---
pretty_name: SEA Toxicity Detection
license:
- cc-by-nc-sa-4.0
- cc-by-nc-3.0
- cc-by-nc-4.0
task_categories:
- text-generation
- text-classification
language:
- id
- th
- vi
dataset_info:
- config_name: id
  features:
  - name: label
    dtype: string
  - name: prompts
    list:
    - name: text
      dtype: string
  - name: metadata
    struct:
    - name: Abusive
      dtype: int64
    - name: HS
      dtype: int64
    - name: HS_Gender
      dtype: int64
    - name: HS_Group
      dtype: int64
    - name: HS_Individual
      dtype: int64
    - name: HS_Moderate
      dtype: int64
    - name: HS_Other
      dtype: int64
    - name: HS_Physical
      dtype: int64
    - name: HS_Race
      dtype: int64
    - name: HS_Religion
      dtype: int64
    - name: HS_Strong
      dtype: int64
    - name: HS_Weak
      dtype: int64
    - name: language
      dtype: string
  - name: id
    dtype: string
  splits:
  - name: eval
    num_bytes: 242657
    num_examples: 1000
  - name: examples
    num_bytes: 1215
    num_examples: 5
  download_size: 106737
  dataset_size: 243872
- config_name: ms
  features:
  - name: id
    dtype: string
  - name: label
    dtype: string
  - name: prompts
    list:
    - name: text
      dtype: string
  - name: metadata
    struct:
    - name: language
      dtype: string
  splits:
  - name: eval
    num_bytes: 146757
    num_examples: 1000
  - name: examples
    num_bytes: 827
    num_examples: 5
  download_size: 96716
  dataset_size: 147584
- config_name: my
  features:
  - name: id
    dtype: string
  - name: label
    dtype: string
  - name: prompts
    list:
    - name: text
      dtype: string
  - name: metadata
    struct:
    - name: language
      dtype: string
  splits:
  - name: eval
    num_bytes: 110107
    num_examples: 400
  - name: examples
    num_bytes: 1421
    num_examples: 5
  download_size: 41320
  dataset_size: 111528
- config_name: th
  features:
  - name: label
    dtype: string
  - name: prompts
    list:
    - name: text
      dtype: string
  - name: metadata
    struct:
    - name: language
      dtype: string
    - name: nontoxic_votes
      dtype: int64
    - name: toxic_votes
      dtype: int64
    - name: tweet_id
      dtype: int64
  - name: id
    dtype: string
  splits:
  - name: eval
    num_bytes: 274301
    num_examples: 1000
  - name: examples
    num_bytes: 1058
    num_examples: 5
  download_size: 126955
  dataset_size: 275359
- config_name: tl
  features:
  - name: id
    dtype: string
  - name: label
    dtype: string
  - name: prompts
    list:
    - name: text
      dtype: string
  - name: metadata
    struct:
    - name: language
      dtype: string
  splits:
  - name: eval
    num_bytes: 51721
    num_examples: 400
  - name: examples
    num_bytes: 660
    num_examples: 5
  download_size: 35427
  dataset_size: 52381
- config_name: vi
  features:
  - name: label
    dtype: string
  - name: prompts
    list:
    - name: text
      dtype: string
  - name: metadata
    struct:
    - name: language
      dtype: string
  - name: id
    dtype: string
  splits:
  - name: eval
    num_bytes: 96345
    num_examples: 1000
  - name: examples
    num_bytes: 539
    num_examples: 5
  download_size: 58271
  dataset_size: 96884
configs:
- config_name: id
  data_files:
  - split: eval
    path: id/eval-*
  - split: examples
    path: id/examples-*
- config_name: ms
  data_files:
  - split: eval
    path: ms/eval-*
  - split: examples
    path: ms/examples-*
- config_name: my
  data_files:
  - split: eval
    path: my/eval-*
  - split: examples
    path: my/examples-*
- config_name: th
  data_files:
  - split: eval
    path: th/eval-*
  - split: examples
    path: th/examples-*
- config_name: tl
  data_files:
  - split: eval
    path: tl/eval-*
  - split: examples
    path: tl/examples-*
- config_name: vi
  data_files:
  - split: eval
    path: vi/eval-*
  - split: examples
    path: vi/examples-*
size_categories:
- 1K<n<10K
---

# SEA Toxicity Detection
SEA Toxicity Detection evaluates a model's ability to identify toxic content such as hate speech and abusive language in text.  It is sampled from [MLHSD](https://aclanthology.org/W19-3506/) for Indonesian, [TTD](http://lrec-conf.org/workshops/lrec2018/W32/pdf/1_W32.pdf) for Thai, and [ViHSD](https://link.springer.com/chapter/10.1007/978-3-030-79457-6_35) for Vietnamese.

