lablup/tariff_trade_domain.synthetic_trade_qa_kr
Korean Trade Domain QA Dataset A Korean-language question-answering dataset for the international trade domain, combining 21,399 QA pairs from three complementary sources: official 무역영어 1급 certification exam questions, trade terminology definitions, and lecture-derived QA pairs. Designed for fine-tuning and evaluating LLMs on Korean trade domain knowledge. Dataset Description The dataset covers vocabulary, regulations, procedures, and concepts from Korean… See the full description on the dataset page: https://huggingface.co/datasets/lablup/tariff_trade_domain.synthetic_trade_qa_kr.
Korean Trade Domain QA Dataset
A Korean-language question-answering dataset for the international trade domain, combining 21,399 QA pairs from three complementary sources: official 무역영어 1급 certification exam questions, trade terminology definitions, and lecture-derived QA pairs. Designed for fine-tuning and evaluating LLMs on Korean trade domain knowledge.
Dataset Description
The dataset covers vocabulary, regulations, procedures, and concepts from Korean international trade — including HS codes, Incoterms, letters of credit, customs law, and trade finance. Three source types are included and identified by the source_type field.
Data Fields
Examples
One representative row per source_type (short → medium → longer):
Source Details
exam_mcq — 무역영어 1급 Certification Exams
Official multiple-choice questions from 22 sessions of the Korean 무역영어 1급 (International Trade English) national certification exam administered by KITA (한국무역협회). Sessions covered: 27회, 28회, 29회, 30회, 31회, 32회, 33회, 34회, 35회, 38회, 39회, 41회, 42회, 43회, 44회, 45회, 46회, 47회, 48회, 50회, 51회, 52회 (22 sessions, 1,430 questions).
Topics span Korean trade law (대외무역법), export/import procedures, customs regulation, Incoterms, UCP 600 (letters of credit), international contracts, and trade insurance.
Note on `context` format for `exam_mcq`: The context field contains the full question text including all 4 answer options (numbered 1–4) and the correct answer after 정답:. The question field repeats just the question stem and options (without the appended answer).
term_qa — Trade Terminology QA
15,678 QA pairs generated from ~7,800 Korean international trade terms (파생어 inclusive). Each term produces two complementary questions:
- Description → Term: Given the definition, identify the Korean (and English) term name.
- Term → Description: Given the term name, explain its meaning.
Terms cover Incoterms, payment instruments, customs procedures, shipping documents, trade finance, and logistics terminology.
transcript_qa — Lecture Transcript QA
4,291 QA pairs generated by LLM from lecture transcript summaries covering Korean trade education content (신용장, 무역계약, 관세법, 보험, etc.). Questions are free-form; answers are concise.
Usage
from datasets import load_dataset
ds = load_dataset("lablup/synthetic_trade_qa_kr")
print(ds["train"][0])Filter by source type:
exam_only = ds["train"].filter(lambda x: x["source_type"] == "exam_mcq")
term_only = ds["train"].filter(lambda x: x["source_type"] == "term_qa")
transcript_only = ds["train"].filter(lambda x: x["source_type"] == "transcript_qa")Use as an instruction fine-tuning format:
def format_prompt(row):
return {
"prompt": f"Context: {row['context']}\n\nQuestion: {row['question']}\n\nAnswer:",
"response": row["answer"],
}
formatted = ds["train"].map(format_prompt)Data Splits
No validation or test split is provided. Create your own with train_test_split().
License
This dataset uses a mixed license:
- `term_qa` and `transcript_qa` (19,969 rows): CC BY 4.0
- `exam_mcq` (1,430 rows): Educational use only. These are verbatim or lightly reformatted questions from official KITA 무역영어 1급 certification exams. Use for research and educational purposes; redistribution or commercial use of exam content may require permission from KITA (한국무역협회).
Limitations
exam_mcqquestions are from real official exams and reflect the exam format (4-choice MCQ); they are not re-generated or paraphrasedterm_qadefinitions are original paraphrases of trade terminology; they may not match official ICC/WCO wording preciselytranscript_qaanswers are LLM-generated and have not been independently verified by domain experts- Coverage skews toward topics common in Korean trade certification exams and is not representative of all trade domains
