引導式測驗 · 中等 等級
文件審查測驗中的召回率、精確率和排除測試
檢查召回率、精確度、規避、抽樣限制、普遍性、參考編碼以及如何報告文件審查中的不確定性。
問題 1 的 8
A sample shows 80 responsive documents out of 100 documents coded responsive. Which measure is 80%?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- A sample shows 80 responsive documents out of 100 documents coded responsive. Which measure is 80%?
- Coverage: all collected documents divided by the review population
- Precision: responsive documents found divided by all documents labeled responsive
- Recall: responsive items found divided by all truly responsive documents
- Elusion: missed responsive items divided by the nonresponsive sample
- A responsive document was classified nonresponsive. Which cell does it occupy in a confusion matrix?
- True positive
- False positive
- True negative
- False negative
- Which sample most directly estimates elusion?
- A statistically selected sample from documents coded nonresponsive
- A sample of documents already coded responsive
- A judgment sample of the first files in a folder
- A sample drawn only from documents with the highest relevance scores
- Why can a low elusion estimate fail to establish high recall when responsiveness is rare?
- Elusion is calculated only from responsive documents
- Rare responsiveness makes random sampling impossible
- A small number of misses in a large nonresponsive pool can still correspond to many missed responsive items
- Low elusion mathematically guarantees high recall at every prevalence
- Which sampling approach supports a statistical claim about a population?
- A set chosen because its items were easy to code
- A probability-based sample drawn from a defined population with its method documented
- A reviewer’s hand-picked set of likely responsive examples
- The first documents displayed by the review platform
- What does high recall by itself fail to establish?
- That the process also has high precision or that every responsive item was found with certainty
- That the system identified any responsive documents
- That the responsive population was defined
- That review instructions can be documented
- Why should reviewers define a reference coding standard before comparing a model with human review?
- To ensure the model’s label is treated as the legal definition
- To avoid adjudicating any disputed documents
- To raise precision by excluding difficult documents from the population
- To make disagreements interpretable against consistent responsiveness criteria
- What should a validation report include alongside an estimated recall or elusion rate?
- A claim that the sample proves complete production
- The model score without reviewer instructions or sample selection
- Population, sample design, criteria, counts, and uncertainty
- Only the percentage and the software vendor name