Clusters are groups of similar verbatims that the platform groups together using unsupervised learning and displays in the Discover page.
The platform presents 30 clusters at a time in Discover, each containing 12 verbatims. The platform underlines words or phrases that strongly contribute to the verbatim being included in that cluster.
The verbatims in each cluster are grouped according to what the platform believes are themes of similar intents or concepts.
Reviewing and labelling verbatims within these clusters is typically the first stage in the model training process.
An example cluster of an insurance underwriting mailbox in Discover
Previous: | Next: Concept Drift