Knowledge Transfer for Cross-Domain Book Recommender System

Addressing the challenge of sparse data, cross-domain recommender systems offer a promising approach by leveraging knowledge from auxiliary domains to enhance recommendations in a target domain. Many studies have concentrated on situations in which the auxiliary domains overlap with the target domain in terms of shared users or items. However, acquiring such auxiliary data is frequently constrained in real-world settings. This paper explores a more general scenario where the auxiliary domains and the target domain have completely different users and items, a situation that is widely applicable due to the ease of acquiring auxiliary data. Introducing a novel approach that involves Knowledge Transfer with residual learning by integrating a triadic interaction pattern between users, items, and domains with the hidden patterns identified by a residual neural network, user, item, and domain factors can be effectively modeled in a unified space using a modified tri-matrix factorization residual network model. The proposed KTRN approach is evaluated through experiments conducted on two real-world datasets: MovieLens and GoodBooks. The effectiveness of the approach is demonstrated by the results, as it outperforms other cross-domain approaches within the CDRSs framework.