Meteor Score

#nlp #deep-learning #interview

METEOR = Metric for Evaluation of Translation with Explicit Ordering
METEOR can be thought as the harmonic mean of BLEU Score and ROUGE-N Score
- More precisely harmonic mean of unigram precision and recall
- You can think of the beta value of 3 in F-Beta Score

METEOR Score

$P r e c i s i o n = \frac{# of unigrams matched}{# of unigrams in generation}$ $R e c a l l = \frac{# of unigrams matched}{# of unigrams in reference}$ $F_{m e a n} = \frac{10 * P r e c i s i o n * R e c a l l}{9 * P r e c i s i o n + R e c a l l}$ $P e n a l t y = 0.5 * (\frac{# of minimum chunks where chunks are matched}{# number of unigrams in the matched chunks})^{3}$ $M E T E O R = F_{m e a n s} * (1 - p e n a l t y)$

Related Notes

Group Normalization
Tanh
Contrastive Loss
ROUGE-N Score
Standardization