中文裁判文书法院观点数据集

# CCVG: Chinese Court View Generation Dataset

[中文](README_zh.md) | 🤗 [huggingface](https://huggingface.co/datasets/TIM0927/CCVG) | 🤖 [modelscope](https://www.modelscope.cn/datasets/ZhitianHou/CCVG) | 📄 [Arxiv](https://arxiv.org/abs/2510.09297) | 💻 [GitHub](https://github.com/ZhitianHou/ShiZhi)

**Now the anonymous test data has been released.**

For privacy protection reasons, the original training data has been removed pending de-identification and anonymization. Only some examples of training data are avaiable.

**CCVG** is a curated Chinese dataset designed for **Criminal Court View Generation (CVG)** and **charge prediction** tasks.
It contains criminal case documents with **fact descriptions** and **court views**, supporting research in legal AI and natural language generation.

---

## Dataset Overview

- **Language:** Chinese
- **Domain:** Legal / Criminal law
- **Task:** Court View Generation & Charge Prediction
- **Size:** 110K+ criminal cases
- **Time Span:** 1985–2021

Each case contains:
1. **system**: The system prompt we used in Training.
2. **query:** The "Fact" section summarying the case fact in legal case documents.
3. **response:** The "Court View" section explaining the judgment in legal case documents.
4. **charge:** The criminal charge corresponding to the case.

## Citation

If you find this project helpful, please consider citing our paper:

```bibtex
@misc{hou2025shizhichineselightweightlarge,
title={ShiZhi: A Chinese Lightweight Large Language Model for Court View Generation},
author={Zhitian Hou and Kun Zeng},
year={2025},
eprint={2510.09297},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2510.09297},
}
```

# CCVG：中国法院裁判观点生成数据集

[中文文档](README_zh.md) | 🤗 [Hugging Face](https://huggingface.co/datasets/TIM0927/CCVG) | 🤖 [ModelScope](https://www.modelscope.cn/datasets/ZhitianHou/CCVG) | 📄 [Arxiv论文](https://arxiv.org/abs/2510.09297) | 💻 [GitHub仓库](https://github.com/ZhitianHou/ShiZhi)

**目前匿名测试数据集已正式发布。**

出于隐私保护的相关要求，原始训练数据集已暂予移除，待完成去标识化与匿名化处理；当前仅开放部分训练数据示例供参考。

**CCVG**是一款精心整理的中文数据集，专为**刑事法院裁判观点生成（Criminal Court View Generation, CVG）**与**罪名预测**两类任务打造。该数据集收录了包含**事实描述**与**法院裁判观点**的刑事案卷文档，可为法律人工智能与自然语言生成领域的相关研究提供支撑。

---

## 数据集概览

- **语言：** 中文
- **领域：** 法律 / 刑法
- **任务：** 法院裁判观点生成与罪名预测
- **规模：** 11万+ 刑事案例
- **时间跨度：** 1985–2021

每个案例包含以下内容：
1. **system**：训练过程中所使用的系统提示词。
2. **query**：刑事案卷文档中用于概括案件事实的“事实”章节内容。
3. **response**：刑事案卷文档中用于阐释裁判理由与结果的“法院裁判观点”章节内容。
4. **charge**：对应案件的刑事罪名。

---

## 引用说明

若本项目对您的研究有所帮助，请引用我们的论文：

bibtex
@misc{hou2025shizhichineselightweightlarge,
title={ShiZhi: A Chinese Lightweight Large Language Model for Court View Generation},
author={Zhitian Hou and Kun Zeng},
year={2025},
eprint={2510.09297},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2510.09297},
}

搜集汇总

背景与挑战

背景概述

CCVG是一个专门用于刑事法院观点生成和罪名预测任务的中文数据集，包含超过11万条刑事案例，每个案例都提供案件事实描述和法院观点。由于隐私保护，原始训练数据已移除，仅保留部分示例。