Build natural language processing domains and continuously refine and evolve your NLU model based on real‑world usage data. Define user intents ('book a flight') and entities ('from JFK to LAX next Wednesday') and provide sample sentences to train the DNN‑based NLU engine.
Build NLU Models
NLU starter packs
When starting a new project, pick from a set of NLU starter packs with predefined intents and samples to be added to your project.
This will give you a head start both with business intents (banking, telco, etc.) and ‘social’ intents (greetings, apologies, emotions, fun questions, and more).
NLU start pack offers include vertical-specific options, such as telco, banking, or utilities as well as generic social intents or chat.
Define intents and entities
Train your NLU model with sample phrases to learn to distinguish between dozens or hundreds of different user intents. For each intent, define the entities required to fulfill the customer request. Create custom entities based on word lists and everyday expressions or leverage ready‑made entities for numbers, currency, and date/time that understand the variety of ways customers express that information.
Mix.nlu supports the process of ‘tagging’ sample messages/utterances from end users with an auto-intent feature that automatically categorizes them by intent.
Auto‑intent groups new samples into existing intents where it finds a close match, increasing the accuracy for existing intents.
And where no good match is found in the existing model, it will suggest new intents—candidates for additional automation.
Multi‑language NLU models
Mix allows you to create and manage multi‑language applications in a single project. In Mix.nlu, you can create a single set of intents and entities across multiple languages, with language‑specific training sets.
This helps provide a more consistent multi‑language user experience at lower cost.
Test and tune
NLU feedback loop
For the best possible natural language understanding results, you need an automated, AI‑based, scalable feedback cycle that uses data from your end user conversations to continuously improve accuracy and scope of the NLU model.
Mix.nlu supports you with this by providing access to production data, AI‑based auto‑intent discovery, support for manual reviews, and an easy way to measure the success of changes in regression tests and production.
NLU and speech recognition tuning
Deploy the trained NLU model both to the NLU engine and at the same time, as a domain language model, to the speech‑to‑text transcription engine. This provides the highest accuracy in speech recognition results, semantic parsing, and understanding of user utterances based on your application’s specific language domain.