korean-law-articles / README.md
Wookhyeon's picture
auto-update from discovery-engine
e1d3de3 verified
metadata
language:
  - ko
license: cc0-1.0
size_categories:
  - 1K<n<10K
task_categories:
  - text-generation
  - text-classification
  - question-answering
  - summarization
  - feature-extraction
tags:
  - korean
  - law
  - legal
  - statute
  - nlp
  - legal-nlp
  - korean-nlp
pretty_name: Korean Law Articles (Statutes)
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: law_id
      dtype: string
    - name: mst
      dtype: string
    - name: title
      dtype: string
    - name: title_short
      dtype: string
    - name: law_type
      dtype: string
    - name: ministry
      dtype: string
    - name: phone
      dtype: string
    - name: amendment_type
      dtype: string
    - name: promulgation_date
      dtype: string
    - name: promulgation_no
      dtype: string
    - name: effective_date
      dtype: string
    - name: articles
      list:
        - name: number
          dtype: string
        - name: title
          dtype: string
        - name: content
          dtype: string
        - name: effective_date
          dtype: string
    - name: full_text
      dtype: string
    - name: supplementary
      dtype: string
    - name: amendment_reason
      dtype: string
    - name: amendment_text
      dtype: string
    - name: url
      dtype: string
    - name: crawled_at
      dtype: string
  splits:
    - name: train
      num_bytes: 2146091
      num_examples: 102
  download_size: 766954
  dataset_size: 2146091

Korean Law Articles — 한국 현행 법령 본문 전수 데이터셋

법제처 국가법령정보 OPEN API (lawService.do) 로 수집한 대한민국 현행 법령 5,583건의 본문 전수 데이터셋입니다. (전체 5,584건 중 1건은 출처 서버에서 본문 미제공)

🎯 라이브 Q&A 챗봇

Korean Law Q&A (Gradio Space)

본 데이터셋 위에 RAG 챗봇이 가동 중입니다 — 자연어 질문에 답변하고 인용한 법령을 클릭 가능 링크로 제시합니다.

  • Retrieval: sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 시맨틱 + BM25 hybrid
  • LLM: Llama 3.3 70B (Groq, free tier)
  • 인용 강제 + hallucination 방지: 데이터셋에 없는 내용은 "확인되지 않습니다" 반환

규모

항목
총 법령 수 5,583
Parquet 크기 약 54 MB
평균 조문 수 약 8개 / 법령
평균 본문 길이 약 2,200자 / 법령
한국어 (ko) 100%
라이선스 CC0-1.0 + 출처 표기 (공공누리 제1유형)

출처

스키마

필드 타입 설명
law_id string 법령ID
mst string 법령일련번호 (lawService.do MST 파라미터)
title string 법령명 (한글)
title_short string 법령명 약칭
law_type string 법종구분 (법률/대통령령/총리령/부령 등)
ministry string 소관부처
phone string 소관부처 연락처
amendment_type string 제개정구분 (제정/일부개정/전부개정/타법개정)
promulgation_date string 공포일자 (YYYYMMDD)
promulgation_no string 공포번호
effective_date string 시행일자 (YYYYMMDD)
articles list[dict] 조문 목록: number, title, content, effective_date
full_text string 모든 조문을 줄바꿈으로 join 한 본문 (LLM 입력용)
supplementary string 부칙 (평탄화)
amendment_reason string 제개정이유 (평탄화)
amendment_text string 개정문 (평탄화)
url string 법제처 원문 링크
crawled_at string 수집 시각

사용 예

from datasets import load_dataset

ds = load_dataset("Wookhyeon/korean-law-articles", split="train")
print(f"loaded {len(ds)} laws")
print(ds[0]["title"])
print(ds[0]["full_text"][:500])

# 조문 단위 평탄화 (1 row = 1 article)
import itertools
articles = list(itertools.chain.from_iterable(
    [{**a, "law_id": row["law_id"], "law_title": row["title"]} for a in row["articles"]]
    for row in ds
))
print(f"total articles: {len(articles):,}")

활용 사례

  • 한국어 법령 도메인 LLM 사전학습/파인튜닝: 한국어 법률 도메인 특화 모델
  • 법령 검색·임베딩 벤치마크: bge/sentence-transformers 한국어 법률 평가 코퍼스
  • 법령 QA 시스템 평가: 시민용 법률 챗봇의 정답 코퍼스
  • 법률 NER/요약 라벨링 시드: 한국어 법률 NLP 태스크의 base data
  • 소관부처별 분석: 부처별 법령 통계, 시계열 변화 분석

갱신 정책

  • 자동 수집: 6시간마다 GitHub Actions (crawl.yml)
  • HF 동기화: 매주 일요일 02:00 UTC + manual dispatch (upload-hf.yml)
  • 출처 시스템(법제처)에 신규/개정 법령이 등록되면 다음 cron 회차에 자동 반영

알려진 제한

  • 본문 미제공 1건 (MST 200755): 법제처 lawService.do 가 404 반환 (출처 시스템 자체 문제)
  • 본문 외 별표/서식은 미포함 (lawBylSearch.do 별도 API 필요)
  • 영문 법령은 미포함 (elaw target 별도 신청 필요)
  • 판례/헌재결정례/법령해석례는 별도 데이터셋으로 분리 예정

인용

이 데이터셋을 학술 또는 상용으로 활용하시면 출처를 표기해 주세요:

법제처 국가법령정보 OPEN API (https://open.law.go.kr/)
discovery-engine 프로젝트 (https://github.com/Wook-AI/discovery-engine)

변경 로그

  • 2026-04-30: 5,583건 전수 수집 완료 (이전 100건 → 55배 확장) + Gradio Space 데모 공개
  • 2026-04-29: 초기 100건 데이터셋 공개