### Supported Tasks and Leaderboards
SEA Toxicity Detection is designed for evaluating chat or instruction-tuned large language models (LLMs). It is part of the [SEA-HELM](https://leaderboard.sea-lion.ai/) leaderboard from [AI Singapore](https://aisingapore.org/).

### Languages
- Indonesian (id)
- Thai (th)
- Vietnamese (vi)

### Dataset Details
SEA Toxicity Detection is split by language, with additional splits containing fewshot examples. Below are the statistics for this dataset. The number of tokens only refer to the strings of text found within the `prompts` column.

| Split | # of examples | # of GPT-4o tokens | # of Gemma 2 tokens | # of Llama 3 tokens |
|-|:-|:-|:-|:-|
| id | 1000 | 34416 | 34238 | 40537
| th | 1000 | 38189 | 35980 | 42901
| vi | 1000 | 17540 | 16904 | 18287
| id_fewshot | 5 | 183 | 174 | 216
| th_fewshot | 5 | 130 | 121 | 150
| vi_fewshot | 5 | 104 | 97 | 104
| **total** | 3015 | 90562 | 87514 | 102195 |

### Data Sources

| Data Source | License | Language/s | Split/s
|-|:-|:-| :-|
| [MLHSD](https://github.com/okkyibrohim/id-multi-label-hate-speech-and-abusive-language-detection) | [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/) | Indonesian | id, id_fewshot
| [TTD](https://huggingface.co/datasets/tmu-nlp/thai_toxicity_tweet) | [CC BY-NC 3.0](https://creativecommons.org/licenses/by-nc/3.0/) | Thai |th, th_fewshot
| [ViHSD](https://github.com/sonlam1102/vihsd) | [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) | Vietnamese | vi, vi_fewshot

### License

For the license/s of the dataset/s, please refer to the data sources table above.

We endeavor to ensure data used is permissible and have chosen datasets from creators who have processes to exclude copyrighted or disputed data. 


## Acknowledgement

This project is supported by the National Research Foundation Singapore and Infocomm Media Development Authority (IMDA), 
Singapore under its National Large Language Model Funding Initiative.

### References

```bibtex
@inproceedings{ibrohim-budi-2019-multi,
    title = "Multi-label Hate Speech and Abusive Language Detection in {I}ndonesian {T}witter",
    author = "Ibrohim, Muhammad Okky  and
      Budi, Indra",
    editor = "Roberts, Sarah T.  and
      Tetreault, Joel  and
      Prabhakaran, Vinodkumar  and
      Waseem, Zeerak",
    booktitle = "Proceedings of the Third Workshop on Abusive Language Online",
    month = aug,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/W19-3506",
    doi = "10.18653/v1/W19-3506",
    pages = "46--57",
}

@inproceedings{sirihattasak2018annotation,
  title={Annotation and classification of toxicity for Thai Twitter},
  author={Sirihattasak, Sugan and Komachi, Mamoru and Ishikawa, Hiroshi},
  booktitle={TA-COS 2018: 2nd Workshop on Text Analytics for Cybersecurity and Online Safety},
  pages={1},
  year={2018}
}

@InProceedings{10.1007/978-3-030-79457-6_35,
      author="Luu, Son T.
      and Nguyen, Kiet Van
      and Nguyen, Ngan Luu-Thuy",
      editor="Fujita, Hamido
      and Selamat, Ali
      and Lin, Jerry Chun-Wei
      and Ali, Moonis",
      title="A Large-Scale Dataset for Hate Speech Detection on Vietnamese Social Media Texts",
      booktitle="Advances and Trends in Artificial Intelligence. Artificial Intelligence Practices",
      year="2021",
      publisher="Springer International Publishing",
      address="Cham",
      pages="415--426",
      isbn="978-3-030-79457-6"
}

@misc{leong2023bhasaholisticsoutheastasian,
      title={BHASA: A Holistic Southeast Asian Linguistic and Cultural Evaluation Suite for Large Language Models}, 
      author={Wei Qi Leong and Jian Gang Ngui and Yosephine Susanto and Hamsawardhini Rengarajan and Kengatharaiyer Sarveswaran and William Chandra Tjhi},
      year={2023},
      eprint={2309.06085},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2309.06085}, 
}
```