Datasets:

Modalities:
Tabular
Text
Languages:
Greek
DOI:
Libraries:
Datasets
License:
toxic-el / README.md
andefined's picture
Initial commit
27f8656
metadata
license: gpl-3.0
task_categories:
  - text-classification
language:
  - el
tags:
  - text-classification
  - toxicity
size_categories:
  - 10K<n<100K
pretty_name: Toxic Tweets, Greek Dataset
multilinguality:
  - monolingual

Toxic Tweets, Greek Dataset

Dataset Description

A frequent debate regarding research on media platforms revolves around the term of toxicity. However, the term itself is poorly defined and often contradictory, encompassing everything between online harassment and bullying to a negative commentary. In order to map online toxicity on Twitter we formed 7 different categories: i) hateful ii) insulting iii) threatening iv) racist v) sexist vi) using anti-refugee rhetoric vii) using nationalistic language (table 4). These categories were clustered into wider groups. Every tweet that contained at least one of the above mentioned categories was marked as toxic. Further on, tweets that included hateful language, threats and / or insults were marked as severe toxic. Tweets targeting an individual or a group based on identity characteristics such as gender, ethnic minority and / or religion were marked as identity hate.

Dataset Curators

Published by Dimitris Papaevagelou (Civic Information Office), Ioanna Archontaki (Civic Information Office), Stefanos Loukopoulos (VouliWatch), Maria Nathanail (VouliWatch) and Konstantinos Mentzelos (VouliWatch).

Annotation Process

VouliWatch, a Greek non-profit and non-partisan parliamentary monitoring organisation, currated the annotation process with 15 annotators that marked 112.000 tweets, resulting this dataset.

Citation

@misc {civic_information_office_2023,
    author       = { {Civic Information Office} },
    title        = { toxic-el (Revision 65f60da) },
    year         = 2023,
    url          = { https://huggingface.co/datasets/cvcio/toxic-el },
    doi          = { 10.57967/hf/0744 },
    publisher    = { Hugging Face }
}

Authors

Dimitris Papaevagelou (Civic Information Office) - @andefined

About

Civic Information Office is a Non Profit Organization based in Athens, Greece focusing on creating technology and research products for the public interest.

VouliWatch is a Non Profit parliamentary monitoring and transparency watchdog organisation that promotes political integrity, engages Greek citizens with legislative politics and grants them with the opportunity to communicate, evaluate and hold elected representatives in the Greek and the European Parliament accountable.