Domain Indexing Collaborative Filtering for Recommender Systems

Abstract

In cross-domain recommendation systems, addressing cold-start items remains a significant challenge. Previous methods typically focus on maximizing performance using cross-domain knowledge, often treating the knowledge transfer process as a black box. However, the recent development of domain indexing introduces a new approach to better address such challenges. We have developed an adversarial Bayesian framework, Domain Indexing Collaborative Filtering (DICF), that infers domain indices during cross-domain recommendation. This framework not only significantly improves the recommendation performance but also provides interpretability for cross-domain knowledge transfer. This is verified by our empirical results on both synthetic and real-world datasets.