---
dataset_info:
  features:
  - name: text
    dtype: string
  - name: labels
    sequence: int64
  - name: labels_str
    sequence: string
  - name: labels_source
    sequence: string
  - name: source
    dtype: string
  splits:
  - name: train
    num_bytes: 63512185
    num_examples: 439361
  - name: validation
    num_bytes: 7914411
    num_examples: 54835
  - name: test
    num_bytes: 8492921
    num_examples: 58625
  download_size: 35356426
  dataset_size: 79919517
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
tags:
- text-classification
- emotion
- multi-label
- crowdsourced
license: cc-by-sa-4.0
task_categories:
- text-classification
language:
- en
pretty_name: Super Emotion
size_categories:
- 100K<n<1M
---

# Super Emotion Dataset

![banner](https://huggingface.co/datasets/cirimus/super-emotion/resolve/main/banner.png)

## Associated Paper
We provide full documentation of the dataset construction process in the accompanying paper: 📘 [The Super Emotion Dataset (PDF)](https://huggingface.co/datasets/cirimus/super-emotion/resolve/main/super-emotion.pdf)

## Dataset Summary
The **Super Emotion** dataset is a large-scale, multilabel dataset for emotion classification, aggregated from six prominent emotion datasets:


- [MELD](https://huggingface.co/datasets/declare-lab/MELD)
- [GoEmotions](https://huggingface.co/datasets/google-research-datasets/go_emotions)
- [TwitterEmotion](https://huggingface.co/datasets/dair-ai/emotion)
- [ISEAR](https://www.unige.ch/cisa/research/materials-and-online-research/research-material/)
- [SemEval](https://huggingface.co/datasets/SemEvalWorkshop/sem_eval_2018_task_1)
- [Crowdflower](tasksource/crowdflower)

It contains 552,821 unique text samples and 570,457 total emotion label assignments, reflecting its multilabel nature. Emotion labels are mapped into six primary emotions and a neutral class: joy, sadness, anger, fear, love, surprise, and neutral. Related or ambiguous categories like happiness, hate, or grief are merged accordingly.


## Supported Tasks
This dataset is designed for **emotion classification** and can be used for:
- Multi-label emotion recognition
- Emotion co-occurrence modeling
- Single-label classification
- Fine-tuning language models for affective NLP

## Dataset Structure
The dataset follows the structure:

| Column  | Type   | Description |
|---------|--------|-------------|
| text    | string | The raw input text |
| label   | string | One or more mapped emotion labels |
| source  | string | The original dataset name |

Splits:
- **Train**: 439361 samples
- **Validation**: 54835 samples
- **Test**: 58,625 samples

Class distribution by source:

| Source         | Neutr. | Surp. | Fear  | Sad.   | Joy    | Anger | Love  | Dropped | Total  |
| -------------- | ------ | ----- | ----- | ------ | ------ | ----- | ----- | ------- | ------ |
| MELD           | 6436   | 1636  | 358   | 1002   | 2308   | 1968  | 0     | 0       | 13708  |
| TwitterEmotion | 0      | 719   | 2373  | 5797   | 6761   | 2709  | 1641  | 0       | 20000  |
| ISEAR          | 0      | 14972 | 47712 | 121187 | 141067 | 57317 | 34554 | 0       | 416809 |
| GoEmotions     | 17772  | 4295  | 929   | 4032   | 7646   | 7838  | 15557 | 2723    | 60792  |
| Crowdflower    | 8817   | 2187  | 8457  | 5165   | 9270   | 1433  | 3842  | 827     | 39998  |
| SemEval        | 0      | 566   | 1848  | 3607   | 5065   | 4780  | 1757  | 1527    | 19150  |
| **All**        | 33025  | 24375 | 61677 | 140790 | 172117 | 76045 | 57351 | 5077    | 570457 |

## License

This dataset is distributed under the CC BY-SA 4.0 license, in compliance with the most restrictive upstream dataset (GoEmotions).

Each source dataset is redistributed under its original license:

- [GoEmotions](https://huggingface.co/datasets/go_emotions) — CC BY-SA 4.0
- [MELD](https://huggingface.co/datasets/meld) — CC BY 4.0
- [ISEAR](https://huggingface.co/datasets/isear) — License: research use only
- CARER
- CrowdFlower
- SemEval

Users are encouraged to consult individual dataset licenses before redistribution or commercial use.

## Citation
If you use this dataset, please cite both the original sources and this aggregation:

```
@article{junquedefortuny2025superemotion,
  title   = {The Super Emotion Dataset},
  author  = {Enric Junqu{\'e} de Fortuny},
  journal = {arXiv preprint},
  year    = {2025},
  url     = {https://arxiv.org/abs/2505.15348},
  note    = {Dataset and supplementary materials available at \url{https://huggingface.co/datasets/cirimus/super-emotion}}
}
```