ReGeS: Reciprocal Retrieval-Generation Synergy for Conversational Recommender Systems

2025-09-22

AI Summary

Existing conversational recommendation systems (CRS) face a dilemma in leveraging external knowledge for user preference understanding: domain-specific methods lack generalizability, while purely LLM-based approaches suffer from hallucination; retrieval-augmented generation (RAG) is hindered by dialogue noise and insufficient modeling of fine-grained item distinctions. To address this, we propose a bidirectional retrieval-generation co-optimization framework that establishes a closed-loop synergy between generation-enhanced retrieval (GER) and RAG—enabling joint modeling of dialogue context, external knowledge, and item semantic disparities without additional annotations. Our approach integrates RAG, large language models, and context-aware retrieval. Empirical evaluation on multiple CRS benchmarks demonstrates substantial improvements in recommendation accuracy. Notably, this work is the first to empirically validate, in knowledge-intensive CRS, the dual benefits of bidirectional collaboration: enhanced robustness to dialogue noise and improved precision in user intent parsing.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Learning Preferences or Rankings
• Knowledge Representation and Reasoning: Preferences

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Connecting conversation with external domain knowledge is vital for conversational recommender systems (CRS) to correctly understand user preferences. However, existing solutions either require domain-specific engineering, which limits flexibility, or rely solely on large language models, which increases the risk of hallucination. While Retrieval-Augmented Generation (RAG) holds promise, its naive use in CRS is hindered by noisy dialogues that weaken retrieval and by overlooked nuances among similar items. We propose ReGeS, a reciprocal Retrieval-Generation Synergy framework that unifies generation-augmented retrieval to distill informative user intent from conversations and retrieval-augmented generation to differentiate subtle item features. This synergy obviates the need for extra annotations, reduces hallucinations, and simplifies continuous updates. Experiments on multiple CRS benchmarks show that ReGeS achieves state-of-the-art performance in recommendation accuracy, demonstrating the effectiveness of reciprocal synergy for knowledge-intensive CRS tasks.

Problem

• Enhancing conversational recommender systems with external knowledge integration
• Addressing retrieval noise and item differentiation challenges in CRS
• Reducing hallucination risks without requiring domain-specific engineering

Innovation

• Generation-augmented retrieval distills user intent from conversations
• Retrieval-augmented generation differentiates subtle item features
• Reciprocal synergy framework eliminates extra annotation requirements