Hugging Face's logo

Datasets:
brighter-dataset
/
BRIGHTER-emotion-categories

Modalities:
Tabular
Text
Formats:
parquet
Languages:
Afrikaans
Arabic
German
Size:
100K - 1M
ArXiv:
Libraries:
Datasets
pandas
License:
Dataset card Data Studio Files Community
1
BRIGHTER-emotion-categories / README.md
vgaraujov's picture
vgaraujov
Upload README.md with huggingface_hub
1e05a11 verified
preview code
|
raw
history blame
6.09 kB
metadata
language:
  - af
  - ar
  - de
  - en
  - es
  - ha
  - hi
  - id
  - ig
  - jv
  - mr
  - pcm
  - pt
  - ro
  - ru
  - rw
  - su
  - sv
  - sw
  - tt
  - uk
  - vmw
  - xh
  - yo
  - zh
  - zu
license: cc-by-4.0

BRIGHTER Emotion Categories Dataset

This dataset contains the emotion categories data from the BRIGHTER paper: BRIdging the Gap in Human-Annotated Textual Emotion Recognition Datasets for 28 Languages.

Dataset Description

The BRIGHTER Emotion Categories dataset is a comprehensive multi-language, multi-label emotion classification dataset with separate configurations for each language. It represents one of the largest human-annotated emotion datasets across multiple languages.

  • Total languages: 28 languages
  • Total examples: 139595
  • Splits: train, dev, test

About BRIGHTER

BRIGHTER addresses the gap in human-annotated textual emotion recognition datasets for low-resource languages. While most existing emotion datasets focus on English, BRIGHTER covers multiple languages, including many low-resource ones. The dataset was created by selecting text from various sources and having annotators label six categorical emotions: anger, disgust, fear, joy, sadness, and surprise.

The dataset contains text in the following languages: Afrikaans, Algerian Arabic, Moroccan Arabic, Mandarin Chinese, German, English, Spanish (Ecuador, Colombia, Mexico), Hausa, Hindi, Igbo, Indonesian, Javanese, Kinyarwanda, Marathi, Nigerian Pidgin, Portuguese (Brazil), Portuguese (Mozambique), Romanian, Russian, Sundanese, Swahili, Swedish, Tatar, Ukrainian, Makhuwa, Xhosa, Yoruba, and Zulu.

Language Configurations

Each language is available as a separate configuration with the following statistics:

Original Code ISO Code Train Examples Dev Examples Test Examples Total
afr af 1222 196 2130 3548
arq ar 901 200 1804 2905
ary ar 1608 534 1624 3766
chn zh 2642 400 5284 8326
deu de 2603 400 5208 8211
eng en 2768 232 5534 8534
esp es 1996 368 3390 5754
hau ha 2145 712 2160 5017
hin hi 2556 200 2020 4776
ibo ig 2880 958 2888 6726
ind id 0 156 851 1007
jav jv 0 151 837 988
kin rw 2451 814 2462 5727
mar mr 2415 200 2000 4615
pcm pcm 3728 1240 3740 8708
ptbr pt 2226 400 4452 7078
ptmz pt 1546 514 1552 3612
ron ro 1241 246 2238 3725
rus ru 2679 398 2000 5077
sun su 924 398 1852 3174
swa sw 3307 1102 3312 7721
swe sv 1187 400 2376 3963
tat tt 1000 400 2000 3400
ukr uk 2466 498 4468 7432
vmw vmw 1551 516 1554 3621
xho xh 0 682 1594 2276
yor yo 2992 994 3000 6986
zul zu 0 875 2047 2922

Features

  • id: Unique identifier for each example
  • text: Text content to classify
  • anger, disgust, fear, joy, sadness, surprise: Presence of emotion
  • emotions: List of emotions present in the text

Dataset Characteristics

This dataset provides binary labels for emotion presence, making it suitable for multi-label classification tasks. For regression tasks or fine-grained emotion analysis, please see the companion BRIGHTER-emotion-intensities dataset.

Usage 

from datasets import load_dataset

# Load all data for a specific language
eng_dataset = load_dataset("YOUR_USERNAME/BRIGHTER-emotion-categories", "eng")

# Or load a specific split for a language
eng_train = load_dataset("YOUR_USERNAME/BRIGHTER-emotion-categories", "eng", split="train")

Citation

If you use this dataset, please cite the following papers:

@misc{muhammad2025brighterbridginggaphumanannotated,
      title={BRIGHTER: BRIdging the Gap in Human-Annotated Textual Emotion Recognition Datasets for 28 Languages}, 
      author={Shamsuddeen Hassan Muhammad and Nedjma Ousidhoum and Idris Abdulmumin and Jan Philip Wahle and Terry Ruas and Meriem Beloucif and Christine de Kock and Nirmal Surange and Daniela Teodorescu and Ibrahim Said Ahmad and David Ifeoluwa Adelani and Alham Fikri Aji and Felermino D. M. A. Ali and Ilseyar Alimova and Vladimir Araujo and Nikolay Babakov and Naomi Baes and Ana-Maria Bucur and Andiswa Bukula and Guanqun Cao and Rodrigo Tufiño and Rendi Chevi and Chiamaka Ijeoma Chukwuneke and Alexandra Ciobotaru and Daryna Dementieva and Murja Sani Gadanya and Robert Geislinger and Bela Gipp and Oumaima Hourrane and Oana Ignat and Falalu Ibrahim Lawan and Rooweither Mabuya and Rahmad Mahendra and Vukosi Marivate and Andrew Piper and Alexander Panchenko and Charles Henrique Porto Ferreira and Vitaly Protasov and Samuel Rutunda and Manish Shrivastava and Aura Cristina Udrea and Lilian Diana Awuor Wanzare and Sophie Wu and Florian Valentin Wunderlich and Hanif Muhammad Zhafran and Tianhui Zhang and Yi Zhou and Saif M. Mohammad},
      year={2025},
      eprint={2502.11926},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.11926}, 
}

@misc{muhammad2025semeval2025task11bridging,
      title={SemEval-2025 Task 11: Bridging the Gap in Text-Based Emotion Detection}, 
      author={Shamsuddeen Hassan Muhammad and Nedjma Ousidhoum and Idris Abdulmumin and Seid Muhie Yimam and Jan Philip Wahle and Terry Ruas and Meriem Beloucif and Christine De Kock and Tadesse Destaw Belay and Ibrahim Said Ahmad and Nirmal Surange and Daniela Teodorescu and David Ifeoluwa Adelani and Alham Fikri Aji and Felermino Ali and Vladimir Araujo and Abinew Ali Ayele and Oana Ignat and Alexander Panchenko and Yi Zhou and Saif M. Mohammad},
      year={2025},
      eprint={2503.07269},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2503.07269}, 
}

License

This dataset is licensed under CC-BY 4.0.