README.md

---
license: mit
task_categories:
- text-classification
language:
- ro
pretty_name: REDv2
size_categories:
- 1K<n<10K
---

# Dataset Card for [REDv2]

## Table of Contents
- [Table of Contents](#table-of-contents)
- [Dataset Description](#dataset-description)
  - [Dataset Summary](#dataset-summary)
  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
  - [Languages](#languages)
- [Dataset Structure](#dataset-structure)
  - [Data Instances](#data-instances)
  - [Data Fields](#data-fields)
  - [Data Splits](#data-splits)
- [Dataset Creation](#dataset-creation)
  - [Curation Rationale](#curation-rationale)
  - [Source Data](#source-data)
  - [Annotations](#annotations)
  - [Personal and Sensitive Information](#personal-and-sensitive-information)
- [Considerations for Using the Data](#considerations-for-using-the-data)
  - [Social Impact of Dataset](#social-impact-of-dataset)
  - [Discussion of Biases](#discussion-of-biases)
  - [Other Known Limitations](#other-known-limitations)
- [Additional Information](#additional-information)
  - [Dataset Curators](#dataset-curators)
  - [Licensing Information](#licensing-information)
  - [Citation Information](#citation-information)
  - [Contributions](#contributions)

## Dataset Description

- **Homepage:**
- **Repository:**
- **Paper:**
- **Leaderboard:**
- **Point of Contact:**

### Dataset Summary

This is the second version of the Romanian Emotions Dataset (RED) containing 5449 tweets annotated in a multi-label fashion with the following 7 emotions: Anger (Furie), Fear (Frică), Joy (Bucurie), Sadness (Tristețe), Surprise (Surpriză), Trust (Încredere) and Neutral (Neutru).

### Supported Tasks and Leaderboards

This dataset is intended for multi-class & multi-label emotion classification.

### Languages

The data is in Romanian.

## Dataset Structure

### Data Instances

Each instance is a tweet with a corresponding ID and one or more emotion annotations (or neutral).

### Data Fields

The simplified configuration includes:
```
    text: the tweet
    text_id: unique identifier of the tweet (can be used to look up the entry in the raw dataset)
    agreed_labels: the agreed emotion annotations vector (each value of 1 means that at least two annotators recognized that specific emotion)
    procentual_labels: vector containing three values: 0.33 if one annotator recognised the emotion, 0.66 if two annotators agreed on the emotion, and 0.99 if all annotators recognised the emotion
```
In addition to the above, the raw data includes:
```
    Anger, Fear, Joy, Neutral, Sadness, Surprise, Trust: boolean values - True if the specific emotion is found in the agreed_labels vector
    annotator1, annotator2, annotator3: vectors of zeros of ones - 1 means the annotator recognized the emotion on the corresponding vector index
    sum_labels: the sum of annotator1, annotator2 and annotator3 vectors
```
The arrays of 7 values correspond to the following emotions: ['Sadness', 'Surprise', 'Fear', 'Anger', 'Neutral', 'Trust', 'Joy'].

### Data Splits

This dataset includes a set of train/val/test splits with 4088, 818, and 543 examples respectively.

## Dataset Creation

### Curation Rationale

From the paper introduction:

>Interpreting correctly one’s own emotions, as well as
other people’s emotional states, is a central aspect of
emotional intelligence. Today, people can automate
the process of emotion detection by creating machine
learning models, provided by the fact that the model
training was done on qualitative and sufficient data.
With the constant increase of social media usage there
is also an increase in online public data, freely available
for model creation. Thus, analyzing emotions in online
content naturally has became more and more of a topic
of interest in the recent years.

### Source Data

#### Initial Data Collection and Normalization

Data was collected from Twitter (for more information see Chapter 3.1 of the [paper](https://aclanthology.org/2022.lrec-1.149.pdf)).

#### Who are the source language producers?

Romanian-speaking Twitter users.

### Annotations

#### Annotation process

See Chapter 3.2. in the [paper](https://aclanthology.org/2022.lrec-1.149.pdf).

#### Who are the annotators?

Annotations were produced by 66 Cognitive Science students, University of Bucharest, Faculty of Psichology and Educational Sciences.

### Personal and Sensitive Information

All tweets in this dataset are anonymized by removing usernames and proper nouns.


## Additional Information

### Dataset Curators

Researchers at the University of Bucharest and Adobe (see the authors of the paper [here](http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.149.pdf)).

### Licensing Information

The [GitHub repository](https://github.com/Alegzandra/RED-Romanian-Emotions-Dataset/tree/main/REDv2) of this dataset has an MIT license.

### Citation Information

If you are using this dataset in your research, please cite:
```
@inproceedings{redv2,
  author = "Alexandra Ciobotaru and
               Mihai V. Constantinescu and
               Liviu P. Dinu and
               Stefan Daniel Dumitrescu",
  title = "{RED} v2: {E}nhancing {RED} {D}ataset for {M}ulti-{L}abel {E}motion {D}etection",
  journal = "Proceedings of the 13th Language Resources and Evaluation Conference (LREC 2022)",
  pages = "1392–1399",
  year = "2022",
  address = "Marseille, France",
  publisher = "European Language Resources Association (ELRA)",
  url = "http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.149.pdf",
  language = "English"
}
```
### Contributions

Thanks to [@Alegzandra](https://github.com/<Alegzandra>) for adding this dataset.