Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Junxiong Tong, Mingjia Yin, Hao Wang, Qiushi Pan, Defu Lian, Enhong Chen

TL;DR

This paper tackles cross-domain recommendation by addressing preference heterogeneity and feature-space disparities with MDAP, a framework that disentangles user preferences through a multi-view encoder and adaptively fuses them via a domain-specific gated decoder. The encoder uses Gumbel-Softmax to produce multiple view-specific embeddings, while domain gates tailor the combination for each domain, promoting effective knowledge transfer and reducing negative transfer. Extensive experiments on three benchmarks show MDAP consistently outperforming single-domain, multi-task, and cross-domain baselines, with ablation studies confirming the critical roles of the Gumbel-Softmax, multi-view encoding, and gated fusion. Overall, MDAP offers a scalable, adaptable approach for cross-domain recommendation that yields more accurate and interpretable cross-domain user representations, with strong implications for handling sparse data and cold-start scenarios.

Abstract

Cross-domain Recommendation systems leverage multi-domain user interactions to improve performance, especially in sparse data or new user scenarios. However, CDR faces challenges such as effectively capturing user preferences and avoiding negative transfer. To address these issues, we propose the Multi-view Disentangled and Adaptive Preference Learning (MDAP) framework. Our MDAP framework uses a multiview encoder to capture diverse user preferences. The framework includes a gated decoder that adaptively combines embeddings from different views to generate a comprehensive user representation. By disentangling representations and allowing adaptive feature selection, our model enhances adaptability and effectiveness. Extensive experiments on benchmark datasets demonstrate that our method significantly outperforms state-of-the-art CDR and single-domain models, providing more accurate recommendations and deeper insights into user behavior across different domains.

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

TL;DR

This paper tackles cross-domain recommendation by addressing preference heterogeneity and feature-space disparities with MDAP, a framework that disentangles user preferences through a multi-view encoder and adaptively fuses them via a domain-specific gated decoder. The encoder uses Gumbel-Softmax to produce multiple view-specific embeddings, while domain gates tailor the combination for each domain, promoting effective knowledge transfer and reducing negative transfer. Extensive experiments on three benchmarks show MDAP consistently outperforming single-domain, multi-task, and cross-domain baselines, with ablation studies confirming the critical roles of the Gumbel-Softmax, multi-view encoding, and gated fusion. Overall, MDAP offers a scalable, adaptable approach for cross-domain recommendation that yields more accurate and interpretable cross-domain user representations, with strong implications for handling sparse data and cold-start scenarios.

Abstract

Cross-domain Recommendation systems leverage multi-domain user interactions to improve performance, especially in sparse data or new user scenarios. However, CDR faces challenges such as effectively capturing user preferences and avoiding negative transfer. To address these issues, we propose the Multi-view Disentangled and Adaptive Preference Learning (MDAP) framework. Our MDAP framework uses a multiview encoder to capture diverse user preferences. The framework includes a gated decoder that adaptively combines embeddings from different views to generate a comprehensive user representation. By disentangling representations and allowing adaptive feature selection, our model enhances adaptability and effectiveness. Extensive experiments on benchmark datasets demonstrate that our method significantly outperforms state-of-the-art CDR and single-domain models, providing more accurate recommendations and deeper insights into user behavior across different domains.
Paper Structure (18 sections, 8 equations, 1 figure, 4 tables, 1 algorithm)

This paper contains 18 sections, 8 equations, 1 figure, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Single Domain Recommendation
  4. Dual-Target Cross-Domain Recommendation
  5. Methodology
  6. Problem Definition
  7. Multi-view Preference Encoder
  8. Domain-specific Gated Decoder
  9. Objective Function
  10. Experiment
  11. Experimental Settings
  12. Datasets
  13. Compared Baselines
  14. Evaluation Protocols
  15. Implementation Details
  16. ...and 3 more sections

Figures (1)

  • Figure 1: The Framework of MDAP: (A) The Multi-view Preference Encoder (orange dashed box) employs Gumbel-Softmax to generate diverse user preference embeddings from the rating matrix. (B) The Domain-specific Gated Decoder (blue dashed box) uses domain-specific gated networks to combine these embeddings based on the domain ID, forming a unified user representation that is then decoded to reconstruct the user-item interaction matrix.