Context-Aware Disentanglement for Cross-Domain Sequential Recommendation: A Causal View

2026-04-09

AI Summary

Existing cross-domain sequential recommendation methods often introduce spurious correlations due to neglecting contextual shifts, gradient conflicts, and reliance on strong user overlap assumptions, leading to performance imbalance. To address these issues, this work proposes CoDiS, a causality-driven disentangled framework that eliminates confounding effects through variational context adjustment, alleviates gradient conflicts via an expert-isolation selection mechanism, and disentangles domain-shared and domain-specific preferences using a variational adversarial disentanglement module. Notably, CoDiS operates without requiring strong user overlap assumptions, enabling effective cross-domain knowledge transfer. Experimental results on three real-world datasets demonstrate that CoDiS significantly outperforms state-of-the-art methods, achieving higher recommendation accuracy while effectively mitigating the performance seesaw phenomenon.

Abstract

Cross-Domain Sequential Recommendation (CDSR) aims to en-hance recommendation quality by transferring knowledge across domains, offering effective solutions to data sparsity and cold-start issues. However, existing methods face three major limitations: (1) they overlook varying contexts in user interaction sequences, resulting in spurious correlations that obscure the true causal relationships driving user preferences; (2) the learning of domain- shared and domain-specific preferences is hindered by gradient conflicts between domains, leading to a seesaw effect where performance in one domain improves at the expense of the other; (3) most methods rely on the unrealistic assumption of substantial user overlap across domains. To address these issues, we propose CoDiS, a context-aware disentanglement framework grounded in a causal view to accurately disentangle domain-shared and domain-specific preferences. Specifically, Our approach includes a variational context adjustment method to reduce confounding effects of contexts, expert isolation and selection strategies to resolve gradient conflict, and a variational adversarial disentangling module for the thorough disentanglement of domain-shared and domain-specific representations. Extensive experiments on three real-world datasets demonstrate that CoDiS consistently outperforms state-of-the-art CDSR baselines with statistical significance. Code is available at:https://anonymous.4open.science/r/CoDiS-6FA0.

Problem

• Cross-Domain Sequential Recommendation
• Context Awareness
• Causal Disentanglement
• Gradient Conflict
• Data Sparsity

Innovation

• context-aware disentanglement
• causal inference
• cross-domain sequential recommendation
• gradient conflict resolution
• variational adversarial learning