README.md

---
license: cc-by-4.0
annotations_creators:
- crowdsourced
language_creators:
- crowdsourced
language:
- pt
multilinguality:
- monolingual
size_categories:
- 10K<n<100K
source_datasets:
- original
task_categories:
- text-classification
task_ids: []
pretty_name: TuPy
language_bcp47:
- pt-BR
tags:
- hate-speech-detection
configs:
- config_name: multilabel
  data_files:
  - split: train
    path: multilabel/multilabel_train.csv
  - split: test
    path: multilabel/multilabel_test.csv
  default: true
  
- config_name: binary
  data_files:
  - split: train
    path: binary/binary_train.csv
  - split: test
    path: binary/binary_test.csv
---

# Portuguese Hate Speech Dataset (TuPy)

The Portuguese hate speech dataset (TuPy) is an annotated corpus designed to facilitate the development of advanced hate speech detection models using machine learning (ML) 
and natural language processing (NLP) techniques. TuPy is comprised of 10,000 (ten thousand) unpublished, annotated, and anonymized documents collected 
on Twitter (currently known as X) in 2023. 
This repository is organized as follows:

```sh
root.
    ├── binary     : binary dataset (including training and testing split)
    ├── multilabel : multilabel dataset (including training and testing split)
    └── README.md  : documentation and card metadata
```
## Security measures
To safeguard user identity and uphold the integrity of this dataset, all user mentions have been anonymized as "@user," and any references to external websites have been omitted

## Annotation and voting process
To generate the binary matrices, we employed a straightforward voting process. Three distinct evaluations were assigned to each document. In cases where a document received two or more identical classifications, the adopted value is set to 1; otherwise, it is marked as 0.Raw data can be checked into the repository in the [project repository](https://github.com/Silly-Machine/TuPy-Dataset)
The subsequent table provides a concise summary of the annotators' profiles and qualifications:

#### Table 1 – Annotators

| Annotator    | Gender | Education                                     | Political  | Color  |
|--------------|--------|-----------------------------------------------|------------|--------|
| Annotator 1  | Female | Ph.D. Candidate in civil engineering           | Far-left   | White  |
| Annotator 2  | Male   | Master's candidate in human rights             | Far-left   | Black  |
| Annotator 3  | Female | Master's degree in behavioral psychology       | Liberal    | White  |
| Annotator 4  | Male   | Master's degree in behavioral psychology       | Right-wing | Black  |
| Annotator 5  | Female | Ph.D. Candidate in behavioral psychology       | Liberal    | Black  |
| Annotator 6  | Male   | Ph.D. Candidate in linguistics                 | Far-left   | White  |
| Annotator 7  | Female | Ph.D. Candidate in civil engineering           | Liberal    | White  |
| Annotator 8  | Male   | Ph.D. Candidate in civil engineering           | Liberal    | Black  |
| Annotator 9  | Male   | Master's degree in behavioral psychology       | Far-left   | White  |

## Data structure
A data point comprises the tweet text (a string) along with thirteen categories, each category is assigned a value of 0 when there is an absence of aggressive or hateful content and a value of 1 when such content is present. These values represent the consensus of annotators regarding the presence of aggressive, hate, ageism, aporophobia, body shame, capacitism, lgbtphobia, political, racism, religious intolerance, misogyny, xenophobia, and others. An illustration from the multilabel TuPy dataset is depicted below:

```python
{
text: "e tem pobre de direita imbecil que ainda defendia a manutenção da política de preços atrelada ao dólar link",
aggressive: 1, hate: 1, ageism: 0, aporophobia: 1, body shame: 0, capacitism: 0, lgbtphobia: 0, political: 1, racism : 0,
religious intolerance : 0, misogyny : 0, xenophobia : 0, other : 0
}
```

# Dataset content

Table 2 provides a detailed breakdown of the dataset, delineating the volume of data based on the occurrence of aggressive speech and the manifestation of hate speech within the documents

#### Table 2 - Count of non-aggressive and aggressive documents

| Label                | Count  |
|----------------------|--------|
| Non-aggressive       | 8013   |
| Aggressive - Not hate| 689    |
| Aggressive - Hate    | 1298   |
| Total                | 10000  |

Table 3 provides a detailed analysis of the dataset, delineating the data volume in relation to the occurrence of distinct categories of hate speech.

#### Table 3 - Count of documents for hate categories

| Label                    | Count |
|--------------------------|-------|
| Ageism                   | 53    |
| Aporophobia              | 61    |
| Body shame               | 120   |
| Capacitism               | 92    |
| LGBTphobia               | 96    |
| Political                | 532   |
| Racism                   | 38    |
| Religious intolerance    | 28    |
| Misogyny                 | 207   |
| Xenophobia               | 70    |
| Other                    | 1     |
| Total                    | 1298  |

# Acknowledge
The TuPy project is the result of the development of Felipe Oliveira's thesis and the work of several collaborators. This project is financed by the Federal University of Rio de Janeiro ([UFRJ](https://ufrj.br/)) and the Alberto Luiz Coimbra Institute for Postgraduate Studies and Research in Engineering ([COPPE](https://coppe.ufrj.br/)).