ConvLab/metalwoz
收藏资源简介:
--- language: - en license: [] multilinguality: - monolingual pretty_name: MetaLWOZ size_categories: - 10K<n<100K task_categories: - conversational --- # Dataset Card for MetaLWOZ - **Repository:** https://www.microsoft.com/en-us/research/project/metalwoz/ - **Paper:** https://www.microsoft.com/en-us/research/publication/results-of-the-multi-domain-task-completion-dialog-challenge/ - **Leaderboard:** None - **Who transforms the dataset:** Qi Zhu(zhuq96 at gmail dot com) To use this dataset, you need to install [ConvLab-3](https://github.com/ConvLab/ConvLab-3) platform first. Then you can load the dataset via: ``` from convlab.util import load_dataset, load_ontology, load_database dataset = load_dataset('metalwoz') ontology = load_ontology('metalwoz') database = load_database('metalwoz') ``` For more usage please refer to [here](https://github.com/ConvLab/ConvLab-3/tree/master/data/unified_datasets). ### Dataset Summary This large dataset was created by crowdsourcing 37,884 goal-oriented dialogs, covering 227 tasks in 47 domains. Domains include bus schedules, apartment search, alarm setting, banking, and event reservation. Each dialog was grounded in a scenario with roles, pairing a person acting as the bot and a person acting as the user. (This is the Wizard of Oz reference—using people behind the curtain who act as the machine). Each pair were given a domain and a task, and instructed to converse for 10 turns to satisfy the user’s queries. For example, if a user asked if a bus stop was operational, the bot would respond that the bus stop had been moved two blocks north, which starts a conversation that addresses the user’s actual need. - **How to get the transformed data from original data:** - Download [metalwoz-v1.zip](https://www.microsoft.com/en-us/download/58389) and [metalwoz-test-v1.zip](https://www.microsoft.com/en-us/download/100639). - Run `python preprocess.py` in the current directory. - **Main changes of the transformation:** - `CITI_INFO`, `HOME_BOT`, `NAME_SUGGESTER`, and `TIME_ZONE` are randomly selected as the valiation domains. - Remove the first utterance by the system since it is "Hello how may I help you?" in most case. - Add goal description according to the original task description: user_role+user_prompt+system_role+system_prompt. - **Annotations:** - domain, goal ### Supported Tasks and Leaderboards RG, User simulator ### Languages English ### Data Splits | split | dialogues | utterances | avg_utt | avg_tokens | avg_domains | cat slot match(state) | cat slot match(goal) | cat slot match(dialogue act) | non-cat slot span(dialogue act) | |------------|-------------|--------------|-----------|--------------|---------------|-------------------------|------------------------|--------------------------------|-----------------------------------| | train | 34261 | 357092 | 10.42 | 7.48 | 1 | - | - | - | - | | validation | 3623 | 37060 | 10.23 | 6.59 | 1 | - | - | - | - | | test | 2319 | 23882 | 10.3 | 7.96 | 1 | - | - | - | - | | all | 40203 | 418034 | 10.4 | 7.43 | 1 | - | - | - | - | 51 domains: ['AGREEMENT_BOT', 'ALARM_SET', 'APARTMENT_FINDER', 'APPOINTMENT_REMINDER', 'AUTO_SORT', 'BANK_BOT', 'BUS_SCHEDULE_BOT', 'CATALOGUE_BOT', 'CHECK_STATUS', 'CITY_INFO', 'CONTACT_MANAGER', 'DECIDER_BOT', 