtering(A-GCF)model.GCF model generalizes the implicit feedback to item side based on the user-item bipartite graph;W-GCF model flexibly learns the weights of individuals in the implicit feedback by matrix factorization and proves the effectiveness of weight mechanism;A-GCF model accomplishes weight mechanism through deep learning methods and further improves the performance of the model.Comprehensive experiments on NetFlix dataset show that our proposed models outperform state-of-the-art models without increasing the model complexity too much.For sparse implicit feedback scenarios,we for the first time leverage the step-two implicit feedback information into the model.By this means we solve the performance loss caused by sparse implicit feedback.