Definition
A metric used in RAG evaluation that measures the ratio of relevant information to irrelevant noise within retrieved context chunks, specifically focusing on whether the most pertinent data is ranked at the top. High precision reduces LLM hallucinations but often necessitates architectural trade-offs such as increased latency from secondary Reranking stages.
Distinguishes 'signal quality' within retrieved chunks from 'Context Recall', which measures if the answer exists at all.
"A gold prospector's pan that successfully filters out silt and gravel, leaving only high-value nuggets at the surface."
- Context Recall(Counterpart)
- Reranking(Component)
- Mean Reciprocal Rank (MRR)(Prerequisite)
- Faithfulness(Downstream metric)
Conceptual Overview
A metric used in RAG evaluation that measures the ratio of relevant information to irrelevant noise within retrieved context chunks, specifically focusing on whether the most pertinent data is ranked at the top. High precision reduces LLM hallucinations but often necessitates architectural trade-offs such as increased latency from secondary Reranking stages.
Disambiguation
Distinguishes 'signal quality' within retrieved chunks from 'Context Recall', which measures if the answer exists at all.
Visual Analog
A gold prospector's pan that successfully filters out silt and gravel, leaving only high-value nuggets at the surface.