Datasets:

Dataset Viewer

The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

Dataset Card for JMultiWOZ

Dataset Summary

JMultiWOZ is a large-scale Japanese multi-domain task-oriented dialogue dataset. The dataset is collected using the Wizard-of-Oz (WoZ) methodology, where two human annotators simulate the user and the system. The dataset contains 4,246 dialogues across 6 domains, including restaurant, hotel, attraction, shopping, taxi, and weather. Available annotations include user goal, dialogue state, and utterances.

Supported Tasks

  • Dialogue State Tracking: The dataset can be used to train models for dialogue state tracking, which is the task of predicting the user's belief state at each turn in the dialogue.
  • Dialogue Generation: The dataset can be used to train models for dialogue generation, which is the task of generating a response given the dialogue history.

Languages

The text in the dataset is in Japanese (ja).

Dataset Usage

from datasets import load_dataset

dataset = load_dataset("nu-dialogue/jmultiwoz", trust_remote_code=True)

Dataset Structure

Data Instances

A data instance is a full multi-turn dialogue between a USER and a SYSTEM. Each turn has an utterance:

[
  "福岡へ行くよていなのですが、値段が普通くらいの宿泊施設を探してもらっていいですか?",
  "かしこまりました。ではWITH THE STYLE FUKUOKAはいかがでしょうか。"
]

SYSTEM turn also has a dialogue_state which contains belief_state, book_state, db_result, and book_result:

belief_state:

{
  "domain": ["general", "general", "hotel", ...],
  "slot": ["active_domain", "city", "pricerange", ...],
  "value": ["hotel", "福岡", "普通", ...]
}

book_state:

{
  "domain": ["hotel", "hotel", "hotel", ...],
  "slot": ["people", "day", "stay", ...],
  "value": [None, None, None, ...]
}

db_result:

{
  "candidate_entities": ["WITH THE STYLE FUKUOKA", "ANA クラウンプラザホテル福岡", ...],
  "active_entity": {
    "slot": ["city", "name", "genre", ...],
    "value": ["福岡", "WITH THE STYLE FUKUOKA", "リゾートホテル", ...]
}

Data Fields

Each dialogue instance has the following fields:

  • dialogue_id (int32): A unique identifier for the dialogue.
  • dialogue_name (string): A name for the dialogue.
  • system_name (string): The name of the wizard.
  • user_name (string): The name of the user.
  • goal (sequence): The user's goal for the dialogue.
    • domain (string): The domain of the goal.
    • task (string): The task of the goal.
    • slot (string): The slot of the goal.
    • value (string): The value of the goal.
  • goal_description (sequence): A description of the user's goal.
    • domain (string): The domain of the goal.
    • text (string): The text of the goal.
  • turns (sequence): The turns in the dialogue.
    • turn_id (int32): A unique identifier for the turn.
    • speaker (string): The speaker of the turn.
    • utterance (string): The utterance of the turn.
    • dialogue_state (struct): The dialogue state of the turn.
      • belief_state (sequence): The belief state of the turn.
        • domain (string): The domain of the belief state.
        • slot (string): The slot of the belief state.
        • value (string): The value of the belief state.
      • book_state (sequence): The book state of the turn.
        • domain (string): The domain of the book state.
        • slot (string): The slot of the book state.
        • value (string): The value of the book state.
      • db_result (struct): The database result of the turn.
        • candidate_entities (sequence): The candidate entities of the database result.
          • entity_name (string): The name of the entity.
        • active_entity (sequence): The active entity of the database result.
          • slot (string): The slot of the active entity.
          • value (string): The value of the active entity.
      • book_result (sequence): The book result of the turn.
        • domain (string): The domain of the book result.
        • success (string): The success of the book result.
        • ref (string): The reference of the book result.

Data Splits

The dataset is split into a train, validation, and test split with the following sizes:

train validation test
Number of dialogues 3646 300 300
Number of turns 52,405 4,346 4,435

Citation Information

@inproceedings{ohashi-etal-2024-jmultiwoz,
    title = "JMultiWOZ: A Large-Scale Japanese Multi-Domain Task-Oriented Dialogue Dataset",
    author = "Ohashi, Atsumoto and Hirai, Ryu and Iizuka, Shinya and Higashinaka, Ryuichiro",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation",
    year = "2024",
    url = "",
    pages = "",
}
Downloads last month
85

Models trained or fine-tuned on nu-dialogue/jmultiwoz