TuniziBigBench: The Largest Open-Source Tunizi (Darija) Dataset

Overview

This dataset was created by scraping over 14,000 Tunisian YouTube videos, providing a rich repository of Tunisian language data. It covers a wide range of topics and text types, including politics, news, football, and more. The dataset is particularly valuable for training and fine-tuning natural language processing models specific to Tunisian Arabic and other local dialects. We are working One 1B dataset stay tuned

• Curated by: Nehdi Taha Mustapha & Oussama Sassi.
• Contact: https://www.linkedin.com/in/taha-mustapha-nehdi-240585203/

Dataset Structure

The dataset is structured in the following way:

• Fields: The dataset contains fields for the text content

Licensing

This dataset is released under the CreativeML Open RAIL-M license.

Usage

from datasets import load_dataset

# Load the dataset from Hugging Face
dataset = load_dataset('Nehdi/TuniziBigBench')

# Example: Print the first 5 entries from the dataset
for entry in dataset['train'][:5]:
    print(entry['text'])

Citation

If you use this dataset in your research, please cite it as follows:

@dataset{TuniziBigBench,
  author = {Nehdi Taha Mustapha},
  title = {Tunizi Tunisian Dialectic Derja Dataset},
  year = {2025},
  url = {https://huggingface.co/datasets/Nehdi/TuniziBigBench}
}