All terms
Evaluation
ROUGE
A text-overlap metric family used mainly to score summarization against references.
Definition
ROUGE is a family of automatic metrics for evaluating summarization and text generation by measuring overlap with one or more human-written reference texts. Common variants count shared n-grams (runs of consecutive words, as in ROUGE-N) or the longest run of words appearing in the same order in both (ROUGE-L). Like BLEU, it rewards matching the exact words rather than the meaning, so it can be gamed by outputs that echo reference wording and misses valid paraphrases.