Datasets:
id int64 0 5k | input stringlengths 8 17 | output stringdate 2020-01-01 00:00:00 2030-12-29 00:00:00 |
|---|---|---|
0 | 2021.11.07 | 2021-11-07 |
1 | 2026년 Jun 19일 | 2026-06-19 |
2 | 2022년 12월18일 | 2022-12-18 |
3 | 2026/08/24 | 2026-08-24 |
4 | 2021/2/11 | 2021-02-11 |
5 | 28년 3월 7일 | 2028-03-07 |
6 | 2022/12/25 | 2022-12-25 |
7 | 날짜: 2020년 5월 31일 | 2020-05-31 |
8 | 2025년 Mar 9일 | 2025-03-09 |
9 | 2028년 십이월 이십구일 | 2028-12-29 |
10 | 2025.6.24 | 2025-06-24 |
11 | 2026년도 9월 3일 | 2026-09-03 |
12 | 2029년 Oct 8일 | 2029-10-08 |
13 | 2020.07.19 | 2020-07-19 |
14 | 28.05.01 | 2028-05-01 |
15 | 2026/1/9 | 2026-01-09 |
16 | 이천이십이년 구월 십일 | 2022-09-10 |
17 | 5/17/2030 | 2030-05-17 |
18 | 30.02.13 | 2030-02-13 |
19 | 2021년 12월 25일 | 2021-12-25 |
20 | 이천이십이년 일월 십구일 | 2022-01-19 |
21 | 2020년 Sep 6일 | 2020-09-06 |
22 | 2021/03/28 | 2021-03-28 |
23 | 2024년 9월22일 | 2024-09-22 |
24 | 2020-4-28 | 2020-04-28 |
25 | 22/03/16 | 2022-03-16 |
26 | [2024.04.05] | 2024-04-05 |
27 | 2024 년 4 월 15 일 | 2024-04-15 |
28 | 2029년 Sep 9일 | 2029-09-09 |
29 | 2023 년 11 월 25 일 | 2023-11-25 |
30 | Date: 2027/07/09 | 2027-07-09 |
31 | 2024년 4월 02일 | 2024-04-02 |
32 | 2028年1月20日 | 2028-01-20 |
33 | 이천이십일년 십일월 칠일 | 2021-11-07 |
34 | 날짜: 2029년 2월 20일 | 2029-02-20 |
35 | 2026-10-20 | 2026-10-20 |
36 | [2025.07.03] | 2025-07-03 |
37 | (2023년 5월 23일) | 2023-05-23 |
38 | 6/18/2024 | 2024-06-18 |
39 | 4/27/2025 | 2025-04-27 |
40 | 2025년 5월 17일 | 2025-05-17 |
41 | 2029 년 11 월 5 일 | 2029-11-05 |
42 | 30년 9월 5일 | 2030-09-05 |
43 | 22.03.26 | 2022-03-26 |
44 | 2024년 사월 십육일 | 2024-04-16 |
45 | 2027.4.3 | 2027-04-03 |
46 | 24년 9월 29일 | 2024-09-29 |
47 | 2029년도 5월 1일 | 2029-05-01 |
48 | 20/04/04 | 2020-04-04 |
49 | Date: 2027/06/02 | 2027-06-02 |
50 | 20년 5월 20일 | 2020-05-20 |
51 | 24/12/08 | 2024-12-08 |
52 | 2028. 5. 19. | 2028-05-19 |
53 | 2025년 십일월 이십이일 | 2025-11-22 |
54 | 2020.04.27 | 2020-04-27 |
55 | 2026.10.14 | 2026-10-14 |
56 | 27/03/27 | 2027-03-27 |
57 | Date: 2028/02/10 | 2028-02-10 |
58 | 2030년 1월10일 | 2030-01-10 |
59 | 2022.04.24 | 2022-04-24 |
60 | (2027년 4월 30일) | 2027-04-30 |
61 | 2024.07.08 | 2024-07-08 |
62 | 2029.02.11 | 2029-02-11 |
63 | 날짜: 2026년 7월 31일 | 2026-07-31 |
64 | 2021. 6. 14 | 2021-06-14 |
65 | Date: 2021/02/07 | 2021-02-07 |
66 | 2027년 9월 6일 | 2027-09-06 |
67 | 2027년 팔월 이십육일 | 2027-08-26 |
68 | 25년 7월 13일 | 2025-07-13 |
69 | 2025년 8월 26일 | 2025-08-26 |
70 | 7/4/2026 | 2026-07-04 |
71 | 3/27/2027 | 2027-03-27 |
72 | 2022년도 4월 6일 | 2022-04-06 |
73 | 5월 1일 2028년 | 2028-05-01 |
74 | 2024년 일월 이십삼일 | 2024-01-23 |
75 | 날짜: 2026년 5월 8일 | 2026-05-08 |
76 | 2030. 8. 4. | 2030-08-04 |
77 | 2028년 십일월 이십오일 | 2028-11-25 |
78 | 2023년도 12월 17일 | 2023-12-17 |
