Datasets:
license: cc-by-nc-4.0
language:
- ps
- ur
- ar
- fa
tags:
- offensive
- not-offensive
POLD: Pashto Offensive Language Dataset
POLD is a benchmark dataset developed to train and evaluate NLP models for detecting offensive textual content on online social networks (OSNs) in the Pashto language. It was collected from Twitter and manually labeled for offensive language detection. The dataset consists of 34,400 instances categorized into two classes: offensive (represented by 1) and not-offensive (represented by 0). For reference, the texts are also translated into English. The dataset contains three columns: text, translation, and label.
Usage
POLD dataset is available at zirak-ai/pold and can be loaded into any NLP pipeline with a single line of code:
from datasets import load_dataset
dataset = load_dataset("zirak-ai/pold")
Kaggle
The dataset is also available on Kaggle at: POLD – Pashto Offensive Language Dataset
Citation
Please cite the following work if you use this dataset:
Ijazul Haq, Weidong Qiu, Jie Guo, Peng Tang. “Pashto Offensive Language Detection: A Benchmark Dataset and Monolingual Pashto BERT.” PeerJ Computer Science, vol. 9, 2023, e1617. https://doi.org/10.7717/peerj-cs.1617
@article{haq2023pold,
title={Pashto offensive language detection: a benchmark dataset and monolingual Pashto BERT},
author={Ijazul Haq and Weidong Qiu and Jie Guo and Peng Tang},
journal={PeerJ Computer Science},
volume={9},
pages={e1617},
year={2023},
issn={2376-5992},
doi={10.7717/peerj-cs.1617}
}