➡ Use labeled data to explore unstructured data

Where this can be useful: you've trained an NLU classifier, and want to find see it perform in real-time on new data

➡ Find clusters of semantically similar data

Where this can be useful: you want to explore and drill down into specific topics or intents within unstructured data

➡ Discover utterances least similar to labeled data

Where this can be useful: You have an existing chatbot, and want to discover that things that people are asking that it wasn't trained to understand yet

➡ Find coverage % of "X" in raw data

➡ Identify errors in noisy-labeled data

Where this can be useful: you have a human labeling team, and want to identify quality issues

➡ Build taxonomy of intents for new project

Where this can be useful: You're building the initial set of intents for a new project, and want to build it collaboratively with different stakeholders

➡ Bulk label training examples

Where this can be useful: You've got a production NLU, and want to add new training examples from the on-going conversations to improve its performance

➡ Evaluate performance of NLU model

Where this can be useful: You want to understand the exact performance of your NLU model at the intent and training example level.

➡ Verify effect of data changes to NLU

Where this can be useful: You're doing a lot of work on the NLU data, and want to understand and track the effect of changes on the underlying NLU's performance