79 | 10월 22일, 2029년 | 2029-10-22 |
80 | 2월 15일, 2026년 | 2026-02-15 |
81 | 2024년 사월 십일일 | 2024-04-11 |
82 | (2026년 9월 25일) | 2026-09-25 |
83 | 2026/3/21 | 2026-03-21 |
84 | 2028/03/10 | 2028-03-10 |
85 | 1월 19일, 2020년 | 2020-01-19 |
86 | 2029. 8. 02 | 2029-08-02 |
87 | 2027. 1. 26 | 2027-01-26 |
88 | 이천이십육년 오월 사일 | 2026-05-04 |
89 | 2020/02/13 | 2020-02-13 |
90 | 11/16/2029 | 2029-11-16 |
91 | 날짜: 2027년 1월 14일 | 2027-01-14 |
92 | 11/22/2029 | 2029-11-22 |
93 | 2022년 3월25일 | 2022-03-25 |
94 | 2028/10/8 | 2028-10-08 |
95 | 2028 년 4 월 25 일 | 2028-04-25 |
96 | 2027년도 1월 15일 | 2027-01-15 |
97 | Date: 2025/04/30 | 2025-04-30 |
98 | 2025.12.20 | 2025-12-20 |
99 | Date: 2024/06/10 | 2024-06-10 |
End of preview. Expand
in Data Studio
Korean Date Translation Dataset
한국어 날짜 표현을 ISO 8601 형식(YYYY-MM-DD)으로 변환하는 데이터셋입니다.
Dataset Description
이 데이터셋은 자연어처리 실습을 위해 제작된 교육용 합성 데이터셋입니다. 다양한 한국어 날짜 표현을 표준화된 형식으로 변환하는 Sequence-to-Sequence (N-to-M) 과제용입니다.
Dataset Summary
- 언어: 한국어 (Korean)
- 도메인: 날짜 표현 정규화
- 태스크: Text-to-Text Generation (Date Parsing / Translation)
- 생성 방식: 템플릿 기반 합성 데이터 (Synthetic)
Supported Tasks
- Text-to-Text Generation: 입력 날짜 표현 -> ISO 8601 형식 변환
- N-to-M Task: 가변 길이 입력 -> 고정 형식 출력 (YYYY-MM-DD)
Languages
한국어 (Korean, ko)
Dataset Structure
Data Instances
{
"id": 0,
"input": "2025년 12월 3일",
"output": "2025-12-03"
}
Data Fields
| Field | Type | Description |
|---|---|---|
id |
int | 샘플 고유 ID |
input |
str | 한국어 날짜 표현 (다양한 형식) |
output |
str | ISO 8601 형식의 날짜 (YYYY-MM-DD) |
Input Variations
이 데이터셋은 다양한 한국어 날짜 표현을 포함합니다:
| 카테고리 | 예시 |
|---|---|
| 기본 한글 | "2025년 12월 3일", "2025년 12월 03일" |
| 2자리 연도 | "25년 12월 3일" |
| 슬래시 구분 | "2025/12/3", "2025/12/03" |
| 점 구분 | "2025.12.3", "2025.12.03" |
| 하이픈 구분 | "2025-12-3" |
| 순서 변형 | "12월 3일 2025년", "12월 3일, 2025년" |
| 한글 숫자 | "2025년 십이월 삼일" |
| 완전 한글 | "이천이십오년 십이월 삼일" |
| 공백 변형 | "2025 년 12 월 3 일" |
| 접미사 변형 | "2025년도 12월 3일" |
| 축약형 | "2025. 12. 3." |
| 괄호 사용 | "(2025년 12월 3일)", "[2025.12.03]" |
| 한자 | "2025年12月3日" |
| 접두어 | "날짜: 2025년 12월 3일" |
| 영문 혼합 | "2025년 Dec 3일" |
Data Splits
| Split | Examples | Description |
|---|---|---|
| train | 5,000 | 학습용 데이터 |
| validation | 500 | 검증용 데이터 |
| test | 500 | 테스트용 데이터 |
Date Range
- 연도: 2020년 ~ 2030년
- 월: 1월 ~ 12월
- 일: 1일 ~ 31일 (월별 유효 범위 내)
Dataset Creation
Curation Rationale
이 데이터셋은 다음 목적으로 제작되었습니다:
- 교육용: 딥러닝/NLP 입문자를 위한 Seq2Seq 실습 데이터
- N-to-M 과제 학습: 가변 길이 입력 -> 고정 형식 출력 변환
- T5/GPT Fine-tuning: Encoder-Decoder 또는 Decoder-only 모델 실습
- Date Parsing: 자연어 날짜 표현의 정규화 학습
Source Data
Initial Data Collection and Normalization
- 생성 방식: 템플릿 기반 합성 (Template-based Synthetic Generation)
- 날짜 범위: 2020-2030년
- 변형 유형: 15개 이상의 다양한 날짜 표현 패턴
Who are the source language producers?
