Matrix Factorization-Based Unify Multiple Interactions for Cross-Domain Recommendation Services
Luong Vuong Nguyen, Tri‐Hai Nguyen, Ho-Trong-Nguyen Pham, Thi-Thu-Hong Phan · 2023
The interactions of users in many different domains on e-commercial sites lead to research in cross-domain recommendation systems (CDRS) considered an emerging topic. Such studies aim to use the information from auxiliary source domains to increase the accuracy of recommendations in the target domain. The recommendations problem, such as recommending movies or genres, can be viewed as a distinct task or, more broadly, a particular domain. Furthermore, explicit interactions may be unavailable for such applications, whereas implicit interactions are readily available. This paper uses the implicit interactions as the knowledge during the transfer between the target and the source domains. Our approach is based on a regularized latent factor model in cross-domain collaborative filtering. The novelty of our process is that we control the transferred knowledge through the regularization parameters to prevent negative transfer. The experiments deployed in two cross-domain datasets collected from Amazon demonstrate that the proposed method performs better in generating recommendations than other methods.