README.md

metadata

pretty_name: KLAID
viewer: true
language: ko
multilinguality:
  - monolingual
license: cc-by-nc-nd-4.0
task_categories:
  - text-classification
task_ids:
  - multi-class-classification

Dataset Card for KLAID

Table of Contents

• Table of Contents
• Dataset Description
  • Dataset Summary
  • Supported Tasks and Leaderboards
  • Languages
• Dataset Structure
  • Data Instances
  • Data Fields
  • Data Splits
• Dataset Creation
  • Curation Rationale
  • Source Data
  • Personal and Sensitive Information
• Considerations for Using the Data
  • Other Known Limitations
• Additional Information
  • Other Inquiries
  • Licensing Information
  • Contributions

Dataset Description

• Homepage: https://klaid.net
• Leaderboard: https://klaid.net
• Point of Contact: [email protected]

Dataset Summary

Korean Legal Artificial Intelligence Datasets(KLAID) is a dataset for the development of Korean legal artificial intelligence technology. This time we offer 1 task, which is legal judgment prediction(LJP).

Supported Tasks and Leaderboards

Legal Judgment Prediction(LJP)

Languages

korean

How to use

from datasets import load_dataset
# legal judgment prediction
dataset = load_dataset("lawcompany/KLAID", 'ljp')

Dataset Structure

Data Instances

ljp

An example of 'train' looks as follows.

{
   'fact': '피고인은 2022. 11. 14. 혈중알콜농도 0.123%의 술에 취한 상태로 승용차를 운전하였다.',
   'laws_service': '도로교통법 제148조의2 제3항 제2호,도로교통법 제44조 제1항',
   'laws_service_id': 7
}

Other References You can refer to each label's 'laws service content' here. 'Laws service content' is the statute(source) corresponding to each label.

Data Fields

ljp

• "fact": a string feature
• "laws_service": a string feature
• "laws_service_id": a classification label, with 177 legal judgment values

More Information Needed

Data Splits

ljp

• train: 161,192

Dataset Creation

Curation Rationale

The legal domain is arguably one of the most expertise fields that require expert knowledge to comprehend. Natural language processing requires many aspects, and we focus on the dataset requirements. As a gold standard is necessary for the testing and the training of a neural model, we hope that our dataset release will help the advances in natural language processing in the legal domain, especially for those for the Korean legal system.

Source Data

These are datasets based on Korean legal case data.

Personal and Sensitive Information

Due to the nature of legal case data, personal and sensitive information may be included. Therefore, in order to prevent problems that may occur with personal and sensitive information, we proceeded to de-realize the legal case.

Considerations for Using the Data

Other Known Limitations

We plan to upload more data and update them as some of the court records may be revised from now on, based on the ever-evolving legal system.

Additional Information

Other Inquiries

[email protected]

Licensing Information

Copyright 2022-present Law&Company Co. Ltd.

Licensed under the CC-BY-NC-ND-4.0

Contributions

[More Information Needed]