Question

A team builds a foundation-model application that extracts specific data fields (such as invoice numbers and totals) from scanned business documents. Each extraction is either present-and-correct or not, and the team wants a single automated metric that balances how many correct fields the model finds against how many of its extractions are actually right. Which evaluation metric best fits this need?

Accepted Answer

F1 score, which combines precision and recall into a single balanced measure

Answer

BLEU score, which measures n-gram overlap between generated and reference text

Answer

Perplexity, which measures how confidently the model predicts the next token

Answer

ROUGE score, which measures recall-oriented overlap for summaries