Recall (AI metric)
The share of all correct items that the system actually found: of everything it should have caught, how much it did.
Recall is true positives divided by all items that really were positive. A fraud detector with high recall misses few frauds, even if it raises some false alarms. In RAG, recall@k asks whether the passages needed for the answer are among the first k retrieved; if they are not, the model cannot use them. Recall is measured against a labeled set, because you have to know what should have been found.
Related terms
- Precision (AI metric): The share of returned results that are actually correct: of everything the system flagged or retrieved, how much was right.
- F1 score: One number that combines precision and recall as their harmonic mean, so it is high only when both are high.
- Confusion matrix: A table that counts a classifier’s right and wrong decisions by type: true and false positives, true and false negatives.
- Context recall: A RAG retrieval metric: how much of the information needed for the correct answer was actually retrieved.
- Top-k (retrieval): The number of highest-ranked search results passed on to the next step; in RAG, how many passages go into the prompt.
