Meta-Learning Wizard-of-Oz Meta-Learning Multi-Domain Dialogue Dataset is a multi-domain dialogue dataset released by Microsoft Research in 2020, with the related research paper being "Fast Domain Adaptation For Goal-Oriented Dialogue Using A Hybrid Generative-Retrieval Transformer", aimed at developing rapid adaptation methods for dialogue models.

The dataset contains 37,884 two-person dialogues recorded by crowdworkers, covering 47 domains and 227 tasks, with each dialogue consisting of at least 10 turns.

Dataset Composition

Dialogue Data:

• id: Unique identifier for the dialogue
• user_id: Unique identifier for the user
• bot_id: Unique identifier for the user playing the bot
• domain: Identifier for the domain to which the dialogue belongs
• task_id: Identifier for the specific task
• turns: Sequence of alternating bot and user utterances

Task Data:

Defines the specific instructions and roles for each task, divided into training and test sets:

• task_id: Unique identifier for the task
• domain: Domain identifier
• bot_prompt: Specific task instructions assigned to the user playing the bot
• bot_role: Description of the bot's role in the domain
• user_prompt: Specific task instructions assigned to the real user
• user_role: Description of the user's role in the domain

Citation

@InProceedings{shalyminov2020fast,
author = {Shalyminov, Igor and Sordoni, Alessandro and Atkinson, Adam and Schulz, Hannes},
title = {Fast Domain Adaptation For Goal-Oriented Dialogue Using A Hybrid Generative-Retrieval Transformer},
booktitle = {2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
year = {2020},
month = {April},
url = {https://www.microsoft.com/en-us/research/publication/fast-domain-adaptation-for-goal-oriented-dialogue-using-a
-hybrid-generative-retrieval-transformer/},
}