PEARL
收藏资源简介:
PEARL数据集是由韩国延世大学创建的一个大规模对话推荐数据集,包含超过57,277个对话,覆盖4,000多名用户和9,000多种商品。该数据集通过整合真实世界评论中的用户角色和知识,使用大型语言模型(LLM)增强模拟器,以提高对话中用户偏好的具体性和推荐的相关性。PEARL数据集旨在解决现有对话推荐数据集中用户偏好表达不具体和推荐解释不足的问题,通过模拟具有明确和一致偏好的用户,以及提供基于丰富商品知识的推荐,来提升对话推荐系统的质量和用户体验。
The PEARL dataset is a large-scale conversational recommendation dataset created by Yonsei University in the Republic of Korea. It contains over 57,277 conversations, covering more than 4,000 users and over 9,000 products. This dataset integrates user personas and domain knowledge from real-world reviews, and uses large language models (LLMs) to enhance the simulator, so as to improve the specificity of user preference expressions in conversations and the relevance of recommendations. The PEARL dataset aims to address two key limitations of existing conversational recommendation datasets: the lack of specific user preference expressions and insufficient recommendation explanations. By simulating users with clear and consistent preferences and providing recommendations grounded in rich product knowledge, it seeks to elevate the quality of conversational recommendation systems and enhance user experience.




