metadata

license: cc-by-sa-4.0
task_categories:
  - text-classification
language:
  - en
tags:
  - disaggregated
  - perspectivism
  - hate speech
  - hs
  - offensiveness
  - aggressiveness
  - stereotype
  - immigration
  - xenophobia
  - Brexit
  - islamophobia
pretty_name: BREXIT
size_categories:
  - 1K<n<10K

Dataset Card for Dataset Name

The dataset contains 1120 tweets related to immigration, racism, islamophobia, and xenophobia in the context of the BREXIT online discussion.

Each tweet has been annotated by 6 annotators, 3 of which belong to the group targeted by the discriminatory content (immigrants and Muslim annotators living in the UK), and 3 of which are part of a "control" group, not directly targeted by the discriminatory content.

We release the dataset in a disaggregated manner (each line corresponds to the annotation of a single annotator).

Dataset Details

Dataset Description

Language(s) (NLP): en
License: cc-by-sa-4.0

Uses

Direct Use

Possible uses of the dataset include:

Training and testing of NLP and ML systems for the automatic classification of hate speech, aggressive, offensive, and stereotypical content
Training and testing of NLP and ML systems learning from disaggregated data
Annotators' disagreement and polarization analysis
Analysis of hate speech, aggressive, offensive, and stereotypical content

Out-of-Scope Use

The dataset is not intended to generate offensive or discriminatory content or similar misuses.

Dataset Structure

Each row in the dataset corresponds to an annotations. The provided fields are the following:

tweet: The text of the tweet
instance_id: the ID of the tweet, unique to each tweet
annotator_group: target or control. Target annotators are Muslim immigrants living in the UK
annotator_id: the id of the annotator
hs: whether or not the tweet contains hate speech according to the annotator
offensiveness: whether or not the tweet is offensive according to the annotator
stereotype: whether or not the tweet contains a stereotype according to the annotator
aggressiveness: whether or not the tweet is aggressive, according to the annotator

The guidelines given to the annotators will be made public.

Dataset Creation

Curation Rationale

The dataset is created to better study consistent disagreement among annotators, specifically in the context when a group of annotators is targeted by discriminatory content.

We observe systematic disagreement and polarization between the target and the control group.

Source Data

The data has been downloaded from Twitter, using the #Brexit hashtag filtering by using a set of immigration, islamophobia, and xenophobia keywords.

Data Collection and Processing

[More Information Needed]

Annotations

Annotation process

The dataset has been annotated by six annotators, 3 of which belong to the group targeted by the discriminatory content. Each annotator provided a single binary label for Hate Speech, Offensivness, Aggressivness and Stereotype.

Who are the annotators?

Three of the six annotators are Muslim immigrants living in the UK at the time of Brexit and thus targeted by the discriminatory content. Three annotators are a control group and are not directly targeted by the discriminatory content.

Personal and Sensitive Information

The data is anonymized, and direct user mentions have been substituted by the "" token.

Bias, Risks, and Limitations

The dataset contains derogatory content, including racist and Islamophobic slurs.

Citation

BibTeX:

[More Information Needed]