A machine learning approach where the model identifies which unlabeled examples would be most valuable to label, reducing annotation costs. Active learning strategies may involve human labelers viewing sensitive content, raising labor and content moderation considerations.
See: Annotation; Human evaluation; Labeling