| --- |
| license: mit |
| language: |
| - en |
| tags: |
| - nlp |
| - text-classification |
| - scam-detection |
| - crypto |
| - pidgin |
| - low-resource-languages |
| - africa |
| pretty_name: pidginshield |
| size_categories: |
| - n<1K |
| --- |
| |
| # pidginshield: pidgin crypto scam detection dataset |
|
|
| ## overview |
| this dataset is designed to detect crypto scams in nigerian pidgin, english, and mixed-language communication. |
|
|
| it captures realistic scam patterns commonly found in west african crypto communities, where informal language and slang are widely used. |
|
|
| ## dataset structure |
| each entry contains: |
| - text |
| - language_type |
| - label (scam or legit) |
| - scam_category |
| - intent |
| - explanation |
|
|
| ## use cases |
| - training scam detection models |
| - improving ai safety in fintech and web3 |
| - moderation tools for crypto communities |
|
|
| ## attribution |
| this dataset, pidginshield, was created for the uncharted data challenge by adaption labs. |
|
|
| it leverages adaptive data, adaption’s data infrastructure for building and evolving datasets. |
