metadata

language:
  - zh
license:
  - apache-2.0
multilinguality:
  - monolingual
pretty_name: CrossWOZ
size_categories:
  - 1K<n<10K
task_categories:
  - conversational

Dataset Card for CrossWOZ

Repository: https://github.com/thu-coai/CrossWOZ
Paper: https://aclanthology.org/2020.tacl-1.19/
Leaderboard: None
Who transforms the dataset: Qi Zhu(zhuq96 at gmail dot com)

To use this dataset, you need to install ConvLab-3 platform first. Then you can load the dataset via:

from convlab.util import load_dataset, load_ontology, load_database

dataset = load_dataset('crosswoz')
ontology = load_ontology('crosswoz')
database = load_database('crosswoz')

For more usage please refer to here.

Dataset Summary

CrossWOZ is the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. It contains 6K dialogue sessions and 102K utterances for 5 domains, including hotel, restaurant, attraction, metro, and taxi. Moreover, the corpus contains rich annotation of dialogue states and dialogue acts at both user and system sides. We also provide a user simulator and several benchmark models for pipelined taskoriented dialogue systems, which will facilitate researchers to compare and evaluate their models on this corpus.

How to get the transformed data from original data:
- Run python preprocess.py in the current directory. Need ../../crosswoz/ as the original data.
Main changes of the transformation:
- Add simple description for domains, slots, and intents.
- Switch intent&domain of General dialog acts => domain == 'General' and intent in ['thank','bye','greet','welcome']
- Binary dialog acts include: 1) domain == 'General'; 2) intent in ['NoOffer', 'Request', 'Select']; 3) slot in ['酒店设施']
- Categorical dialog acts include: slot in ['酒店类型', '车型', '车牌']
- Non-categorical dialogue acts: others. assert intent in ['Inform', 'Recommend'] and slot != 'none' and value != 'none'
- Transform original user goal to list of {domain: {'inform': {slot: [value, mentioned/not mentioned]}, 'request': {slot: [value, mentioned/not mentioned]}}}, stored as user_state of user turns.
- Transform sys_state_init (first API call of system turns) without selectedResults as belief state in user turns.
- Transform sys_state (last API call of system turns) to db_query with domain states that contain non-empty selectedResults. The selectedResults are saved as db_results (only contain entity name). Both stored in system turns.
Annotations:
- user goal, user state, dialogue acts, state, db query, db results.
- Multiple values in state are separated by spaces, meaning all constraints should be satisfied.

Supported Tasks and Leaderboards

NLU, DST, Policy, NLG, E2E, User simulator

Languages

Chinese

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	5012	84674	16.89	20.55	3.02	99.67	-	100	94.39
validation	500	8458	16.92	20.53	3.04	99.62	-	100	94.36
test	500	8476	16.95	20.51	3.08	99.61	-	100	94.85
all	6012	101608	16.9	20.54	3.03	99.66	-	100	94.43

6 domains: ['景点', '餐馆', '酒店', '地铁', '出租', 'General']

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

@article{zhu2020crosswoz,
  author = {Qi Zhu and Kaili Huang and Zheng Zhang and Xiaoyan Zhu and Minlie Huang},
  title = {Cross{WOZ}: A Large-Scale Chinese Cross-Domain Task-Oriented Dialogue Dataset},
  journal = {Transactions of the Association for Computational Linguistics},
  year = {2020}
}

Licensing Information

Apache License, Version 2.0