---
license: apache-2.0
language:
- en
- fr
- ar
- hi
tags:
- web-text
- text
- web-corpus
- web-data
pretty_name: CleanSite
size_categories:
- 100K<n<1M
task_categories:
- text-generation
---

# CleanSite

CleanSite is a multilingual web-text dataset published by **Nebulixlabs** for research, experimentation, and machine learning workflows involving large-scale web text.

The current dataset contains a single `train` split with **564,349 rows**. The dataset viewer exposes three primary fields:

* `text` — extracted web-page text
* `source` — source URL associated with the record
* `date` — timestamp associated with the record

## Dataset Summary

| Property  | Value        |
| --------- | ------------ |
| Dataset   | CleanSite    |
| Publisher | Nebulixlabs  |
| Modality  | Text         |
| Split     | train        |
| Rows      | 564,349      |
| Format    | Parquet      |
| Languages | Multilingual |

## Data Structure

### `text`

The textual content associated with a web-page or web document.

### `source`

The source URL associated with the record.

### `date`

A timestamp associated with the source record.

## Intended Use

CleanSite can be used for:

* web-text and corpus research
* text preprocessing experiments
* language-model and NLP research
* web-data analysis
* information retrieval experiments
* dataset prototyping and benchmarking

## Limitations

The exact source collection, filtering, cleaning, deduplication, and licensing pipeline should be documented here by the dataset maintainer.

Because the data originates from web content, records may contain duplicated material, low-quality text, outdated information, offensive material, copyrighted text, or other content that requires additional filtering before downstream use.

Users should independently validate the dataset for their intended application.

## Data Quality and Preprocessing

The current public repository does not fully document the complete preprocessing pipeline.

Before using this dataset for training or evaluation, users should consider:

* deduplication
* language identification
* HTML and boilerplate removal
* quality filtering
* personally identifiable information filtering
* unwanted or unsafe-content filtering
* train/evaluation leakage checks

## Licensing and Attribution

The dataset repository currently does not provide enough licensing information to make a definitive statement about the legal status of every source document.

Users are responsible for verifying the applicable licenses, permissions, and terms for data used in their projects.

## Maintainer

**Nebulixlabs**

Hugging Face: https://huggingface.co/Nebulixlabs

## Citation

If you use CleanSite in research or software, please cite the Hugging Face repository:

```text
Nebulixlabs. CleanSite. Hugging Face Datasets.
https://huggingface.co/datasets/Nebulixlabs/CleanSite
```

## Additional Information

For questions, improvements, or collaboration, please use the repository discussion/community features on Hugging Face.