합성 데이터로, 실제 문서나 사용자 생성 콘텐츠가 아닙니다.
Annotations
Annotation process
템플릿 기반 자동 생성으로, 입력 생성 시 정답(ISO 형식)이 함께 결정됩니다.
Who are the annotators?
자동 생성된 레이블 (생성 규칙에 따라 결정)
Personal and Sensitive Information
이 데이터셋은 합성 데이터로, 개인정보나 민감 정보를 포함하지 않습니다.
Considerations for Using the Data
Social Impact of Dataset
교육 목적으로 제작된 데이터셋으로, 실제 문서의 날짜 표현과 다를 수 있습니다.
Discussion of Biases
- 템플릿 기반 생성으로 표현 패턴이 제한적일 수 있습니다
- 실제 텍스트에서 발생하는 오타나 비표준 표현은 포함되지 않습니다
- 교육 및 실습 목적에 적합하며, 실제 서비스 적용에는 추가 검증이 필요합니다
Other Known Limitations
- 합성 데이터의 한계: 실제 텍스트의 다양성 부족
- 문맥 없음: 단독 날짜 표현만 포함 (문장 내 날짜 추출 미지원)
- 시간 미포함: 시간 정보(시:분:초) 없음
- 기원전/기원후: 음력, 기원전 연도 등 미지원
Additional Information
Dataset Curators
정상근 (hugmanskj@gmail.com)
Licensing Information
CC BY 4.0 (Creative Commons Attribution 4.0 International)
이 데이터셋은 교육 목적으로 자유롭게 사용할 수 있습니다.
Citation Information
@dataset{korean_date_translation,
author = {정상근},
title = {Korean Date Translation Dataset},
year = {2024},
publisher = {Hugging Face},
note = {Educational dataset for deep learning textbook}
}
How to Use
Loading the Dataset
from datasets import load_dataset
# Hugging Face Hub에서 로드
dataset = load_dataset("YOUR_USERNAME/korean-date-translation")
# 또는 로컬 파일에서 로드
dataset = load_dataset("json", data_files={
"train": "train.json",
"validation": "valid.json",
"test": "test.json"
})
Example Usage with T5
from transformers import T5Tokenizer, T5ForConditionalGeneration
from transformers import Seq2SeqTrainer, Seq2SeqTrainingArguments
# 모델 및 토크나이저 로드
model_name = "google/mt5-small" # 또는 한국어 T5 모델
tokenizer = T5Tokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)
# 데이터 전처리
def preprocess_function(examples):
# 입력에 task prefix 추가
inputs = ["날짜 변환: " + inp for inp in examples["input"]]
targets = examples["output"]
model_inputs = tokenizer(
inputs,
max_length=64,
truncation=True,
padding="max_length"
)
labels = tokenizer(
targets,
max_length=16,
truncation=True,
padding="max_length"
)
model_inputs["labels"] = labels["input_ids"]
return model_inputs
# 데이터셋 전처리
tokenized_dataset = dataset.map(preprocess_function, batched=True)
# 학습
training_args = Seq2SeqTrainingArguments(
output_dir="./results",
num_train_epochs=5,
per_device_train_batch_size=16,
per_device_eval_batch_size=16,
evaluation_strategy="epoch",
predict_with_generate=True,
)
trainer = Seq2SeqTrainer(
model=model,
args=training_args,
train_dataset=tokenized_dataset["train"],
eval_dataset=tokenized_dataset["validation"],
tokenizer=tokenizer,
)
trainer.train()
Example Usage with GPT-style Model
from transformers import AutoTokenizer, AutoModelForCausalLM
# 모델 로드
model_name = "skt/kogpt2-base-v2" # 한국어 GPT
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
# 프롬프트 형식
def format_prompt(input_date):
return f"날짜 변환\n입력: {input_date}\n출력:"
# 추론
input_text = "2025년 12월 3일"
prompt = format_prompt(input_text)
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=12)
result = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(result) # 날짜 변환\n입력: 2025년 12월 3일\n출력: 2025-12-03
Sample Data
Example 1: Basic Korean Format
Input: "2025년 12월 3일"
Output: "2025-12-03"
Example 2: Slash Separated
Input: "2025/12/03"
Output: "2025-12-03"
Example 3: Full Korean Numerals
Input: "이천이십오년 십이월 삼일"
Output: "2025-12-03"
Example 4: Order Variation
Input: "12월 3일, 2025년"
Output: "2025-12-03"
Example 5: With Prefix
Input: "날짜: 2025년 12월 3일"
Output: "2025-12-03"
Evaluation Metrics
이 데이터셋에서 사용할 수 있는 평가 지표:
- Exact Match (EM): 정확히 일치하는 비율
- Character Error Rate (CER): 문자 수준 오류율
- Token Accuracy: 토큰별 정확도
권장 평가 방식: Exact Match가 가장 적합합니다. 날짜 형식이 정확해야 의미가 있기 때문입니다.
- Downloads last month
- 16