KTCDR-DUP: Knowledge Transfer in Cross-Domain Recommender Systems via Dynamic User Profiles

• Zhengli Zhai,
• Chao Zong,
• Shiya Xu,

Vol. 19, No. 9, September 30, 2025
10.3837/tiis.2025.09.004, Download Paper (Free):

• cross-domain recommendation
• Recommender Systems
• knowledge transfer
• dynamic user profiles
• Probabilistic Graphical Model

Abstract

Cross-Domain Recommendation (CDR) is a widely recognized approach that efficiently addresses sparsity issues in target domain recommendations through leveraging knowledge from additional domains. Nevertheless, effectively leveraging knowledge from other domains remains challenging in recommendation systems. proposes the Knowledge Transfer for CDR using Dynamic User Profiles (KTCDR-DUP) model, which enables knowledge transfer across domains via dynamic user profiles. First, dynamic user profiles are constructed by integrating demographic information, explicit ratings, and content details. Subsequently, calculate the similarity between these profiles. Based on this foundation, utilize a probabilistic graphical framework to estimate the latent elements of users and items spanning different domains by optimizing the posterior likelihood. Predictions for unrated items are derived by computing the dot product of the respective latent factors. Furthermore, in order to verify the superiority of proposed KTCDR-DUP approach, carry out a set of evaluations on cross-domain datasets having different degrees of sparsity. Regarding precision, the experimental outcomes demonstrate that our proposed method surpasses both non-transfer learning and transfer learning-based approaches.

Statistics

Cite this article