Definition
The harmonic mean of precision and recall, used in RAG pipelines to quantitatively measure the accuracy of a model's response by balancing its ability to include all necessary information (recall) with its ability to exclude irrelevant or hallucinated content (precision).
Focuses on token-level or concept-level overlap between the RAG output and ground truth, rather than simple binary classification accuracy.
"A balanced scale where one side is 'The Whole Truth' (Recall) and the other is 'Nothing But The Truth' (Precision)."
- Precision(Component)
- Recall(Component)
- Exact Match (EM)(Rigid alternative metric)
- Faithfulness(RAG-specific quality it helps quantify)
Conceptual Overview
The harmonic mean of precision and recall, used in RAG pipelines to quantitatively measure the accuracy of a model's response by balancing its ability to include all necessary information (recall) with its ability to exclude irrelevant or hallucinated content (precision).
Disambiguation
Focuses on token-level or concept-level overlap between the RAG output and ground truth, rather than simple binary classification accuracy.
Visual Analog
A balanced scale where one side is 'The Whole Truth' (Recall) and the other is 'Nothing But The Truth' (Precision).