MANtIS

MANtIS是由代尔夫特理工大学创建的一个大规模多领域信息寻求对话数据集，包含80,324条对话，覆盖14个不同领域，如物理、旅行和IT等。该数据集从Stack Exchange社区问答门户中提取，旨在支持复杂信息需求的对话系统研究。MANtIS数据集满足了多项理想的数据集特征，如多轮对话、多意图语句、澄清问题等，适用于训练和评估对话搜索系统的能力。

MANtIS is a large-scale multi-domain information-seeking conversation dataset developed by Delft University of Technology. It contains 80,324 conversations spanning 14 distinct domains including physics, travel, IT, and others. Extracted from the Stack Exchange community Q&A portal, this dataset is designed to support research on conversational systems for complex information needs. The MANtIS dataset possesses several desirable characteristics for conversation datasets, such as multi-turn dialogues, multi-intent utterances, clarification questions, and more, making it suitable for training and evaluating the capabilities of conversational search systems.

搜集汇总

背景与挑战

背景概述

MANtIS是一个多领域信息寻求对话数据集，包含超过80,000个来自Stack Exchange社区问答门户的对话，覆盖14个不同领域，如物理、旅行和世界构建，所有对话都基于网页URL提供grounding。该数据集适用于对话响应排名/生成和用户意图预测任务，并提供了手动标注的意图标签，体现了多意图和多样性特点。