'EDIT_PLAYLIST', 'EVENT_RESERVE', 'GAME_RULES', 'GEOGRAPHY', 'GUINESS_CHECK', 'HOME_BOT', 'HOW_TO_BASIC', 'INSURANCE', 'LIBRARY_REQUEST', 'LOOK_UP_INFO', 'MAKE_RESTAURANT_RESERVATIONS', 'MOVIE_LISTINGS', 'MUSIC_SUGGESTER', 'NAME_SUGGESTER', 'ORDER_PIZZA', 'PET_ADVICE', 'PHONE_PLAN_BOT', 'PHONE_SETTINGS', 'PLAY_TIMES', 'POLICY_BOT', 'PRESENT_IDEAS', 'PROMPT_GENERATOR', 'QUOTE_OF_THE_DAY_BOT', 'RESTAURANT_PICKER', 'SCAM_LOOKUP', 'SHOPPING', 'SKI_BOT', 'SPORTS_INFO', 'STORE_DETAILS', 'TIME_ZONE', 'UPDATE_CALENDAR', 'UPDATE_CONTACT', 'WEATHER_CHECK', 'WEDDING_PLANNER', 'WHAT_IS_IT', 'BOOKING_FLIGHT', 'HOTEL_RESERVE', 'TOURISM', 'VACATION_IDEAS'] - **cat slot match**: how many values of categorical slots are in the possible values of ontology in percentage. - **non-cat slot span**: how many values of non-categorical slots have span annotation in percentage. ### Citation ``` @inproceedings{li2020results, author = {Li, Jinchao and Peng, Baolin and Lee, Sungjin and Gao, Jianfeng and Takanobu, Ryuichi and Zhu, Qi and Minlie Huang and Schulz, Hannes and Atkinson, Adam and Adada, Mahmoud}, title = {Results of the Multi-Domain Task-Completion Dialog Challenge}, booktitle = {Proceedings of the 34th AAAI Conference on Artificial Intelligence, Eighth Dialog System Technology Challenge Workshop}, year = {2020}, month = {February}, url = {https://www.microsoft.com/en-us/research/publication/results-of-the-multi-domain-task-completion-dialog-challenge/}, } ``` ### Licensing Information [Microsoft Research Data License Agreement](https://msropendata-web-api.azurewebsites.net/licenses/2f933be3-284d-500b-7ea3-2aa2fd0f1bb2/view)
--- 语言: - 英语 许可证: - 无 多语言属性: - 单语言 美观名称:MetaLWOZ 规模类别: - 10K < n < 100K 任务类别: - 对话式 --- # MetaLWOZ数据集卡片 - **仓库地址**:https://www.microsoft.com/en-us/research/project/metalwoz/ - **论文地址**:https://www.microsoft.com/en-us/research/publication/results-of-the-multi-domain-task-completion-dialog-challenge/ - **排行榜**:无 - **数据集转换者**:朱琪(邮箱:zhuq96@gmail.com) 如需使用本数据集,请先安装[ConvLab-3](https://github.com/ConvLab/ConvLab-3)平台,随后可通过以下代码加载数据集: from convlab.util import load_dataset, load_ontology, load_database dataset = load_dataset('metalwoz') ontology = load_ontology('metalwoz') database = load_database('metalwoz') 更多使用方法请参阅[此处](https://github.com/ConvLab/ConvLab-3/tree/master/data/unified_datasets)。 ### 数据集概述 本大规模数据集通过众包构建了37884条面向任务的对话,涵盖47个领域的227项任务,涉及公交时刻表查询、公寓搜索、闹钟设置、银行业务、活动预约等多个领域。每段对话均基于带有明确角色分工的场景搭建,由一名扮演机器人(bot)的参与者与一名扮演用户的参与者组成对话双方。(此处采用了绿野仙踪(Wizard of Oz)实验范式——即由幕后人类模拟机器行为)。每组参与者会被分配一个领域与一项任务,并被要求进行至多10轮对话以满足用户的查询需求。例如,若用户询问某公交站是否正常运营,机器人可回复该公交站已向北迁移两个街区,由此展开一段贴合用户实际需求的对话。 ### 如何从原始数据获取转换后的数据集 1. 下载[metalwoz-v1.zip](https://www.microsoft.com/en-us/download/58389)与[metalwoz-test-v1.zip](https://www.microsoft.com/en-us/download/100639) 2. 在当前目录下执行`python preprocess.py` ### 转换过程的主要改动 - 随机选取`CITI_INFO`、`HOME_BOT`、`NAME_SUGGESTER`与`TIME_ZONE`作为验证领域 - 移除系统发出的第一条话语,该语句在多数场景下为“Hello how may I help you?” - 根据原始任务描述添加目标说明:用户角色+用户提示词+系统角色+系统提示词 ### 标注信息 标注字段包含:领域(domain)、目标(goal) ### 支持的任务与排行榜 任务类型:RG、用户模拟器(User simulator) ### 语言 英语 ### 数据划分 | 数据集划分 | 对话数 | 话语总数 | 平均每对话轮数 | 平均每轮Token数 | 平均领域数 | 分类槽匹配(状态) | 分类槽匹配(目标) | 分类槽匹配(对话行为) | 非分类槽跨度(对话行为) | |------------|-------------|--------------|-----------|--------------|---------------|-------------------------|------------------------|--------------------------------|-----------------------------------| | 