Build

Available in Classic and VPC

There are three types of build as follows.

• Conversation model build: The type where a build is done by learning through a deep learning model.
• Quick build: The type where a build is done without learning through a deep learning model.
• Apply changed settings: This feature enables instant application of partial changes to the settings to the chatbot service without learning a conversation model.

Conversation model build

The following describes how to build a conversation model.

1. Complete creating a chatbot by referring to Create domain and Create conversation.
2. Click the Conversation model build > [Conversation model build] button at the upper right of the chatbot builder.
3. When the Conversation model build window appears, click the [OK] button.
   • You can document the changes of the build in the memo field.
4. You can check the progress of the build in the Chatbot settings > [Build history] tab.

Quick build

Quick build is a type of build where a build is done without learning through a deep learning model. Since only the complete match engines and regular engines are built based on the input question data, the build is completed in approximately 5 minutes.
Unlike the conversation model build, a quick build does not learn the deep learning model, so the chatbot can't respond with an answer, even if a user's question which has similar intent with the registered questions is entered. Since the quick build can respond to less coverage compared to the conversation model build, it is recommended that you use the quick build when testing a composed scenario in a quick manner. Once enough tests have been conducted, complete the conversation model build and proceed with service deployment.
See the following for how to conduct a quick build.

1. Complete creating a chatbot by referring to Create domain and Create conversation.
2. Click the Conversation model build > [Quick build] button at the upper right of the chatbot builder.

Apply changed settings

The "Apply changed settings" feature is used when applying some configuration features that don't require learning conversation models instantly to the service. It's supported only for the most recently built conversation model, and it will be automatically beta-deployed once the settings have been applied. The changed settings application won't be counted as builds.
The following describes how you can apply changed settings.

1. Change the chatbot's settings.
   • Common message, link messengers and services, edit form, edit action method
2. Click the Conversation model build > [Apply changed settings] button at the upper right of the chatbot builder.

Cancel build

If you have to cancel a build after starting the conversation model build due to modifications that need to be made, then click the [Cancel build] button from Chatbot settings > Build history tab.

Manage build

A chatbot is not completed with the initial building but it requires a process where the quality of the chatbot is raised with constant relearning of conversation models after creation. In this process, history management related to conversation model learning is required, and this is called build management.
You can see the information of the chatbot which is being built now and history of the chatbot builds so far in the Build history tab. It also provides deployment management feature such as reverting to a previously learning model and deploying the service.

View build information

You can view the build and service deployment information, build memos, deployed file locations, and service availability by build version in the Build history tab.
To see the build history, click the [Build history] button from the chatbot builder's left menu, or click Chatbot settings > [Build history] tab.

• Build status: Check the build status for each step in Build status.
  • Build ready: Request learning build and approve build
  • Preprocessing: Data preprocessing step
  • Learning: Learning based on conversation datasets
  • Available for testing: Step for testing data in the process of learning in a beta environment
  • Learning completed: Conversation dataset learning completed, such as natural language analysis and entity mapping
  • Build applied: Step for applying the learned conversation model build
  • Build completed: Service deployment available
  • Build canceled: Build is available for cancellation before the preprocessing step
  • DATABASE SNAPSHOT ERROR: Build error
• Actions
  • Beta deployment button: Deploys the service to a beta environment, where you can run tests before starting an actual operation.
  • Service deployment button: Deploys the service to a real environment, which will be applied to the actual service with channels linked. Service deployment is available only for versions with the build status of "Build completed."
  • Apply settings button: Applies the settings without a conversation model build. It's the same feature as "Apply changed settings." This is only enabled for the most recently deployed version.
• Build usage statistics
  Check the number of builds and service deployments in graphs. Aggregation is conducted every hour, and you can check the following information.
  • Number of builds and build failures during a specific period
  • Number of service deployments during a specific period