Datasheet

Domain: Evaluation

🎯 Motivation

Documentation for a dataset: what it contains, how it was collected, limitations, and recommended uses.

πŸ“‹ Concrete Example

Datasheet: β€˜Dataset: 50K English news articles. Collected: 2024. Sources: 5 news sites. Limitations: English only, Western bias. Use: text classification training.’

πŸ”— Analogy

An ingredient label on packaged food - what is inside, where it came from, allergen warnings. A datasheet labels a dataset’s contents and limitations.