← All questions
EvaluationMedium· MRR· Ranking· Evaluation Strategy

Your retrieval system’s MRR climbed from 0.62 to 0.81 a...

MRR is a limited metric because it only considers the position of the first relevant document. An improvement in MRR does not guarantee an improvement in overall system quality. If your relevance judgments are graded, you should also evaluate using nDCG@k to ensure that the quality of the top-ranked documents is actually improving, rather than just shifting the first hit to a slightly better position while degrading the rest of the list.

Code playground (Python)