• For each query, the model gives you a list of proposed results
  • The reciprocal rank is the reciprocal of the rank of the first valid result
  • MRR is the mean of the Reciprocal Rank of the queries:
QueryProposed ResultsCorrect responseRankReciprocal rank
catcatten, cati, catscats31/3
torustorii, tori, torusestori21/2
virusviruses, virii, viriviruses11
MRR = (1/3 + 1/2 + 1)/3 = 11/18
  • Note: If the model returns multiple correct answers for a query, it only cares about the rank of the first correct answer