---
title: Chinese_Poetry_Dataset
canonical_url: "https://www.modelscope.cn/datasets/modelscope_mp_827305057/Chinese_Poetry_Dataset"
md_url: "https://www.modelscope.cn/datasets/modelscope_mp_827305057/Chinese_Poetry_Dataset.md"
repository: modelscope_mp_827305057/Chinese_Poetry_Dataset
chinese_name: "中国古诗词全集"
last_updated: 2026-07-26
license: cc-by-nc-4.0
downloads: 45
stars: 0
---

# Chinese_Poetry_Dataset

> Chinese_Poetry_Dataset - modelscope_mp_827305057 在 ModelScope 开源的数据集。中华古诗词数据集 (Chinese Poetry Dataset)

modelscope_mp_827305057/Chinese_Poetry_Dataset 是 ModelScope 魔搭社区上的数据集，采用 cc-by-nc-4.0 许可。

- **Repository**: modelscope_mp_827305057/Chinese_Poetry_Dataset
- **License**: cc-by-nc-4.0
- **Downloads**: 45
- **Stars**: 0
- **Last updated**: 2026-07-26

Source: https://www.modelscope.cn/datasets/modelscope_mp_827305057/Chinese_Poetry_Dataset

---

# 中华古诗词数据集 (Chinese Poetry Dataset)

> 从中华古诗词开源数据库整理而成，共 **388,883** 首诗词，涵盖唐、宋、五代、元、先秦、周、汉、清等多个朝代。

## 数据格式

JSONL 格式（每行一条 JSON 记录），字段仅含 `text`，将标题、朝代、作者、正文融合为一句自然语言描述：

```json
{"text": "登鹳雀楼是唐朝王之涣写的：白日依山尽，黄河入海流。欲穷千里目，更上一层楼。"}
{"text": "静夜思是唐朝李白写的：床前明月光，疑是地上霜。举头望明月，低头思故乡。"}
{"text": "春晓是唐朝孟浩然写的：春眠不觉晓，处处闻啼鸟。夜来风雨声，花落知多少。"}
```

格式说明：

```
{text} = {title}是{dynasty}朝{author}写的：{content}
```

| 组成部分 | 说明 | 示例 |
|---------|------|------|
| title | 诗词标题/词牌名 | 登鹳雀楼、清平乐·上陽春晚 |
| dynasty | 朝代 | 唐、宋、五代、元、先秦、周、汉、清 |
| author | 作者 | 李白、杜甫、温庭筠、佚名 |
| content | 诗词正文 | 白日依山尽，黄河入海流…… |

## 数据统计

| 项目 | 数值 |
|------|------|
| 总记录数 | 388,883 条 |
| 文件大小 | 93.89 MB |
| 文件格式 | JSONL (UTF-8) |
| 平均每条长度 | 80 字符 |
| 最长记录 | 22,946 字符 |
| 最短记录 | 9 字符 |

### 朝代分布

| 朝代 | 记录数 | 占比 |
|------|--------|------|
| 唐 | 355,535 | 91.4% |
| 宋 | 21,333 | 5.5% |
| 元 | 10,890 | 2.8% |
| 五代 | 506 | 0.1% |
| 周（诗经） | 305 | 0.1% |
| 先秦（楚辞） | 65 | <0.1% |
| 汉（曹操） | 26 | <0.1% |
| 清（纳兰性德） | 4 | <0.1% |
| 其他（蒙学等） | 219 | 0.1% |

## 数据来源

基于以下开源项目整理转换：

- **[chinese-poetry/chinese-poetry](https://github.com/chinese-poetry/chinese-poetry)** - 最全中华古诗词数据库（GitHub 52.8k+ Stars）

原始数据包含：

- 《全唐诗》约 5.5 万首
- 《全宋诗》约 26 万首
- 《全宋词》约 2.1 万首
- 《五代诗词》（花间集、南唐）
- 《元曲》
- 《楚辞》
- 《诗经》305 篇
- 《论语》、《四书五经》
- 蒙学经典（百家姓、弟子规、古文观止、千家诗、千字文、三字经等）
- 纳兰性德诗集
- 曹操诗集
- 幽梦影
- 水墨唐诗
- 御定全唐诗 900 卷

## 数据处理说明

### 转换规则

1. **统一格式**：将原始数据中多种不同的 JSON 结构（含 `paragraphs`、`content`、`rhythmic` 等字段）统一转换为 `{"text": "..."}` 格式。
2. **朝代标注**：根据原始文件路径自动推断朝代信息。
3. **正文拼接**：将多段 `paragraphs`/`content` 数组拼接为单行文本。
4. **编码统一**：保留原始繁简体混合文本，未做转换。

### 已排除的数据

以下非诗词内容未包含在数据集中：

- `loader/` - 加载器配置
- `strains/` - 平仄结构数据
- `error/` - 校验错误记录
- `表面结构字` - 汉字结构数据
- `authors.*` - 作者列表（已内嵌到每条记录中）

## 使用方法

### Python

```python
import json

with open("poetry.jsonl", "r", encoding="utf-8") as f:
    for line in f:
        record = json.loads(line)
        print(record["text"])
```

### ModelScope / 魔搭社区

```python
from modelscope.msdatasets import MsDataset

dataset = MsDataset.load("your-username/chinese-poetry-dataset")
print(dataset[0])
```

### Hugging Face Datasets

```python
from datasets import load_dataset

dataset = load_dataset("json", data_files="poetry.jsonl")
print(dataset["train"][0])
```

### LLM 微调

本数据集格式兼容大语言模型指令微调场景，可直接用于：

- 古诗词续写训练
- 文学风格迁移
- 诗词知识问答
- 中文 NLP 预训练

## 许可证

本数据集采用 [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/deed.zh-hans)（署名-非商业性使用 4.0 国际）协议。

- ✅ 允许：共享、改编、注明出处
- ❌ 禁止：商业用途

原始诗词文本属于公共领域。数据整理转换基于 [chinese-poetry/chinese-poetry](https://github.com/chinese-poetry/chinese-poetry) 开源项目。

## 引用

如果本数据集对您的研究有帮助，请引用原始数据来源：

```bibtex
@misc{chinese-poetry,
  author = {chinese-poetry},
  title  = {The most comprehensive database of Chinese poetry},
  url    = {https://github.com/chinese-poetry/chinese-poetry},
  year   = {2023}
}
```
