swap-uniba/DistillRecDial

DistillRecDial是一个会话推荐系统数据集，通过知识蒸馏管道利用较小的开源语言模型生成，能够模拟不同用户类型的行为多样性，包括他们的意图、偏好表达方式和主动性。这个数据集在对话连贯性和领域特定专业知识方面表现优异，有助于推动更真实和有效的会话推荐系统的发展。

DistillRecDial is a conversational recommendation system dataset generated through a knowledge distillation pipeline using smaller open language models. It simulates the behavioral diversity of different user types, including their intentions, preference expression styles, and levels of initiative. The dataset excels in terms of dialogue coherence and domain-specific expertise, contributing to the development of more realistic and effective conversational recommendation systems.