Recall measures how many actual positives a system correctly identifies. It emphasizes coverage: a classifier with high recall misses fewer members of a target class, while a search system with high recall retrieves more of the relevant documents available. section 4.2.3 section 7.2.3
Use recall differently depending on what the system must avoid missing:
- For classification, increasing recall captures more true members of a class but can wrongly assign members of other classes to it. section 4.2.3
- For search, recall is the share of all relevant documents that the system retrieves; low recall means potentially important information is missing. section 7.2.3
- Balance recall against precision when false positives or missed positives carry different product consequences. section 4.2.3
Search example
If a topic has 15 relevant documents but the search system returns only 8, its recall is low because it misses 7 relevant documents. section 7.2.3
