AI summaryⓘ
The authors study how to better predict what items a user might want next when looking at their activity across different areas or categories (domains). They found that existing models using large language models (LLMs) focus too much on word-level details of item descriptions and miss bigger patterns about how items relate, causing problems when mixing information from different domains. To fix this, the authors created DuELRec, which uses two expert models—one focusing on single domains and one on all domains—with a system that smartly combines their advice to avoid confusing signals. They also teach the model to understand item relationships better by comparing items within and across domains. Their experiments show this method works better than many existing approaches.
Cross-Domain Sequential RecommendationLarge Language Models (LLMs)Autoregressive ModelingNegative TransferAttention MechanismContrastive LearningItem-Level RepresentationsDomain GatingCollaborative SignalsSequential Recommendation
Authors
Hyeongjun Yun, Kihyuk Song, Jaegul Choo, Chung Park
Abstract
Cross-Domain Sequential Recommendation (CDSR) predicts the next item a user will interact with based on their historical interaction sequences across multiple domains. Recent approaches leverage Large Language Models (LLMs) finetuned on textual representations of cross-domain user sequences to retrieve the recommended items, referred to as LLMRec. However, LLMRec primarily models the autoregressive patterns of token-level item texts, while overlooking item-level collaborative signals. This semantic misalignment often leads to distorted knowledge transfer across domains-termed negative transfer degrading performance in the CDSR task. To address this issue, we propose a novel LLM-based CDSR model, DuELRec: Domain-Gated Dual Experts with LLMs for Cross-Domain Sequential Recommendation. We propose a domain-gated dual-expert framework, equipped with an item-aware attention transformation module, which aggregates textual subtokens into item-level representations and enforces block-level attention masking. The single-domain expert restricts autoregressive attention to items within the same domain, while the cross-domain expert allows it across all domains. A gating mechanism adaptively fuses their outputs, using single-domain signals to reduce cross-domain noise that causes negative transfer. Second, we introduce a dual-sampling token-to-item contrastive learning objective that allows LLMs to capture the item-level collaborative signals from both single- and cross-domains. This is achieved by transforming token-level item texts into item-level representations and applying stochastic negative sampling from both single- and cross-domain item pools for contrastive learning. Extensive experiments on two real-world datasets across ten domains show that our model outperforms 26 state-of-the-art methods in recommendation performance.