Definition
A retrieval evaluation metric that measures the proportion of relevant documents successfully captured within the top K results returned by a vector database or search engine. In RAG pipelines, it serves as a benchmark for the retriever's ability to locate the necessary ground-truth context before the generation phase.
Focuses on 'finding all relevant items' rather than 'how many retrieved items are relevant'.
"A fisherman's net: a wider net (higher K) is more likely to catch all the target fish (relevant data), but also gathers more unwanted seaweed (noise)."
Conceptual Overview
A retrieval evaluation metric that measures the proportion of relevant documents successfully captured within the top K results returned by a vector database or search engine. In RAG pipelines, it serves as a benchmark for the retriever's ability to locate the necessary ground-truth context before the generation phase.
Disambiguation
Focuses on 'finding all relevant items' rather than 'how many retrieved items are relevant'.
Visual Analog
A fisherman's net: a wider net (higher K) is more likely to catch all the target fish (relevant data), but also gathers more unwanted seaweed (noise).