NLU is a tool we train and use to analyze phrases from people, their reactions to certain events, their behavior, goals and intentions.
¶ Preparatory Steps
Before you start creating an agent, review recommendations for creating a quality NLU model and follow these steps:
- Register and fund company balance.
- Create a bot into which NLU will be integrated.
¶ Create NLU agent
- Click Create Agent. The settings window will open.
- Name — come up with a name for the new agent.
When choosing a name, pay attention to the following rules: use only Latin letters without spaces and start the name with a capital letter.
- Description — a short description.
- In the Text Processing block, enable the desired switch, which will be responsible for the basic system of recognizing intent in customer phrases:
- Lemma — to bring each word in each training example to its initial form (for example, for nouns the initial form is the singular form of the nominative case). Recommended to install - then when filling NLU-agent you won't have to specify phrases containing the same words with all possible endings.
- Stemma — to remove endings and some simple suffixes in each word in each training example so that only the base of each word remains.
- Correct typos - to correct typos in words based on context and word composition. Not recommended if specific terms are used in the agent.
- In the Additional block:
-
Training Configuration: select a configuration from the drop-down list.
-
Activity: the switch is active by default. To further integrate the agent into scenarios leave the switch active.
-
Recognition Accuracy Threshold: set the percentage of matches between a customer phrase and a phrase from the intent content. To set this threshold, enable the switch and move the slider from 0 to 100. The number you choose will be displayed to the left of the scale.
With this setting, you can control the flexibility and accuracy of trusting the system's assumptions about what a phrase spoken or written by the customer refers to. The accuracy threshold can have different values depending on the importance of the client's answer to a particular question. The selected threshold applies in all scenarios in which the NLU agent is used.
The higher the threshold for accuracy, the more cautious the robot is.
- Click Save. The agent you created will appear on the section page with the status and date of creation.
¶ Copying the NLU agent identifier
Copying the identifier can be useful, for example, to pass it on to technical support.
To copy the NLU agent ID, click on 
in the agent block in the list or on the agent page - at the top of the screen, next to its name.
The ID will then be copied to the clipboard on your computer.
¶ Editing an NLU agent
To change agent parameters:
- Click on
in the agent block in the list or on the agent page - at the top of the screen, next to its name. - Select Edit. The agent settings window opens.
- Make the desired changes in the settings window.
- Click Save.
¶ Copying the agent
To copy a previously created agent:
- Click on in the agent block in the list or on the agent page - at the top of the screen, next to its name.
- Select Copy. The copy window opens.
- Specify the company to copy the agent to. The current company is selected by default.
The company selection is not displayed for all users. If the field is unavailable, the copy will be made to the current company.
- Click Continue — the agent will be copied to the selected company.
¶ Delete an NLU agent
Delete an unwanted agent:
- Click on in the block of the desired agent in the list or on the agent page - at the top of the screen, next to its name.
- Click Delete. A message will be displayed on the screen confirming the deletion.
- Click Delete.
¶ Exporting an NLU agent
To export an agent to a file:
- Click on the in the block of the desired agent in the list or on the agent page — at the top of the screen, next to its name.
- Click Export. The agent configuration file will begin downloading.
Direct editing of the downloaded file is not recommended, as this may corrupt the agent configuration when re-importing. All changes should be made in the NLU editor.
¶ Importing an NLU agent
To import an agent into a file:
- Click the Importing an Agent button in the upper right corner of the agent list.
- In the window that opens, select the file from which the agent configuration will be downloaded.
After selecting the file, a new agent will be automatically created, whose name and content will be loaded from the file. - Click on in the agent block in the list or on the agent page — at the top of the screen, next to its name.
- Select Open. The agent settings page will open.
- Click the Train button in the upper right corner of the agent settings page.
Once training is complete, the agent can be used in bot scenarios.