Deep Matrix Factorization for Cross-Domain Recommendation
Hailan Kuang, Weiyi Xia, Xiaolin Ma, Xinhua Liu · 2021
Recommender system can make personalized recommendation services for all users. Collaborative filtering is the main method can select suitable items to specific users among the recommendation algorithm. Common CF-based approaches use the users' history behaviors which include explicit information or implicit feedback to make recommendation decisions. Matrix Factorization is the most popular idea in the field of collaborative filtering. Currently, many MF-based methods have been proposed and achieve great improvement. But matrix factorization can only fit linear features which limit the their performance in real-world dataset which contains complex and nonlinear feature, moreover, the sparsity of user-item interaction information is another bottleneck of matrix factorization methods. Recently, Deep learning is widely applied to many fields, there are many research has applied deep learning to recommender system. In this paper, we use multi-layer perceptron structures to learn the representation of users and items in ML-based method. On the other hand, in order to address the data sparsity problem exists in collaborative filtering, Cross-Domain Recommendation is a promising way. Combining with collaborative approach to extract the latent feature, we propose Deep Matrix Factorization for Cross Domain Recommendation (DMF-CDR). We test the proposed method on real-world dataset and show that it outperforms several recent popular models.