AI Glossary Term
What does “Retrieval Precision” mean?
Definition
The proportion of retrieved items that are relevant to the user's query.
How Retrieval Precision works
Example
A search step returns ten passages for a question, and seven are actually relevant. Retrieval precision is 70 percent. The three irrelevant passages can distract the model.
Why it matters
Irrelevant passages waste context space and can lead to wrong answers. Precision is one of the main checks on retrieval quality.
Common questions
Retrieval precision vs recall?
Precision measures how many returned items are relevant; recall measures how many of the relevant items were found.
How do I improve it?
Better embeddings, reranking, smaller chunk sizes, or filters on metadata.
Do I need labeled data to measure it?
Yes, someone has to mark which results are relevant for a set of test queries.