Refining Text Generation for Realistic Conversational Recommendation via Direct Preference Optimization

2025-08-27

AI Summary

Current conversational recommendation systems (CRS) suffer from unnatural interaction patterns—prioritizing rapid recommendations in minimal turns at the expense of human-like dialogue rhythm and logical information progression. To address this, we propose a naturalness-oriented CRS optimization framework. Our method jointly models dialogue history and item descriptions using large language models to generate fine-grained dialogue summaries and preference-aware recommendation texts. Additionally, we incorporate direct preference optimization (DPO) to explicitly align generated responses with critical recommendation signals, thereby enhancing coherence and relevance. Evaluated on the Redial and TG-ReDial benchmarks, our approach achieves significant improvements: +12.3% in human-rated dialogue naturalness and +5.8% in Recall@10. The implementation is publicly available.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Natural Language Processing: Conversational AI/Dialog Systems
• Machine Learning: Learning Preferences or Rankings

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economics and fairness of platforms and recommendation systems

Abstract

Conversational Recommender Systems (CRSs) aim to elicit user preferences via natural dialogue to provide suitable item recommendations. However, current CRSs often deviate from realistic human interactions by rapidly recommending items in brief sessions. This work addresses this gap by leveraging Large Language Models (LLMs) to generate dialogue summaries from dialogue history and item recommendation information from item description. This approach enables the extraction of both explicit user statements and implicit preferences inferred from the dialogue context. We introduce a method using Direct Preference Optimization (DPO) to ensure dialogue summary and item recommendation information are rich in information crucial for effective recommendations. Experiments on two public datasets validate our method's effectiveness in fostering more natural and realistic conversational recommendation processes.Our implementation is publicly available at:https://github.com/UEC-InabaLab/Refining-LLM-Text

Problem

• Generates conversational summaries from dialogue history
• Infers implicit user preferences from dialogue context
• Optimizes information richness for effective recommendations

Innovation

• Using Direct Preference Optimization for refinement
• Generating dialogue summaries from conversation history
• Extracting explicit and implicit user preferences