Train and test your NLU model in Virtual Agent Designer

  • 릴리스 버전: Australia
  • 업데이트 날짜 2026년 03월 12일
  • 소요 시간: 3분

    • Use the NLU Intent tab to train and try a Natural Language Understanding (NLU) model that is mapped to a topic without leaving Virtual Agent Designer.

    시작하기 전에

    Role required: virtual_agent_admin or admin

    이 태스크 정보

    As you make changes to a topic's intent, train the model so that your updates are incorporated. Then try and test the model to review prediction results based on the latest changes. Continue to test and refine your model until you're satisfied with your changes.

    프로시저

    1. Navigate to All > Conversational Interfaces > Virtual Agent > Designer.
    2. Set the discovery type toggle switch to NLU/Keyword.
    3. Open a topic and view the NLU Intent tab.
    4. Select Train model.
      A message indicates that the model was successfully trained.
    5. Select Try model.
      In the Try model panel, manually enter an utterance to test. The prediction results display in the panel.
      그림 1. Example of testing an utterance in the NLU Try model panel
      The Try model panel showing results of testing an utterance manually.
      If prediction confidence score results meet or surpass the confidence threshold you established for the model, the results appear under Top prediction(s). In this example, the threshold value is 65%.

      The default value of the com.glide.cs.intent_confidence_threshold property is 0.7.

      The default value of the com.glide.cs.entity_confidence_threshold property is 0.5.

      주:
      You can compare the draft NLU model to its most recent published version. For more information, see Compare draft and published versions of your NLU model.
    6. When you're satisfied with the prediction results, publish the model from NLU Workbench or select Publish in the topic Header bar and select NLU Model.

    다음에 수행할 작업

    You can publish the model at the same time as the Virtual Agent topic if the model meets publication criteria. Make sure your model has passed through all phases of development and is ready to publish. For more information, see Model management.

    For more information on providing feedback using thumbs icons in the results, see Test panel feedback.