Active Learning Sampling

When using HumanFirst NLU as the active NLU engine, you can filter your unlabeled data by Margin Score, Uncertainty and Entropy.

Margin score

Exposes utterances that have equal match scores across two intents in your workspace. It helps uncover utterances that are ambiguous between two intents (labeling them to the appropriate one will help disambiguate the model). It also helps discover new intents that should be created to further disambiguate the model.

Uncertainty

Exposes utterances that have the lowest confidence of matching any existing intent in your workspace. It helps discover new intents or utterances that are under represented in your training data.

Entropy

Exposes utterances that have strong matches across multiple intents in the workspace. It helps uncover utterances that are ambiguous between many intents (labeling them to the appropriate one will disambiguate the model). It also helps discover new intents that should be created to further disambiguate the model.

note

Make sure you have trained the NLU from the NLU section (to get updated results) and that the active NLU engine is HumanFirst NLU.