ducut91/legal_mcp

Dataset Description

A comprehensive Korean legal dataset for evaluating RAG systems, containing court cases, statutes, and bar examination questions.

Dataset Summary

Data Type	Count	Description
Court Cases	193,276	Korean court judgments (including 29,730 Constitutional Court decisions)
Statutes (metadata)	5,474	Current Korean statutes
Statute Articles	200,633	Individual statute articles
Benchmark Questions	150	14th Korean Bar Examination (2025) multiple-choice questions

Languages

Korean (ko)

Dataset Structure

Files

legal_mcp/
├── court_cases.csv          # 193,276 court judgments
├── statutes.csv             # 200,633 statute articles
├── statutes_metadata.csv    # 5,474 statute metadata
└── benchmark.csv            # 150 bar exam questions

Court Cases (court_cases.csv)

193,276 Korean court judgments including Supreme Court and Constitutional Court decisions.

Field	Type	Description
case_id	string	Unique identifier
case_number	string	Case number (e.g., "2022헌마884")
case_name	string	Case name
case_type	string	Case type (헌마, 헌바, 민사, 형사, etc.)
court_name	string	Court name (헌법재판소, 대법원, etc.)
decision_date	string	Decision date (YYYYMMDD)
case_content	string	Full judgment text
judgment_summary	string	Judgment summary
issues	string	Issue summary (may be empty)
reference_statute	string	Referenced statutes
judged_statute	string	Adjudicated statutes (Constitutional Court)
reference_case	string	Referenced cases (comma-separated)
reference_case_count	int	Number of referenced cases
case_detail_link	string	Link to official source
token_count	int	Token count

Statutes (statutes.csv)

200,633 individual statute articles from current Korean law.

Field	Type	Description
statute_id	string	Statute unique identifier
law_name	string	Statute name
law_type	string	Statute type (법률, 대통령령, 시행령, etc.)
department	string	Governing department
abbreviation	string	Abbreviation (e.g., 개인정보보호법)
revision_type_name	string	Revision type (전부개정, 일부개정, 타법개정, etc.)
effective_date	string	Effective date (YYYYMMDD)
promulgation_date	string	Promulgation date (YYYYMMDD)
clause_number	string	Article number (e.g., "1", "2의2")
clause_title	string	Article title
clause_content	string	Article content (full text)
clause_count	int	Total number of articles in statute
reference_case_count	int	Number of cases citing this article
law_detail_link	string	Link to official source

Statutes Metadata (statutes_metadata.csv)

5,474 statute-level metadata records.

Field	Type	Description
statute_id	string	Statute unique identifier
law_name	string	Statute name
law_type	string	Statute type
department	string	Governing department
abbreviation	string	Abbreviation
revision_type_name	string	Revision type
effective_date	string	Effective date (YYYYMMDD)
promulgation_date	string	Promulgation date (YYYYMMDD)
clause_count	int	Total number of articles
description	string	Statute description (typically Article 1 - Purpose)
reference_case_count	int	Total number of citing cases
law_detail_link	string	Link to official source
token_count	int	Token count

Benchmark Questions (benchmark.csv)

150 multiple-choice questions from the 14th Korean Bar Examination (2025).

Field	Type	Description
category	string	Legal domain (민사법, 형사법, 공법)
prompt	string	Question text with answer choices
answer	int	Correct answer (1-5)

Category Distribution

Category	Count	Description
민사법 (Civil Law)	70	Civil Act, Civil Procedure, Commercial Act
형사법 (Criminal Law)	40	Criminal Act, Criminal Procedure
공법 (Public Law)	40	Constitutional Law, Administrative Law

Usage

Loading the Dataset

from datasets import load_dataset

# Load all data
dataset = load_dataset("ducut91/legal_mcp")

# Load specific files
cases = load_dataset("ducut91/legal_mcp", data_files="court_cases.csv")
statutes = load_dataset("ducut91/legal_mcp", data_files="statutes.csv")
benchmark = load_dataset("ducut91/legal_mcp", data_files="benchmark.csv")

Example: Benchmark Evaluation

import pandas as pd

benchmark = pd.read_csv("benchmark.csv")

for idx, row in benchmark.iterrows():
    question = row["prompt"]
    correct_answer = row["answer"]
    category = row["category"]

    # Your model inference here
    predicted = model.predict(question)

    is_correct = (predicted == correct_answer)

Dataset Creation

Source Data

• Court Cases: Collected from the Supreme Court of Korea Legal Information System
• Statutes: Collected from the Korea Law Information Center (National Law Information Center)
• Benchmark Questions: 14th Korean Bar Examination (January 2025)

Data Processing

1. Court cases and statutes were collected via official APIs
2. Text was cleaned and normalized
3. Embeddings were generated using the Upstage embedding model (4,096 dimensions)
4. Data was indexed in Elasticsearch for hybrid retrieval (BM25 + semantic search)

License

This dataset is released under CC BY-NC 4.0 (Creative Commons Attribution-NonCommercial 4.0 International).

• Court Cases: Public domain (Korean government public data)
• Statutes: Public domain (Korean government public data)
• Benchmark Questions: Used for academic research purposes

Contact

• Eul Young Lim - ey_lim@o.cnu.ac.kr
• Jihun Park - jihun.park@cnu.ac.kr

Chungnam National University, Daejeon, South Korea