F2 Score Measure

An F2 Score Measure is an evaluation measure that weights recall twice as heavily as precision in harmonic mean calculation for retrieval performance assessment.

• AKA: F-Beta Score Measure with Beta=2, Recall-Weighted F-Measure, F2 Measure.
• Context:
  • It can typically compute Weighted Harmonic Means with beta parameter of 2.
  • It can typically emphasize Recall Performance over precision performance.
  • It can typically evaluate Information Retrieval Systems with completeness priority.
  • It can often assess Legal Retrieval Tasks where missing relevant documents are costly.
  • It can often measure Medical Diagnosis Systems where false negatives are critical.
  • It can often evaluate Safety-Critical Systems with high recall requirements.
  • It can often compare Retrieval Algorithms with different precision-recall tradeoffs.
  • It can range from being a Micro-Averaged F2 Score to being a Macro-Averaged F2 Score, depending on its aggregation method.
  • It can range from being a Binary F2 Score to being a Multi-Class F2 Score, depending on its classification type.
  • It can range from being a Document-Level F2 Score to being a Passage-Level F2 Score, depending on its evaluation granularity.
  • It can range from being a Weighted F2 Score to being a Unweighted F2 Score, depending on its class importance.
  • ...
• Examples:
  • Legal Retrieval F2 Scores, such as:
    • COLIEE 2025 Task 3 F2 Score, evaluating statute retrieval completeness.
    • E-Discovery F2 Score, measuring relevant document recall.
  • General F2 Score Applications, such as:
    • Medical Screening F2, Fraud Detection F2, Safety Alert F2.
  • ...
• Counter-Examples:
  • F1 Score, which equally weights precision and recall.
  • F0.5 Score, which weights precision twice as heavily as recall.
  • Precision Metric, which ignores recall entirely.
• See: F1 Score, F-Beta Score, Precision-Recall Tradeoff, Evaluation Measure, Information Retrieval Evaluation, Legal Statute Retrieval Task, Harmonic Mean, Legal Evaluation Measure.

---