---
license: cc-by-nc-sa-4.0
language:
- zh
tags:
- legal-ai
- chinese
- text-classification
- nlp
---

# Refined Chinese Legal Dataset (RCLD)

## Dataset Description

This dataset contains refined excerpts from Chinese legal documents, primarily sourced from the **CAIL2018 benchmark** (The China AI and Law Challenge). It is specifically curated for natural language processing tasks within the legal domain, such as legal judgment prediction, charge prediction, and as a factual basis for dialogue systems and text analysis.

The dataset consists of factual descriptions of criminal cases (`fact`) along with associated metadata (`meta`), including the accusation, relevant legal articles, and sentencing information.

## Data Processing

The raw data from CAIL2018 was processed through a meticulous multi-stage pipeline to ensure quality and consistency:

1.  **Initial Filtering and Cleaning**: Raw case files were merged and cleaned. This involved removing textual artifacts, boilerplate text (e.g., "公诉机关指控"), and filtering records based on content and length to ensure each entry is a self-contained factual description.

2.  **Fact Refinement via LLM**: To enhance the clarity and narrative consistency of the case descriptions, we employed a Large Language Model to refine each `fact` entry. This summarization and rephrasing step converts dense legal jargon into a more structured and natural language format suitable for NLP modeling.

3.  **Data Splitting**: The final, cleaned dataset was partitioned into training, validation, and test sets to facilitate standardized model training and evaluation.

## Data Fields

Each record in the dataset is a single JSON object with the following structure:

-   `fact`: (string) A textual description of the case facts, refined by an LLM.
-   `meta`: (object) A dictionary containing structured metadata about the case.
    -   `relevant_articles`: (list of int) A list of relevant law article numbers from the Chinese Criminal Law.
    -   `accusation`: (list of string) The criminal charge(s), e.g., `["盗窃"]` (theft).
    -   `punish_of_money`: (int) The monetary penalty imposed, in CNY.
    -   `criminals`: (list of string) The name(s) of the defendant(s).
    -   `term_of_imprisonment`: (object) An object detailing the prison sentence:
        -   `death_penalty`: (boolean) Whether the death penalty was imposed.
        -   `imprisonment`: (int) The term of imprisonment in months. A value of 0 may indicate a suspended sentence or other non-custodial penalty.
        -   `life_imprisonment`: (boolean) Whether life imprisonment was imposed.

## Data Splits

The dataset is divided into three splits with the following number of samples:

| Split      | Number of Samples |
| ---------- | ----------------- |
| train      | 122,928           |
| validation | 16,451            |
| test       | 31,173            |

## How to use

You can load the dataset using the Hugging Face `datasets` library:

```python
from datasets import load_dataset

dataset = load_dataset("...")

print(dataset['train'][0])