Precision and Recall, explained.
Precision is the relevant share of retrieved items; recall is the retrieved share of all relevant items. F1 is their harmonic mean when defined.
Why it happens
Twenty ranked documents contain six relevant items at ranks 1,3,4,7,10,15. Choose a top-k prefix and reveal labels. Precision=TP/k, recall=TP/6 and F1=2TP/(k+6). Empty retrieval has undefined precision and F1 set to zero.
Precision is the relevant share of retrieved items; recall is the retrieved share of all relevant items. F1 is their harmonic mean when defined.
Read the result
Follow the same ranked prefix as k grows. Recall cannot fall here, while precision may rise or fall as each newly retrieved document changes the mix.
A worked example
A search that needs five useful documents
Open the top five results and reveal labels. Three are relevant, so precision is sixty percent and recall is fifty percent.
Expand to ten results. Five relevant documents are now included, giving fifty-percent precision and about eighty-three-percent recall.
Compare the extra reading cost with the recovered material. Ten is the smallest prefix reaching the activity's five-document goal.
OPTIONAL DEEPER DETAILGo deeper: inside the model
Inside this model
Twenty ranked documents contain six relevant items at ranks 1,3,4,7,10,15. Choose a top-k prefix and reveal labels. Precision=TP/k, recall=TP/6 and F1=2TP/(k+6). Empty retrieval has undefined precision and F1 set to zero.
Where this idea is useful
A practical use
A search tool may retrieve many relevant documents while burying them among irrelevant ones; a narrow answer can be precise but incomplete.
A common misconception
“Higher recall always means a better search experience.”
Recall ignores irrelevant material the reader must inspect. Whether a wider result set is worthwhile depends on the cost of misses and unnecessary reading.
What this explanation leaves out
- Binary relevance is stipulated and known for the full list. Real relevance can be graded and incomplete. No single metric encodes every cost of missed or irrelevant material.
Why is precision undefined for zero results?
Precision divides relevant retrieved documents by all retrieved documents. With no documents the denominator is zero; the interface reports that explicitly rather than claiming perfect precision.
What matters more in your search: missing one useful item or reading several irrelevant ones?
Associated thinkers
Associations marked provisional are awaiting source review.
Further reading
Explore the original research or the teaching reference behind this experiment.