SIMMC (Situated and Interactive Multimodal Conversations)

位置交互式多模式对话 (SIMMC) 的任务是除了对话历史记录之外，采取基于共同发展的多模式输入内容的多模式动作。此数据集包含两个SIMMC数据集，在两个购物域上使用多模态Oz向导 (WoZ) 设置，总计约13k人-人对话 (约169K话语): (a) 家具 (基于共享的虚拟环境) 和 (b) 时尚 (基于不断发展的图像集)。The task of Situated Interactive Multi-Modal Conversation (SIMMC) is to perform multi-modal actions based on both dialogue history and co-evolving multi-modal inputs. This dataset comprises two SIMMC datasets, developed using the Multimodal Oz Wizard-of-Oz (WoZ) setup across two shopping domains, totaling approximately 13,000 human-human conversations (around 169,000 utterances): (a) Furniture (based on a shared virtual environment) and (b) Fashion (based on an evolving image collection).

搜集汇总

背景与挑战

背景概述

SIMMC数据集专注于位置交互式多模式对话任务，基于多模态输入和动作。它包含两个购物领域的约13k人-人对话（约169K话语），分别是家具和时尚域，采用多模态Oz向导设置。