Active Learning & Technology-Assisted Review Flashcards

In an Active Learning project, which review queue serves the documents the model ranks as most likely to be relevant first?

Answer: Prioritized Review

Prioritized Review gives reviewers the highest-ranked (most likely relevant) documents first, so relevant material is found early.

What is the main goal of the Coverage Review queue in Active Learning?

Answer: To serve the documents the model is least certain about, so it learns as quickly as possible

Coverage Review serves documents near the model's decision boundary (ranks around the middle). This trains the classifier quickly across the whole data set.

When you set up an Active Learning project, which kind of review field does it require?

Answer: A single choice field with a designated positive choice and negative choice

Active Learning needs a single choice review field, with one choice set as positive (relevant) and one as negative (not relevant), to train the classifier.

What does project validation in Active Learning mainly measure?

Answer: Elusion and recall: how many relevant documents remain in the unreviewed, low-ranked set

Project validation samples the unreviewed documents below the rank cutoff to estimate elusion and recall. This supports a defensible decision to stop review.

What is the range of document ranks that Active Learning assigns?

Answer: 0 to 100

Active Learning gives each document a rank from 0 to 100. Higher ranks mean the model thinks the document is more likely to be relevant.

What kind of index must be built before an Active Learning project can be created?

Answer: An Analytics classification index

Active Learning runs on an Analytics index built for classification. Its classifier is what assigns document ranks.