Abstract:Cross-domain recommendation (CDR), aiming to extract and transfer knowledge across domains, has attracted wide attention for its efficacy in addressing data sparsity and cold-start problems. Despite significant advances in representation disentanglement to capture diverse user preferences, existing methods usually neglect representation enhancement and lack rigorous decoupling constraints, thereby limiting the transfer of relevant information. To this end, we propose a Unified Framework for Adaptive Representation Enhancement and Inversed Learning in Cross-Domain Recommendation (AREIL). Specifically, we first divide user embeddings into domain-shared and domain-specific components to disentangle mixed user preferences. Then, we incorporate intra-domain and inter-domain information to adaptively enhance the ability of user representations. In particular, we propose a graph convolution module to capture high-order information, and a self-attention module to reveal inter-domain correlations and accomplish adaptive fusion. Next, we adopt domain classifiers and gradient reversal layers to achieve inversed representation learning in a unified framework. Finally, we employ a cross-entropy loss for measuring recommendation performance and jointly optimize the entire framework via multi-task learning. Extensive experiments on multiple datasets validate the substantial improvement in the recommendation performance of AREIL. Moreover, ablation studies and representation visualizations further illustrate the effectiveness of adaptive enhancement and inversed learning in CDR.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are data sparsity and cold - start problems in cross - domain recommendation systems. Specifically, the paper focuses on how to improve the performance of recommendation systems by extracting and transferring knowledge from different domains. Although significant progress has been made in representation disentanglement techniques to capture diverse user preferences, existing methods usually overlook representation enhancement and lack strict decoupling constraints, which limit the effective transfer of relevant information. Therefore, the paper proposes a unified framework - Adaptive Representation Enhancement and Inverse Learning for Cross - Domain Recommendation (AREIL), aiming to solve the following two key problems: 1. **Adaptive Representation Enhancement**: Existing work usually assumes that the shared representations of the same user in different domains are exactly the same, resulting in the inability to capture the differences in embedding quality caused by unbalanced data distribution and inconsistent shared user preferences among different domains. Exploring the representational correlation between domains and adaptively transferring important and generalized knowledge is a nontrivial challenge. 2. **Inverse Representation Learning**: Although some studies use self - supervised methods such as Variational Auto - Encoder (VAE) and contrastive learning to achieve mutual exclusion of disentangled representations, they cannot ensure that domain - shared and domain - specific factors are assigned to the corresponding representations. Ideally, these factors should exhibit a complementary relationship and encode complementary information. Implementing such inverse constraints within a unified framework to learn disentangled user representations remains a key challenge. To address the above challenges, the paper proposes the AREIL framework. The specific steps include: - Initializing item and user representations and decomposing user representations into domain - shared and domain - specific components to disentangle mixed user preferences. - Enhancing user representations through the Adaptive Representation Enhancement Module (AREM), especially using the graph convolution module to capture high - order information, and the self - attention module to reveal inter - domain correlations and complete adaptive fusion. - Introducing a domain classifier and a Gradient Reversal Layer (GRL) to achieve inverse representation learning within a unified framework. The GRL reverses the gradient direction to achieve the inverse constraint objective, and the domain classifier distinguishes the input sources and generates comprehensive supervision signals to constrain the disentangling process. - Finally, optimizing the entire framework using the multi - task learning paradigm in an end - to - end manner. Through extensive experimental verification, the recommendation performance of AREIL on multiple datasets has been significantly improved. Further ablation studies and representation visualizations also prove the effectiveness of adaptive enhancement and inverse learning in cross - domain recommendation.

A Unified Framework for Adaptive Representation Enhancement and Inversed Learning in Cross-Domain Recommendation

Not All Embeddings are Created Equal: Towards Robust Cross-domain Recommendation via Contrastive Learning

Decoupled domain-specific and domain-conditional representation learning for cross-domain recommendation

Beyond the Overlapping Users: Cross-Domain Recommendation via Adaptive Anchor Link Learning

Contrastive Cross-domain Recommendation in Matching

One for All, All for One: Learning and Transferring User Embeddings for Cross-Domain Recommendation

DTCDR: A Framework for Dual-Target Cross-Domain Recommendation

Domain Disentanglement with Interpolative Data Augmentation for Dual-Target Cross-Domain Recommendation

MDAP: A Multi-view Disentangled and Adaptive Preference Learning Framework for Cross-Domain Recommendation

Heterogeneous Graph-based Framework with Disentangled Representations Learning for Multi-target Cross Domain Recommendation

REMIT: Reinforced Multi-Interest Transfer for Cross-Domain Recommendation

Gromov-Wasserstein Guided Representation Learning for Cross-Domain Recommendation.

Towards Robust Cross-Domain Recommendation with Joint Identifiability of User Preference

Towards Knowledge-aware and Deep Reinforced Cross-Domain Recommendation over Collaborative Knowledge Graph

DDCDR: A Disentangle-based Distillation Framework for Cross-Domain Recommendation

Transferring Causal Mechanism over Meta-representations for Target-unknown Cross-domain Recommendation

CDR-Adapter: Learning Adapters to Dig Out More Transferring Ability for Cross-Domain Recommendation Models

Learning Partially Aligned Item Representation for Cross-Domain Sequential Recommendation

Explicitly modeling relationships between domain-specific and domain-invariant interests for cross-domain recommendation

Towards Universal Cross-Domain Recommendation

Prompt-enhanced Federated Content Representation Learning for Cross-domain Recommendation