README.md

metadata

language:
  - it
language_details: it-IT
license: cc-by-nc-sa-4.0
task_ids:
  - hate-speech-detection
task_categories:
  - text-classification
configs:
  - config_name: default
    data_files:
      - split: dev
        path: hs_dev.jsonl
      - split: test_all
        path: hs_test_all.jsonl
      - split: test_news
        path: hs_test_news.jsonl
      - split: test_tweets
        path: hs_test_tweets.jsonl
size_categories:
  - 1K<n<10K

HaSpeeDe2

The HaSpeeDe2 dataset collects 8,012 tweets and 500 news headlines annotated for the presence of hate speech, stereotypes and nominal utterance.

The dataset has been used in the context of the HaSpeeDe task (http://www.di.unito.it/~tutreeb/haspeede-evalita20/index.html), organized as part of the EVALITA 2020 evaluation campaign (http://www.evalita.it/2020).

In order to meet the GDPR requirements, texts have been pseudonymized replacing all original IDs in both datasets with newly-generated ones. Mentions, emails, person names (excluded public person names), and phone numbers have been masked with, respectively, the labels MENTION, EMAIL, PERSON, PHONE, followed by a number to distinguish between different entities of the same kind within the same text.

Original files available here: https://live.european-language-grid.eu/catalogue/corpus/7498/overview/

Available splits:

• dev: development set
• test_news: news test set
• test_tweets: tweets test set
• test_all: combines test_news and test_tweets

If you find this dataset useful please cite:

@article{sanguinetti2020haspeede,
  title={Haspeede 2@ evalita2020: Overview of the evalita 2020 hate speech detection task},
  author={Sanguinetti, Manuela and Comandini, Gloria and Di Nuovo, Elisa and Frenda, Simona and Stranisci, Marco and Bosco, Cristina and Caselli, Tommaso and Patti, Viviana and Russo, Irene},
  journal={Evaluation Campaign of Natural Language Processing and Speech Tools for Italian},
  year={2020},
  publisher={CEUR Workshop Proceedings (CEUR-WS. org)}
}