MIntRec2.0 多模态意图识别对话数据集

MIntRec2.0 是清华大学等提出的一个大规模多模态多方基准数据集，专门用于识别对话中的意图和检测非意图内容。相较于先前的 MIntRec，MIntRec2.0 的数据量增至 15K，涵盖 30 种意图类别，并包含约 9.3K 个意图内及 5.7K 个意图外的标注语句，涉及文本、视频和音频等多种模态。

MIntRec2.0 is a large-scale multimodal multi-party benchmark dataset proposed by Tsinghua University and other institutions, which is specifically designed for intent recognition in conversations and detection of out-of-intent content. Compared with the original MIntRec, MIntRec2.0 has expanded its scale to 15K annotated utterances, covering 30 intent categories. Specifically, it includes approximately 9.3K in-intent utterances and 5.7K out-of-intent utterances, involving multiple modalities such as text, video and audio.

AI搜集汇总

背景与挑战

背景概述

MIntRec2.0是一个大规模多模态多方对话数据集，由清华大学等机构提出，专门用于意图识别和非意图内容检测。该数据集包含15K数据量，涵盖30种意图类别，涉及文本、视频和音频多种模态，并引入OOS标签以处理开放世界场景，旨在推动多模态意图理解研究，为更自然的人机交互奠定基础。