Explainable Extraction Performance Measure
(Redirected from Span-Explanation Quality Measure)
An Explainable Extraction Performance Measure is an extraction performance metric that is an interpretability metric evaluating both extraction accuracy and explanation quality in explainable span extraction systems.
- AKA: Interpretable Extraction Metric, Span-Explanation Quality Measure.
- Context:
- It can typically assess Explanation Correctness for span-decision links.
- It can typically measure Explanation Completeness across all extractions.
- It can typically evaluate Explanation Consistency between similar cases.
- It can typically quantify Human Understandability of provided explanations.
- It can typically detect Explanation Faithfulness to actual model behavior.
- ...
- It can often combine Extraction F1 with explanation scores.
- It can often incorporate User Study Results for quality assessment.
- It can often employ Automated Coherence Checks for explanation logic.
- It can often measure Explanation Granularity at multiple levels.
- ...
- It can range from being an Automatic Explainability Measure to being a Human-Evaluated Explainability Measure, depending on its assessment method.
- It can range from being a Binary Quality Measure to being a Scaled Quality Measure, depending on its scoring approach.
- ...
- It can evaluate Explainable Span Extraction Task performance.
- It can guide Algorithm Development for interpretable systems.
- It can support Model Selection in explainable AI applications.
- It can diagnose Explanation Weaknesses in extraction systems.
- ...
- Example(s):
- Span-Purpose Alignment Score, measuring correctness of mappings.
- Explanation Plausibility Score, assessing human agreement.
- Extraction-Explanation F1, combining span accuracy and explanation quality.
- Counterfactual Validity Score, testing explanation robustness.
- Multi-Rater Agreement Score for explanation assessment.
- ...
- Counter-Example(s):
- Pure Extraction F1, which ignores explanation quality.
- Speed Metrics, which measure efficiency not interpretability.
- Coverage Metrics, which assess completeness not explanation.
- See: Interpretability Evaluation Metric, Extraction Quality Measure, Explainability Assessment, Joint Performance Metric.