Abstract
Cross-domain recommendation (CDR) has been attracting increasing attention of researchers for its ability to alleviate the data sparsity problem in recommender systems. However, the existing single-target or dual-target CDR methods often suffer from two drawbacks, the assumption of at least one rich domain and the heavy dependence on domain-invariant preference, which are impractical in real world where sparsity is ubiquitous and might degrade the user preference learning. To overcome these issues, we propose a Multi-Sparse-Domain Collaborative Recommendation (MSDCR) model for multi-target cross-domain recommendation. Unlike traditional CDR methods, MSDCR treats the multiple relevant domains as all sparse and can simultaneously improve the recommendation performance in each domain. We propose a Multi-Domain Separation Network (MDSN) and a Gated Aspect Preference Enhancement (GAPE) module for MSDCR to enhance a user's domain-specific aspect preferences in a domain by transferring the complementary aspect preferences in other domains, during which the uniqueness of the domain-specific preference can be preserved through the adversarial training offered by MDSN and the complementarity can be adaptively determined by GAPE. Meanwhile, we propose a Multi-Domain Adaptation Network (MDAN) for MSDCR to capture a user's domain-invariant aspect preference. With the integration of the enhanced domain-specific aspect preference and the domain-invariant aspect preference, MSDCR can reach a comprehensive understanding of a user's preference in each sparse domain. At last, the extensive experiments conducted on real datasets demonstrate the remarkable superiority of MSDCR over the state-of-the-art single-domain recommendation models and CDR models.
Introduction: Beyond the "Rich Domain" Assumption
Cross-domain recommendation (CDR) is a vital strategy for addressing the "data sparsity" problem—the challenge where users interact with only a tiny fraction of available items. Traditionally, CDR has operated on a mentor-student paradigm, where a data-rich "source" domain provides insights to improve recommendations in a data-sparse "target" domain.
However, this approach faces two significant hurdles in real-world applications. First, the "Rich Domain Myth" assumes the existence of a dense auxiliary domain, which is often unavailable. Second, traditional models over-rely on domain-invariant preferences—interests that remain constant across mediums—while ignoring "complementary" preferences. For instance, a user's interest in martial arts movies might strongly suggest an interest in martial arts novels, even if they have no reading history in that genre.
The MSDCR Framework
To address these limitations, we propose the Multi-Sparse-Domain Collaborative Recommendation (MSDCR) model. Unlike previous methods, MSDCR treats all involved domains as sparse and seeks to improve performance across all of them simultaneously through collaborative learning. The architecture is built upon three core pillars designed to capture a comprehensive view of user preferences.
Core Components
- Multi-Domain Separation Network (MDSN): This component ensures that domain-specific preferences are truly unique. Using adversarial training, a discriminator attempts to identify the source domain of a preference embedding, while the encoder works to make these embeddings distinct enough to represent the "uniqueness" of each specific domain.
- Gated Aspect Preference Enhancement (GAPE): Acting as the transfer engine, GAPE uses a gating mechanism to adaptively determine how much information to pull from other domains. It identifies "complementary" signals—preferences observed in one domain that can help infer potential interests in another.
- Multi-Domain Adaptation Network (MDAN): While MDSN focuses on differences, MDAN targets similarities. It employs an adversarial process to extract "domain-invariant" preferences that are indistinguishable across all domains, such as a general preference for "Suspense" or "Fantastic" themes.
Experimental Setup and Results
The model was evaluated using real-world data from Douban, spanning three domains: Movie, Book, and Music. Each domain exhibited high sparsity (ranging from 0.022% to 0.075%). We tested the model across three dual-domain scenarios and one triple-domain scenario (Movie-Book-Music).
Key Findings
- Superior Performance in Sparsity: MSDCR consistently outperformed state-of-the-art baselines like NCF, CoNet, and BiTGCF. In the Movie-Music scenario, it achieved an NDCG@10 of 0.2815 for Movies and 0.3027 for Music, significantly surpassing dual-target CDR competitors.
- Resilience to Data Loss: As training data was manually reduced by up to 80%, the performance gap between MSDCR and other models widened, demonstrating its robustness in extremely data-poor environments.
- The Benefit of Multi-Domain Collaboration: A striking result was that MSDCR performed better in the triple-domain scenario than in dual-domain ones. This suggests that incorporating more sparse domains leads to a more comprehensive and accurate understanding of user preferences.
Conclusion
MSDCR shifts the CDR paradigm from mentorship to collaboration. By treating sparse domains as peers and utilizing adversarial learning to balance domain-specific and domain-invariant aspects, the model effectively transfers complementary knowledge. This approach not only solves the sparsity problem more effectively but also scales better as more domains are integrated into the recommendation ecosystem.
Summary updated