Korean Toxicity Dataset

When this dataset fits

Korean Toxicity Dataset is a Korean Toxicity Dataset from GitHub. It fits the korean toxicity dataset task. Do not mix it with a benchmark download. License label MIT. Size text is “소용량”, copied from the host.

Open 2 catalog excerpts →

Task / category

Text · Text · classification

Language

Korean

Format · Size

Text · Small (5.8K sentences)

License

MIT(Catalog verified) · License document

Source & original link

GitHub · Original dataset link(Verified)

Verification status

Source: Verified · License: Verified · Last checked: 2026-09-03

Korean Toxicity Dataset: Korean Toxicity Dataset from GitHub. License MIT. Open the original link to download.

Korean Toxicity Dataset (2021). GitHub. license label: MIT catalog commercial mark: allowed Original: https://github.com/2runo/Curse-detection-data. GearDel catalog: https://geardel.com/en/datasets/curse-detection-data. Last checked: 2026-09-03. Author names are not stored in the GearDel catalog and are not invented here.

Differences below use catalog fields only: license, commercial mark, provider, size text, and year.

• HateScoreKorean Toxicity Dataset · GitHubViewLicense labels differ (MIT vs Apache 2.0). Size text differs (Small vs Medium). Those strings are copied from the hosts. Catalog years differ (2021 vs 2022).
• Korean NLP DatasetKorean NLP Dataset · GitHubViewLicense labels differ (MIT vs Apache 2.0). Size text differs (Small vs Medium). Those strings are copied from the hosts. Catalog years differ (2021 vs 2023).
• Smilegate UnSmileKorean NLP Dataset · GitHubViewLicense labels differ (MIT vs CC BY-NC-ND 4.0). Commercial marks differ (Commercial use allowed (catalog) vs Commercial use may be conditional).
• SMSKorean NLP Dataset · GitHubViewCatalog years differ (2021 vs 2022).

EXCERPTS

Catalog excerpts

2 stored samples. GearDel does not host the original files.

Open this row in Try samples →

이딴 식으로 장사하면 망하는 게 당연하지 한심하다

1 (공격성 및 욕설 포함)