LLM模拟用户与人类用户在任务导向对话中的平行对话数据集
收藏资源简介:
本研究创建了一个名为“LLM模拟用户与人类用户在任务导向对话中的平行对话数据集”的数据集,包含1856条对话记录,旨在比较大型语言模型模拟的用户与真实人类用户在任务导向对话中的行为差异。数据集涵盖四种代表性行话场景:准备礼物、旅行计划、食谱计划和技能学习计划。该数据集为评估和改进基于LLM的用户模拟框架提供了宝贵资源。
This study constructs a parallel dialogue dataset titled "Parallel Dialogue Dataset Between LLM-Simulated Users and Human Users in Task-Oriented Dialogue". Comprising 1,856 dialogue records, this dataset aims to compare behavioral disparities between large language model (LLM)-simulated users and real human users during task-oriented conversations. It covers four representative scenarios with domain-specific jargon: gift preparation, travel planning, recipe planning, and skill learning planning. This dataset serves as a valuable resource for evaluating and refining LLM-based user simulation frameworks.




