Update README.md

Overview

Dataset Name: 101 Billion Arabic Words Dataset
Curated by: Clusterlab
Contributors: Manel Aloui, Hasna Chouikhi, Ghaith Chaabane, Haithem Kchaou, and Chehir Dhaouadi
Language(s): Modern Standard Arabic
License: Apache 2.0 (License Link)
Repository: HuggingFace Dataset Page

Description

The 101 Billion Arabic Words Dataset is a comprehensive corpus specifically designed for training and fine-tuning Arabic Language Models (LLMs). It encompasses a vast volume of text, amounting to 101 billion words, sourced primarily from web content through extensive data mining and cleaning processes. This dataset is pivotal for promoting linguistic diversity and technological equity in AI, focusing on enhancing Arabic language processing capabilities.

Authorship

Publishing Organization: Clusterlab
Industry Type: Tech Start-Up
Contact Details: Clusterlab Contact

Data Points

Number of Words: 101 billion
Format: JSONL

Motivations & Intentions

This dataset was curated with the aim of addressing the scarcity of high-quality, large-scale linguistic data for the Arabic language, which has historically been underrepresented in NLP research. By providing this dataset, Clusterlab intends to facilitate the development of models that can understand and generate Arabic with high accuracy, supporting both academic research and practical applications.

Provenance

Methods Used: Data were extensively gathered from Common Crawl archives, with focus on extracting and cleaning Arabic content. The extraction process utilized Rust for efficient processing, and a rigorous deduplication and noise reduction pipeline was applied to ensure data quality.
Methodology Details:

Source: Mainly Common Crawl
Dates of Collection: September 2021 - July 2022
Data Collection Process: Extraction from WET files, filtering, deduplication
Cleaning and Preprocessing Techniques: See section 3.2 of the paper for detailed techniques and tools used
Dataset Version and Maintenance

Maintenance Status: Actively Maintained
Version: 1.0
Maintenance Plan: The dataset will be updated periodically to refine the quality and expand the scope based on user feedback and technological advancements.

Use Cases & Applications

This dataset is intended for use in training and fine-tuning Arabic LLMs capable of performing a variety of NLP tasks, including but not limited to, text classification, sentiment analysis, and machine translation. Its extensive and diverse nature makes it an excellent resource for developing robust models that are well-adapted to the nuances of the Arabic language.

README.md

@@ -1,4 +1,9 @@
 ---
 license: apache-2.0
+language:
+- ar
+pretty_name: 101 Billion Arabic Words Dataset
+size_categories:
+- 100B<n<1T
 ---
 Upload In Progress ...