Aiming at the Target: Filter Collaborative Information for Cross-Domain Recommendation

As recommender systems become pervasive in various scenarios, cross-domain recommenders (CDR) are proposed to enhance the performance of one target domain by data from other related source domains. However, irrelevant information from the source domain may instead degrade target domain performance, which is known as the negative transfer problem. There have been some attempts to address this problem, mostly by designing adaptive representations for overlapped users. Whereas, these methods rely on the learned representations of the model, lacking explicit constraint to filter irrelevant source-domain collaborative information for the target domain, which limits their cross-domain transfer capability.

In this paper, we propose a novel Collaborative information regularized User Transformation (CUT) framework to tackle the negative transfer problem by directly filtering users' collaborative information. In CUT, target domain user similarity is adopted as a constraint for user transformation to filter user collaborative information from the source domain. First, CUT learns user similarity relationships from the target domain. Then, source-target information transfer is guided by the user similarity, where we design a user transformation layer to learn target-domain user representations and a contrastive loss to supervise the user collaborative information transferring. As a flexible and lightweight framework, CUT can be applied with various single-domain recommender systems as the backbone and extend them to multi-domain tasks. We conduct extensive experiments of CUT with two single-domain backbone recommenders on six CDR tasks from two real-world datasets. The results show significant performance improvement of CUT compared with SOTA single and cross-domain methods. Further analysis illustrates that CUT can effectively alleviate the negative transfer problem.