---
language:
  - fr
license: mit
pretty_name: French National Commission on Informatics and Liberty (CNIL)
size_categories:
  - n>1M
source_datasets: []
task_categories:
  - text-classification
  - text-generation
task_ids:
  - document-retrieval
  - document-question-answering
tags:
  - legal
  - french-law
  - CNIL
  - judicial-decisions
 
dataset_info:
  features:
    - name: id
      dtype: string  
    - name: origine
      dtype: string 
    - name: url
      dtype: string 
    - name: nature
      dtype: string 
    - name: titrefull
      dtype: string 
    - name: numero
      dtype: string 
    - name: nature_delib
      dtype: string 
    - name: date_texte
      dtype: string 
    - name: date_publi
      dtype: string 
    - name: etat_juridique
      dtype: string 
    - name: contenu
      dtype: string       
---

# French National Commission on Informatics and Liberty (CNIL) Dataset

## Dataset Description

The **CNIL Dataset** is a curated collection of documents from the French National Commission on Informatics and Liberty (CNIL). This dataset provides detailed records of decisions and deliberations made by CNIL, which governs data privacy and personal data regulation in France. It serves as a rich resource for researchers, legal practitioners, and machine learning engineers interested in exploring regulatory texts, legal reasoning, and related applications in the legal domain.

### Source Data

The data originates from the official DILA (Direction de l'Information Légale et Administrative) platform through their open data initiative. The original data can be accessed at: [DILA Open Data Platform](https://echanges.dila.gouv.fr/OPENDATA/)

## Content Description

Each record in the dataset corresponds to a decision or deliberation and includes the following fields:

- **id**: Unique identifier for the document.
- **origine**: Origin of the document.
- **url**: URL pointing to the original document.
- **nature**: Nature of the document.
- **titre**: Full title of the document.
- **numero**: Document or case number.
- **nature_delib**: Nature of the deliberation.
- **date_texte**: Date when the text was created.
- **date_publi**: Publication date of the text.
- **etat_juridique**: Legal status of the document.
- **contenu**: Full textual content of the document.

### Key Features

- **Comprehensive Metadata**: Each entry is enriched with detailed metadata extracted from the original XML files.
- **Full Content**: The complete text of each document is included for analysis.
- **Structured Data**: The dataset is stored in JSONL format, making it easy to integrate into data pipelines and machine learning workflows.

## Dataset Size and Format

- **File Format**: JSONL (JSON Lines)
- **Estimated Size**: Dependent on the volume of available CNIL documents.

## Usage

This dataset can be loaded using the Hugging Face datasets library for machine learning and natural language processing tasks.

### Example Code:

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("La-Mousse/CNIL-18-01-2025")

# View the first few entries
print(dataset["train"][0])
```

### Applications

- **Text Classification**: Train models to classify legal documents by type or category.
- **Named Entity Recognition (NER)**: Identify entities like dates, names, and legal terms.
- **Question Answering**: Develop systems to answer questions based on the content of regulatory texts.
- **Document Retrieval**: Implement search engines or information retrieval systems for legal documents.

## Processing Steps

### 1. **File Download and Extraction**
The script automates the download and extraction of `tar.gz` files containing XML documents from the CNIL open data repository.

### 2. **XML Parsing**
Each XML file is parsed to extract relevant metadata and document content using `xml.etree.ElementTree`.

### 3. **Dataset Creation**
The extracted data is saved as a JSONL file for easy consumption by data analysis tools and machine learning frameworks.

### 4. **Integration with Hugging Face**
The dataset can be uploaded to the Hugging Face hub for seamless access and sharing.

## Acknowledgments

This dataset was built using the open data provided by the French DILA initiative. Special thanks to the CNIL for their commitment to transparency and open governance.

## License

The dataset is distributed under the **MIT License**, permitting free use, modification, and distribution with proper attribution.