Virtual Agent and NLU Workbench integration

  • Release version: Washingtondc
  • Updated June 28, 2024
  • 2 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Virtual Agent and NLU Workbench Integration

    The Virtual Agent and NLU Workbench integration allows administrators to access and update Natural Language Understanding (NLU) models directly within the Virtual Agent Designer. This integration facilitates the migration of existing NLU topics to new Large Language Model (LLM) topics for those using Now Assist in Virtual Agent.

    Show full answer Show less

    Key Features

    • Creation and configuration of NLU models and intents in the NLU Workbench, requiring admin or nluadmin roles.
    • Enabling NLU and selecting the NLU service provider in Virtual Agent General Settings.
    • Applying NLU models to conversation topics in Virtual Agent Designer, including setting NLU properties for entity extraction.
    • Activation of Dialog Acts for flexible responses (Modify, Affirm, Negate) based on user interactions during conversations.
    • Updating NLU intent utterances and managing model training, testing, and publishing from within the Virtual Agent Designer.

    Key Outcomes

    Once the NLU model is published in the NLU Workbench, it can be seamlessly utilized in the Virtual Agent Designer. Administrators can publish topics with mapped intents, but publishing may fail if the model is untrained, training is ongoing, or if the intent is not enabled.

    Virtual Agent administrators can access and update their NLU models from within the Virtual Agent Designer user interface.

    Diagram of NLU and Virtual Agent authoring.

    Note:
    If you have Now Assist in Virtual Agent, you can continue to use your existing NLU topics and migrate them into new LLM topics using the topic migration feature within Virtual Agent Designer. For more information on topic migration, see Migrate NLU topics to LLM topics.

    Integration setup tasks, roles, and details

    As Virtual Agent administrators create and configure their conversation topics, they must first create their NLU model and its associated intents in the NLU Workbench. This action requires they use the NLU Workbench and the admin or nlu_admin role.

    In addition, they must also complete the following tasks in Virtual Agent General Settings.
    • Enable NLU.
    • Select the NLU service provider.
    • If using language-specific NLU models, enable the languages for those models.
    Virtual Agent administrators must also apply their NLU model to a conversation topic by completing the following tasks in Virtual Agent Designer:
    • In Topic Properties, select the NLU model, the NLU intent, and the topic switching behavior.
    • For input controls used in the topic flow, set the NLU properties for entity extraction.

    Optionally, admins can activate Dialog Acts to enable Virtual Agent to respond flexibly when users make a modification in mid-conversation. Currently available response types are Modify, Affirm, and Negate, based on the last 5 exchanges in the conversation. Dialog Acts can be configured for English only, in Topic Properties. For more information see Dialog Acts for Virtual Agent.

    After the NLU model is complete and associated with a Virtual Agent conversation topic, administrators with the virtual_agent_admin or admin role can do the following from within the Virtual Agent Designer user interface.
    • Update NLU intent utterances.
    • Train, test, and publish the NLU model.

    For more information, see Natural Language Understanding (NLU) topic discovery in Virtual Agent.

    Publishing topics from Virtual Agent

    When your model is published in the NLU Workbench, it's ready to use in Virtual Agent Designer. When editing a topic, click the Properties tab to select a model and intent to map to that topic.

    Natural Language Understanding section of the Properties tab of Virtual Agent Designer.

    When you click Publish, the model and intent are mapped to that topic and published seamlessly.

    Publishing a topic with a mapped intent fails for the following reasons:
    • Model isn't trained, or training is in progress.
    • The last trained model is already published with a VA topic.
    • The intent is not enabled in the model.