MetaLWOz (Meta-Learning Wizard-of-Oz)

这个大型数据集是由众包37,884面向目标的对话框创建的，涵盖了47个域中的227个任务。域包括公交时刻表，公寓搜索，警报设置，银行业务和活动预订。每个对话框都以具有角色的场景为基础，将充当机器人的人和充当用户的人配对。(这是《绿野仙踪》的参考-使用幕后的人充当机器)。每对都被赋予一个域和一个任务，并指示进行10圈交流以满足用户的查询。例如，如果用户询问公交车站是否在运行，则机器人会回答说该公交车站已向北移动了两个街区，这开始了针对用户实际需求的对话。

This large-scale dataset is constructed from 37,884 goal-oriented dialogues collected through crowdsourcing, covering 227 tasks across 47 domains. The included domains are bus schedules, apartment hunting, alarm configuration, banking services, and event booking. Every dialogue is rooted in scenario-based settings with assigned roles, pairing participants acting as the chatbot and those acting as the user. (This is a reference to *The Wizard of Oz*, where human operators working behind the scenes act as the automated system.) Each pair was assigned a designated domain and task, and instructed to complete 10 rounds of dialogue to satisfy the user's query. For example, if the user asks if a certain bus stop is in service, the chatbot will reply that the stop has relocated two blocks north, thus initiating a dialogue aimed at addressing the user's actual needs.

搜集汇总

背景与挑战

背景概述

MetaLWOz是一个大型对话数据集，包含37,884个面向目标的对话，涵盖47个域和227个任务。它通过众包方式创建，采用《绿野仙踪》方法模拟人机交互，适用于AIGC、自然语言处理和对话生成研究。