SIMMC

该数据集是一组面向任务的对话集，在多模态环境中进行，系统与代理均位于同一虚拟环境内，具体涉及虚拟商店中的时尚和家具领域。数据集被划分为训练集、开发集、开发测试集和标准测试集，分别占比65%、5%、15%和15%。挑战包含四个子任务：消歧、共指解析、对话状态追踪以及响应生成与检索。该数据集规模包含11,244个对话，任务重点在于歧义检测和共指解析。

This dataset is a collection of task-oriented dialogues conducted in a multimodal environment, where both the dialogue system and agents are situated in the same virtual environment, specifically covering the fashion and furniture domains in virtual stores. The dataset is divided into four subsets: training set, development set, development test set, and standard test set, accounting for 65%, 5%, 15%, and 15% of the total dataset respectively. The challenges associated with this dataset include four subtasks: disambiguation, coreference resolution, dialogue state tracking, and response generation and retrieval. The dataset comprises a total of 11,244 dialogues, with the core tasks focusing on ambiguity detection and coreference resolution.

搜集汇总

背景与挑战

背景概述

SIMMC是一个面向任务的多模态对话数据集，专注于虚拟商店中的时尚和家具领域，包含11,244个对话，划分为训练、开发、开发测试和标准测试集。该数据集的核心挑战在于消歧、共指解析、对话状态追踪及响应生成与检索，特别强调歧义检测和共指解析任务。