Jaeuk-Han/korean-traditional-liquor-dataset

Dataset Card for Korean Traditional Liquor Corpus (886–1947)

Dataset Summary

A corpus of Korean traditional liquors and brewing culture curated from 132 historical texts written between 886 and 1947. It includes paragraph-level chunks for IR/RAG and document-/process-level metadata for cultural and historical analyses.

• Documents: 132 historical texts (886–1947)
• Entries: 3,519 total
  • 🍶 Traditional liquors: 2,876
  • 🍞 Fermentation starters (Nuruk): 193
  • 📦 Others (brewing secrets, utensils, drinking culture, etc.): 450

Files

• sool_chunks.jsonl — paragraph-level chunks (JSON Lines)
• sool_docs.csv / sool_docs.parquet — document-level metadata
• sool_steps.csv / sool_steps.parquet — brewing steps / process-level data (optional)

CSV and Parquet mirror the same schema within each pair.

---

Supported Tasks and Use Cases

• Text Retrieval / RAG: build BM25 / dense indexes on sool_chunks
• Question Answering & Summarization: cite sool_docs and step metadata
• Classification / Taxonomy: use category fields and derived stats
• Digital Humanities: study historical brewing practices and terminology

---

Languages

• Korean (ko) with historical forms and occasional Hanja.

---

Units & Terminology (historical)

Term	Description
Doi (되)	Traditional volume; ≈ 1/10 Mal, ~1.8 L (estimated).
Mal (말)	Traditional volume; ~18 L.
Nuruk (누룩)	Grain-based enzymatic/microbial starter used for fermentation.
Hot Mixing (온혼)	Mashing/processing with hot water.
Cold Mixing (냉혼)	Mashing/processing with cold water.

---

Dataset Structure

Files and Schemas

1) sool_chunks.jsonl

Each line is one JSON object (a chunk).

field	type	description
chunk_id	string	Chunk ID (unique)
parent_doc_id	string	Parent document ID (join with sool_docs.doc_id)
title	string	Title
section	string	Section/subheading
text	string	Chunk text
metadata	object	Additional metadata (JSON)

Example

{"chunk_id":"c_000001","parent_doc_id":"d_001","title":"Gyeongmyeon Nokpaju","section":"Brewing","text":"...chunk text...","metadata":{"page":12,"source":"산림경제"}}

---

2) sool_steps.csv / sool_steps.parquet (optional)

field	type	description
doc_id	string	Document ID
book	string	Source book
title	string	Document title
liq	string	Liquor name/type
section	string	Section name
start_day	int	Start day
ferment_days	int	Fermentation days
rice_doi	float	Rice (Doi)
rice_mal	float	Rice (Mal)
water_doi	float	Water (Doi)
water_mal	float	Water (Mal)
water_L	float	Water (Liter)
nuruk_doi	float	Nuruk (Doi)
nuruk_mal	float	Nuruk (Mal)
processing	string	Processing method
nuruk_form	string	Nuruk form
sprinkling_water	bool	Sprinkling water flag
hot_mixing	bool	Hot mixing flag
cold_mixing	bool	Cold mixing flag
notes	string	Notes
has_glutinous	bool	Glutinous rice present
has_non_glutinous	bool	Non-glutinous rice present

---

3) sool_docs.csv / sool_docs.parquet

field	type	description
doc_id	string	Document ID
book	string	Source book
title	string	Title
liq	string	Liquor name/type
hanja	string	Hanja (classical Chinese) form
description	string	Description
main_category	string	Main category
main_category_en	string	Main category (EN)
entry_type	string	Entry type
level1 ~ level4	string	Category levels
level1_en ~ level4_en	string	Category levels (EN)
source_categories	string	Source categories
source_categories_en	string	Source categories (EN)
year_guess	int	Estimated year
rice_total_doi	float	Total rice (Doi)
rice_total_mal	float	Total rice (Mal)
water_total_doi	float	Total water (Doi)
water_total_mal	float	Total water (Mal)
water_total_L	float	Total water (Liter)
nuruk_total_doi	float	Total nuruk (Doi)
nuruk_per_mal	float	Nuruk per Mal ratio
num_stages	int	Number of stages
is_multi_stage	bool	Multi-stage brewing flag
filter_on_or_after_day	int	Filtering day (estimated)
yield_mal	float	Yield (Mal)
yield_L	float	Yield (Liter)
water_per_rice_mal	float	Water per rice Mal
yield_per_rice_mal	float	Yield per rice Mal
uses_nuruk	bool	Uses nuruk
hot_mixing	bool	Hot mixing flag
cold_mixing	bool	Cold mixing flag
has_glutinous_rice	bool	Glutinous rice present
has_non_glutinous_rice	bool	Non-glutinous rice present
stage_names	string	Stage names
unique_nuruk_forms	string	Unique nuruk forms
text_len_chars	int	Text length (chars)
text_len_words	int	Text length (words)
title_len_chars	int	Title length (chars)
has_hanja	bool	Hanja present
has_digits	bool	Digits present
has_original	bool	Has original text
has_translation	bool	Has translation
created_at	string	Creation timestamp

