Recall (Information Retrieval Metric)

Recall is a way of measuring how completely a system finds the items that are actually relevant to a query or task. It is computed by dividing the number of relevant items that the system returned by the total number of relevant items that exist in the collection, so a value of one means every relevant item was found. The metric tells you what proportion of the true positives were captured, irrespective of how many irrelevant items may also have been included.

The reason recall matters is that many applications care more about missing an important result than about presenting extra material. In medical literature searches, safety inspections, or legal document review, failing to retrieve a single critical item can have serious consequences, so designers aim for high recall even if it means the user sees more noise. Recall therefore complements precision, which measures how much of what is returned is actually useful.

You will see recall used whenever a system is asked to retrieve documents, images, products, or any other items from a large pool based on relevance. Search engines report recall when evaluating ranking algorithms, recommendation services use it to gauge how many suitable suggestions they generate, and machine‑learning classifiers often quote recall for classes that are rare but important. Because the definition depends only on counting relevant versus retrieved items, recall remains a stable concept across domains.