Abstract:Discovering user preferences across different domains is pivotal in cross-domain recommendation systems, particularly when platforms lack comprehensive user-item interactive data. The limited presence of shared users often hampers the effective modeling of common preferences. While leveraging shared items' attributes, such as category and popularity, can enhance cross-domain recommendation performance, the scarcity of shared items between domains has limited research in this area. To address this, we propose a Coherence-guided Preference Disentanglement (CoPD) method aimed at improving cross-domain recommendation by i) explicitly extracting shared item attributes to guide the learning of shared user preferences and ii) disentangling these preferences to identify specific user interests transferred between domains. CoPD introduces coherence constraints on item embeddings of shared and specific domains, aiding in extracting shared attributes. Moreover, it utilizes these attributes to guide the disentanglement of user preferences into separate embeddings for interest and conformity through a popularity-weighted loss. Experiments conducted on real-world datasets demonstrate the superior performance of our proposed CoPD over existing competitive baselines, highlighting its effectiveness in enhancing cross-domain recommendation performance.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively extract and utilize shared item attributes to guide user preference disentanglement in cross - domain recommendation systems so as to improve recommendation performance. Specifically, when the user - item interaction data between different platforms is limited, this will hinder the effective modeling of common preferences. Although using the attributes of shared items (such as category and popularity) can enhance cross - domain recommendation performance, research in this area has been limited due to the scarcity of shared items between different domains. To solve these problems, the authors propose a Coherence - guided Preference Disentanglement (CoPD) method, aiming to improve cross - domain recommendation in the following two aspects: 1. **Explicit extraction of shared item attributes**: Guide the learning of shared user preferences by explicitly extracting shared item attributes. 2. **Preference disentanglement**: Disentangle these preferences to identify specific user interests transferred between different domains. ### Specific problem description - **Data sparsity problem in cross - domain recommendation**: Due to the limited number of shared users between different domains, the performance of cross - domain recommendation systems is restricted. - **Insufficient utilization of shared item attributes**: Although the attributes of shared items (such as category and popularity) can improve recommendation performance, existing methods are insufficient in using these attributes. - **Challenges of user preference disentanglement**: User behavior may be influenced by herd mentality, not just real interests. Therefore, distinguishing users' true interests from herd behavior is a challenge. ### Solutions - **Coherence - guided item attribute alignment**: Promote the spread of feature information between two domains by constructing a global graph connecting the two domains, and use overlapping users as connection points. At the same time, apply coherence constraints so that shared item embeddings can capture shared attributes while filtering out unnecessary information. - **User preference disentanglement**: Disentangle shared user embeddings into interest embeddings and herd embeddings, and use a popularity - weighted loss function to train interest and herd encoders. By establishing a positive correlation between herd preference and popularity, and a negative correlation between interest preference and popularity, effective preference disentanglement is achieved. - **Embedding fusion and recommendation**: Combine shared information and domain - specific information to generate comprehensive user and item embeddings for final recommendation. ### Main contributions - Propose a new Coherence - guided Preference Disentanglement framework (CoPD), aiming to improve the performance of cross - domain recommendation by extracting shared item attributes and applying preference disentanglement. - Design a global - local graph framework, which extracts shared item attributes while retaining important domain - specific information and removing redundant information by adding coherence constraints on embeddings in specific domains and shared domains (the union of the two domains). - Achieve effective disentanglement of user preferences by constructing a popularity - weighted loss function, and use comprehensive item embeddings to guide the preference disentanglement process. - Verify the effectiveness of the proposed method through experiments, and the results show that CoPD outperforms existing competing methods on multiple benchmark datasets. In conclusion, this paper solves the challenges of data sparsity and user preference disentanglement in cross - domain recommendation systems by introducing the Coherence - guided Preference Disentanglement method, and improves the performance of the recommendation system.

Coherence-guided Preference Disentanglement for Cross-domain Recommendations

User Distribution Mapping Modelling with Collaborative Filtering for Cross Domain Recommendation

Co-training Disentangled Domain Adaptation Network for Leveraging Popularity Bias in Recommenders

Cross-domain recommendation via user interest alignment

DisCo: Towards Harmonious Disentanglement and Collaboration between Tabular and Semantic Space for Recommendation

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

Multi-level Attention-based Domain Disentanglement for BCDR

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

Disentangling User Interest and Conformity for Recommendation with Causal Embedding

Preference or Intent? Double Disentangled Collaborative Filtering

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

Curriculum Co-disentangled Representation Learning Across Multiple Environments for Social Recommendation.

Disentangling ID and Modality Effects for Session-based Recommendation

Cross-domain Transfer of Valence Preferences via a Meta-optimization Approach

Unveiling Hidden Implicit Similarities for Cross-Domain Recommendation

Deep Transfer Collaborative Filtering with Geometric Structure Preservation for Cross-Domain Recommendation

Heterogeneous and Clustering-Enhanced Personalized Preference Transfer for Cross-Domain Recommendation

DDTCDR: Deep Dual Transfer Cross Domain Recommendation

Learning Accurate and Bidirectional Transformation Via Dynamic Embedding Transportation for Cross-Domain Recommendation

Disentangled Hypergraph Collaborative Filtering for Social Recommendation