Definition
A retrieval evaluation metric that measures the extent to which the retrieved context contains the information required to reconstruct the ground truth answer. Optimizing for high contextual recall often involves a trade-off where a wider retrieval window increases the likelihood of finding the answer but introduces more noise and higher token costs.
Measures 'is the answer in the retrieved chunks?', whereas Context Precision measures 'how relevant/ranked are those chunks?'.
"A detective's evidence bag: Contextual Recall is the measure of whether the actual murder weapon is inside the bag, regardless of how many other random objects were also picked up."
- Ground Truth(Prerequisite)
- Context Precision(Complementary Metric)
- Faithfulness(Downstream Metric)
Conceptual Overview
A retrieval evaluation metric that measures the extent to which the retrieved context contains the information required to reconstruct the ground truth answer. Optimizing for high contextual recall often involves a trade-off where a wider retrieval window increases the likelihood of finding the answer but introduces more noise and higher token costs.
Disambiguation
Measures 'is the answer in the retrieved chunks?', whereas Context Precision measures 'how relevant/ranked are those chunks?'.
Visual Analog
A detective's evidence bag: Contextual Recall is the measure of whether the actual murder weapon is inside the bag, regardless of how many other random objects were also picked up.