From Reviews to Dialogues: Active Synthesis for Zero-Shot LLM-based Conversational Recommender System

2025-04-21

AI Summary

Conversational recommendation systems (CRS) face significant challenges in low-resource settings, including scarcity of domain-specific dialogue data, high annotation costs, and privacy constraints. To address these, we propose the first active learning–driven dialogue synthesis framework tailored for black-box large language models (LLMs), requiring neither fine-tuning nor access to internal LLM parameters. Leveraging prompt engineering, our method integrates heterogeneous non-dialogue data—such as item metadata, user reviews, and collaborative signals—to enable high-informativeness seed selection and semantic-consistent, structured dialogue generation. Empirically, the approach substantially improves zero-shot LLM-based CRS performance in sparse-data regimes, enhancing both recommendation accuracy and dialogue coherence. Moreover, it enables lightweight supervised models trained on synthesized data to approach the performance of fully supervised baselines. This work constitutes the first demonstration of high-quality CRS construction without any human-annotated dialogue data.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Conversational AI/Dialog Systems

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Conversational recommender systems (CRS) typically require extensive domain-specific conversational datasets, yet high costs, privacy concerns, and data-collection challenges severely limit their availability. Although Large Language Models (LLMs) demonstrate strong zero-shot recommendation capabilities, practical applications often favor smaller, internally managed recommender models due to scalability, interpretability, and data privacy constraints, especially in sensitive or rapidly evolving domains. However, training these smaller models effectively still demands substantial domain-specific conversational data, which remains challenging to obtain. To address these limitations, we propose an active data augmentation framework that synthesizes conversational training data by leveraging black-box LLMs guided by active learning techniques. Specifically, our method utilizes publicly available non-conversational domain data, including item metadata, user reviews, and collaborative signals, as seed inputs. By employing active learning strategies to select the most informative seed samples, our approach efficiently guides LLMs to generate synthetic, semantically coherent conversational interactions tailored explicitly to the target domain. Extensive experiments validate that conversational data generated by our proposed framework significantly improves the performance of LLM-based CRS models, effectively addressing the challenges of building CRS in no- or low-resource scenarios.

Problem

• Addressing lack of domain-specific conversational data for CRS
• Enabling zero-shot LLM-based CRS with scalable data synthesis
• Improving small CRS models via active learning-guided LLM augmentation

Innovation

• Active learning guides LLM data synthesis
• Leverages non-conversational data as seeds
• Generates domain-specific synthetic dialogues