DibraText Dataset

General Information

• Description: DibraText is an aggregated dataset consisting of Albanian language texts collected from various sources. It's designed for natural language processing tasks such as text classification, sentiment analysis, and machine learning model training.
• Author: Florijan Qosja
• Maintainer: Florijan Qosja (florijanqosja@gmail.com)
• Created: 27/04/2024
• Last Updated: 27/04/2024
• Language: Albanian (Language Code: sq)
• Volume: 100,378,107 unique values
• Language: Albanian (Language Code: sq)

How to Access

from datasets import load_dataset

# Load the DibraText dataset
dataset = load_dataset("florijanqosja/dibratext")

# Access the train and test split
train_set = dataset['train']
test_set = dataset['test']

Source Data

• Original Sources:

  1. CulturaX: Hugging Face Dataset
  2. C4: Hugging Face Dataset
  3. Wikipedia: Hugging Face Dataset
  4. Kosovo News Articles Dataset: Kaggle Dataset
  5. OSCAR: Hugging Face Dataset
  6. CC100: Hugging Face Dataset
  7. OSCAR-2201: Hugging Face Dataset

• Inclusion Criteria: Only datasets containing Albanian language content were included. All data subsets are filtered to include only texts in Albanian.

Dataset Structure

• Data Instances: A data instance typically consists of text content in Albanian. [Provide more details if the data instances include more elements such as titles, dates, etc.]
• Data Fields:
  • text: The main body of the text in Albanian.
• Data Volume: 100378107 unique rows
• Data Split:
  • train: 90340296
  • test: 10037811

Licensing Information

• License Details:
  1. CulturaX, mC4, OSCAR: Distributed under their respective licenses.
  2. C4 (allenai/c4): Released under the ODC-BY license, bound by Common Crawl's terms of use.
  3. Wikipedia: Content licensed under the GFDL and CC-BY-SA 3.0.
  4. OSCAR and OSCAR-2201: Released under the CC0 license ("no rights reserved").
  5. CC100: User is bound by the Common Crawl's terms of use.
  6. Kosovo News Articles Dataset: Licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

This dataset is released under multiple licenses due to the combination of various sources. Users must comply with the individual license terms of each component of the dataset:

• Common Crawl terms apply to parts of the data from datasets such as C4 and CC100.
• GFDL and CC-BY-SA for Wikipedia content.
• CC0 for the OSCAR datasets, promoting wide reuse without restrictions.

Limitations and Bias

• Known Limitations: The dataset is limited to texts in the Albanian language and may not represent all dialects or sociolects equally.
• Potential Bias: As the data sources vary in origin and context, biases present in individual sources may be propagated in this aggregated dataset.

Citation Information

@misc{dibratext2023,
  author       = {Florijan Qosja},
  title        = {DibraText: An Aggregated Dataset for Albanian Text Processing},
  year         = 2023,
  publisher    = {HuggingFace},
  journal      = {HuggingFace Dataset},
  howpublished = {\url{https://huggingface.co/datasets/florijanqosja/dibratext}}
}

Usage

• Intended Use: This dataset is intended for academic and research purposes, particularly in the fields of natural language processing and machine learning focusing on the Albanian language.
• Should Not Be Used For: The dataset should not be used as the sole basis for critical decisions without proper validation and additional sources of information.

Contributions

Thanks to @florijanqosja for adding this dataset.