PIVOT chat dataset
收藏资源简介:
PIVOT聊天数据集是由CyberAgent和东京科学研究所创建的,包含650个关于特定主题的聊天实例,这些聊天实例是由大型语言模型(LLM)与人类之间的互动构成的。该数据集旨在分析和发展能够自然地在用户偏好的话题聊天中主动获取用户信息的技术。每个聊天实例都包含一个预定义的话题和一系列问题,这些问题与话题不直接相关,但需要在聊天中自然地获取答案。数据集的应用领域是对话系统研究,特别是在开发能够实现复杂目标的高级对话策略方面具有重要作用。
The PIVOT Chat Dataset was created by CyberAgent and the Tokyo Science Research Institute, and consists of 650 chat instances centered on specific topics. Each of these instances is composed of interactions between large language models (LLMs) and human participants. This dataset is designed to analyze and advance technologies that can proactively acquire user information during natural conversations on topics preferred by users. Every chat instance includes a predefined topic and a series of questions that are not directly linked to the topic, yet require natural extraction of relevant answers throughout the conversation. The dataset is applied in conversational system research, and plays a critical role particularly in the development of advanced dialogue strategies capable of achieving complex goals.

- 1Proactive User Information Acquisition via Chats on User-Favored TopicsCyberAgent, Institute of Science Tokyo · 2025年



