Smaug: A Chat Model with Agent-Generated Data for Conversational Recommendations
Abstract: Large language models (LLMs) have demonstrated remarkable success in general chat scenarios, delivering coherent and contextually appropriate responses to a wide range of questions. However, current chat models struggle to provide high-quality responses for recommendations, particularly when the recommended items belong to specific domains not covered by common knowledge. In this paper, we propose an efficient method for constructing personalized conversations to fine-tune LLMs for conversational recommendations. Based on this method, we provide a high-quality conversation dataset tailored for the shopping scenario. Using this dataset, we fine-tune a chat model and introduce a chat framework that delivers both high-quality conversations and accurate recommendations. Experimental results show that LLMs fine-tuned on our datasets achieve significant improvements in both recommendation performance and generation quality.
Paper Type: Short
Research Area: Dialogue and Interactive Systems
Research Area Keywords: conversational recommender systems,recommendation,
Contribution Types: Publicly available software and/or pre-trained models, Data resources
Languages Studied: English
Submission Number: 785
Loading