训练集 | 34261 | 357092 | 10.42 | 7.48 | 1 | - | - | - | - | | 验证集 | 3623 | 37060 | 10.23 | 6.59 | 1 | - | - | - | - | | 测试集 | 2319 | 23882 | 10.3 | 7.96 | 1 | - | - | - | - | | 全部 | 40203 | 418034 | 10.4 | 7.43 | 1 | - | - | - | - | 本数据集共包含51个领域:`['AGREEMENT_BOT', 'ALARM_SET', 'APARTMENT_FINDER', 'APPOINTMENT_REMINDER', 'AUTO_SORT', 'BANK_BOT', 'BUS_SCHEDULE_BOT', 'CATALOGUE_BOT', 'CHECK_STATUS', 'CITY_INFO', 'CONTACT_MANAGER', 'DECIDER_BOT', 'EDIT_PLAYLIST', 'EVENT_RESERVE', 'GAME_RULES', 'GEOGRAPHY', 'GUINESS_CHECK', 'HOME_BOT', 'HOW_TO_BASIC', 'INSURANCE', 'LIBRARY_REQUEST', 'LOOK_UP_INFO', 'MAKE_RESTAURANT_RESERVATIONS', 'MOVIE_LISTINGS', 'MUSIC_SUGGESTER', 'NAME_SUGGESTER', 'ORDER_PIZZA', 'PET_ADVICE', 'PHONE_PLAN_BOT', 'PHONE_SETTINGS', 'PLAY_TIMES', 'POLICY_BOT', 'PRESENT_IDEAS', 'PROMPT_GENERATOR', 'QUOTE_OF_THE_DAY_BOT', 'RESTAURANT_PICKER', 'SCAM_LOOKUP', 'SHOPPING', 'SKI_BOT', 'SPORTS_INFO', 'STORE_DETAILS', 'TIME_ZONE', 'UPDATE_CALENDAR', 'UPDATE_CONTACT', 'WEATHER_CHECK', 'WEDDING_PLANNER', 'WHAT_IS_IT', 'BOOKING_FLIGHT', 'HOTEL_RESERVE', 'TOURISM', 'VACATION_IDEAS']` - **分类槽匹配(cat slot match)**:指分类槽的取值落在本体预定义可选值范围内的百分比。 - **非分类槽跨度(non-cat slot span)**:指带有跨度标注的非分类槽取值占总非分类槽取值的百分比。 ### 引用文献 @inproceedings{li2020results, author = {Li, Jinchao and Peng, Baolin and Lee, Sungjin and Gao, Jianfeng and Takanobu, Ryuichi and Zhu, Qi and Minlie Huang and Schulz, Hannes and Atkinson, Adam and Adada, Mahmoud}, title = {Results of the Multi-Domain Task-Completion Dialog Challenge}, booktitle = {Proceedings of the 34th AAAI Conference on Artificial Intelligence, Eighth Dialog System Technology Challenge Workshop}, year = {2020}, month = {February}, url = {https://www.microsoft.com/en-us/research/publication/results-of-the-multi-domain-task-completion-dialog-challenge/}, } ### 许可信息 [微软研究院数据许可协议](https://msropendata-web-api.azurewebsites.net/licenses/2f933be3-284d-500b-7ea3-2aa2fd0f1bb2/view)
数据集概述
- 名称: MetaLWOZ
- 语言: 英语
- 大小: 10K<n<100K
- 任务类型: 对话式
- 数据集创建方式: 通过众包方式创建,包含37,884个目标导向的对话,覆盖47个领域的227个任务。
数据集内容
- 领域: 包括公交时刻表、公寓搜索、闹钟设置、银行业务、事件预订等47个领域。
- 对话结构: 每个对话基于一个场景,由两个人扮演,一人扮演机器人,另一人扮演用户,进行10轮对话以满足用户查询。
数据处理
- 原始数据转换:
- 下载
metalwoz-v1.zip和metalwoz-test-v1.zip。 - 运行
python preprocess.py进行数据预处理。
- 下载
- 主要变化:
- 随机选择
CITI_INFO,HOME_BOT,NAME_SUGGESTER,TIME_ZONE作为验证域。 - 移除系统的第一个发言,通常为“Hello how may I help you?”。
- 根据原始任务描述添加目标描述。
- 随机选择
数据集分割
| 分割 | 对话数 | 发言数 | 平均发言数 | 平均令牌数 | 平均领域数 |
|---|---|---|---|---|---|
| 训练 | 34261 | 357092 | 10.42 | 7.48 | 1 |
| 验证 | 3623 | 37060 | 10.23 | 6.59 | 1 |
| 测试 | 2319 | 23882 | 10.3 | 7.96 | 1 |
| 全部 | 40203 | 418034 | 10.4 | 7.43 | 1 |
引用信息
@inproceedings{li2020results, author = {Li, Jinchao and Peng, Baolin and Lee, Sungjin and Gao, Jianfeng and Takanobu, Ryuichi and Zhu, Qi and Minlie Huang and Schulz, Hannes and Atkinson, Adam and Adada, Mahmoud}, title = {Results of the Multi-Domain Task-Completion Dialog Challenge}, booktitle = {Proceedings of the 34th AAAI Conference on Artificial Intelligence, Eighth Dialog System Technology Challenge Workshop}, year = {2020}, month = {February}, url = {https://www.microsoft.com/en-us/research/publication/results-of-the-multi-domain-task-completion-dialog-challenge/}, }
许可信息