---

Loading the Dataset

With datasets

from datasets import load_dataset

ds = load_dataset(
    "Jaeuk-Han/korean-traditional-liquor-dataset",
    data_files={
        "chunks": "sool_chunks.jsonl",
        "docs": ["sool_docs.parquet", "sool_docs.csv"],
        "steps": ["sool_steps.parquet", "sool_steps.csv"],  # optional
    }
)

chunks = ds["chunks"]
docs = ds.get("docs")
steps = ds.get("steps")
chunks[0]

With huggingface_hub + pandas

from huggingface_hub import hf_hub_download
import pandas as pd, json

REPO_ID = "Jaeuk-Han/korean-traditional-liquor-dataset"

# docs (parquet preferred)
p_docs = hf_hub_download(REPO_ID, "sool_docs.parquet")
df_docs = pd.read_parquet(p_docs)

# steps (csv)
p_steps = hf_hub_download(REPO_ID, "sool_steps.csv")
df_steps = pd.read_csv(p_steps)

# chunks (jsonl)
p_chunks = hf_hub_download(REPO_ID, "sool_chunks.jsonl")
with open(p_chunks, "r", encoding="utf-8") as f:
    chunks = [json.loads(line) for line in f]

Minimal Retrieval Example (toy TF-IDF)

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity

texts = [c["text"] for c in chunks]
vec = TfidfVectorizer(min_df=2).fit(texts)
M = vec.transform(texts)

def search(q, topk=5):
    sims = cosine_similarity(vec.transform([q]), M).ravel()
    idx = sims.argsort()[::-1][:topk]
    return [chunks[i] for i in idx]

search("How is Gyeongmyeon Nokpaju brewed?", 5)

---

Ethical Considerations & Limitations

• Reflects the historical context of its sources; some practices/descriptions may be culturally sensitive or misaligned with modern safety/ethical standards.
• Provided as a record; it does not endorse any harmful practice.
• For commercial use, confirm rights with the original sources and applicable policies.

---

Source

Originally provided by the Korean Traditional Liquor DB, maintained by the Korean Traditional Liquor Research Institute / 한국술문헌연구소.
Data source: http://www.koreansool.kr/ktw/php/home.php
Blog: https://blog.naver.com/korean-sool

---

Changes Made

Structured texts into JSONL/CSV/Parquet, normalized historical units (Mal/Doi → L), added English category labels, and derived per-document/step statistics.

---

🔁 Reproducibility & Code

• Preprocessing repo: https://github.com/Jaeuk-Han/korean-traditional-liquor-dataset
• Version pin: use a tagged release (e.g., v1.0.0) or a specific commit SHA (recommended).
• Environment (example):

  python -V          # 3.10+ (3.11 recommended)
  pip install -r requirements.txt
  

• Recreate (template):

  git clone https://github.com/Jaeuk-Han/korean-traditional-liquor-dataset
  cd korean-traditional-liquor-dataset
  git checkout <commit-sha-or-tag>
  
  # adjust paths/args to your environment
  python scripts/preprocess.py \
    --input ./raw_sources \
    --out_chunks ./out/sool_chunks.jsonl \
    --out_docs ./out/sool_docs.parquet \
    --out_steps ./out/sool_steps.parquet
  

Checksums (for this release)

file	size	sha256
sool_chunks.jsonl	14.39 MB	56dc4c10e0945149ee62365551c0cc80a0d185426921352643443e030084253f
sool_docs.csv	1.70 MB	86444b22c156f14dc038747004fa026f487ee6618ed7b8f6d2d845fe394bb6bc
sool_docs.parquet	0.34 MB	bdcfbeb4e983b3c2614778f8f5dbb6a3adc343091b4f66478eb1ddd2ec83e484
sool_steps.csv	1.81 MB	3b3a9e01b7bcc42895ea69e576f4f030fa7ae9aa08c838f8e7f4f1c0bfc27b58
sool_steps.parquet	0.75 MB	0f6d0fd16ef5195ae3784d454c37cfad079f7d320d4b66ed36b4640575fa6b1b

---

Licensing Information

• License: CC BY-NC-SA 2.0 (Korea, KR port)
  Canonical deed: https://creativecommons.org/licenses/by-nc-sa/2.0/kr/
  You may share/adapt non-commercially, with attribution, and you must share alike under the same license (or a later BY-NC-SA version).
• If you prefer to apply a later version for your adaptation, you may use CC BY-NC-SA 4.0 International (as permitted by ShareAlike for 2.0/2.5/3.0 materials); ensure no additional restrictions are added.

---

Citation

If you use this dataset, please cite:

Han, Jaeuk. (2025). Korean Traditional Liquor Corpus (886–1947). Hugging Face Datasets. https://huggingface.co/datasets/Jaeuk-Han/korean-traditional-liquor-dataset

---

Changelog

• v1.0.0 — Initial release (sool_chunks.jsonl, sool_docs.(csv|parquet), sool_steps.(csv|parquet))

---

Dataset Card Authors

• Jaeuk Han (maintainer)

Sources — Data was sourced from the Korean Sool Historical DB / 한국술 고문헌 DB, maintained by the Korean Traditional Liquor Research Institute / 한국술문헌연구소. The archive covers historical documents dated 886–1947, including texts on liquor, fermentation starters, brewing methods, and cultural practices. Blog: https://blog.naver.com/korean-sool

---

Contact / Takedown
Questions or takedown requests: [email protected].