Entity Knowledge Transfer-Oriented Dual-Target Cross-Domain Recommendations

Yakun Li,Qiang Wu,Lei Hou,Juanzi Li

DOI: https://doi.org/10.1016/j.eswa.2022.116591

IF: 8.5

2022-01-01

Expert Systems with Applications

Abstract:Cross-domain recommendations can assist users in selecting suitable items in the target domain by aggregating or transferring the abundant available data from the auxiliary domain, which has gradually become a promising research area. However, most existing recommendations are only single-target cross-domain recommendations, and ignore the analysis of entity-side (user side or item side) knowledge, resulting in recommender systems suffering from low accuracy of rating predictions. To address these concerns, we propose a novel Entity Knowledge Transfer-oriented Dual-Target Cross-Domain Recommendations (EKTDCR) to improve the prediction performance of these two domains simultaneously in our paper. Specifically, the latent factor and denoising autoencoder techniques are first utilized to extract and learn entity interaction embedding and entity-side embedding respectively, in separate domains, which are the basis of the proposed EKTDCR approach for dualdomain knowledge transfer and model training. Then, based on the previous entity embeddings, a dualattention mechanism consisting of intra-domain attention and inter-domain attention modules is designed to transfer user preference knowledge to achieve entity feature fusion across domains. In addition, a deep belief network model is adopted to better train the proposed EKTDCR method, and the learnable regularization constraints are further considered to achieve dual-target cross-domain recommendations. Extensive experiments on four real-world datasets demonstrate that our EKTDCR approach can improve the recommendation performance of both domains simultaneously and significantly outperform the state-of-the-art single-domain and crossdomain baseline methods.