Data collection

Available in Classic and VPC

Data collection is the process through which the model thinks and examines to provide appropriate answers to the user's questions. Through collecting and reviewing data, you can edit your model's thought process.

Data collection screen

The data collection screen is configured as follows:

Run data collection

The following walks you through the model's thought and judgment process to handle user queries through data collection.

To run data collection, follow these steps:

1. From the NAVER Cloud Platform console, click > Services > AI Services > CLOVA Studio in order.
2. From the my product menu, click the [Go to CLOVA Studio] button.
3. In CLOVA Studio, click the Skill trainer menu.
4. Click the skill set you want to collect data for.
5. Click the Data collection tab.
6. Set required information for data collection.
   • Model/version: select a language model or version to use.
   • Skill: list of skills included in the skill set. You can only use skills whose task status is "completed."
   • Setting > Call option: specify a value for API call for user queries. For more information on how to create a call option, see create call option.
7. Enter queries to collect in the user query area.
8. Click the [→] button.
   • The data collection steps are shown in order on the left side of the screen.
   • Click [Suspend] to suspend a specific step. Click [Restart] to restart the suspended task.

Review and edit data

After running the data collection, if there is any inaccurate judgment from the model at each step, you can improve the response by manually editing the parameters to generate more accurate answers.

To review and edit the collected data, follow these steps:

1. Run data collection.
2. Check the skill call results after the query analysis.
   • For more information on each step, see Data collection steps.
3. Click [Next] at each step to proceed to the next.
   • For a new data collection, you can skip to the final response stage without clicking [Next].
4. To edit your data, click a field, edit, and click [Apply]
   • Data creation is restarted for the next steps.
5. Check the final response and click [Task completed].
   • To save the task for later use, click [Temporary save].
   • The [Task completed] button appears after the final response is created.

Load data

You can load, review, and edit data you previously worked on.

To load data, follow these steps:

1. From the NAVER Cloud Platform console, click > Services > AI Services > CLOVA Studio in order.
2. From the my product menu, click [Go to CLOVA Studio].
3. In CLOVA Studio, click the Skill trainer menu.
4. Click a skill set.
5. Click the Data collection tab.
6. Click [Load] in the Run data collection area.
7. Check the temporarily saved task list.
   • You can search based on user queries from the top right of the Load window.
   • To check the data results, click [⋮] > [View data] from the list.
   • To delete a task, click [⋮] > [Delete] from the list.
   • The task status shows "completed" only when you click [Complete task] on the data task screen.
8. To edit the previous query task from the data collection screen, click [Open].
   • The data collection screen appears, and you can continue to work on the previous queries.
   • You can also load the call option applied to the task.

Data collection steps

Data collection consists of query analysis, skill calls and final response, in order. The following describes the tasks run in each data collection step.

Query analysis

This is the step to decide which skill to use by analyzing user queries. Queries are classified by the call and mapped to relevant skills. If 2 or more queries are analyzed, 2 skill calls are made.

Skill calls

This step includes skill calls and the review of the model's thoughts. Skill calls are performed as many times as the number of queries classified during the query analysis step.

Skill calls consist of 3 steps.

• Step 1: querying API for the skill you want to call
  • Call method: model's thought process to find an API
  • Call skill: skill's model name
  • API query details: API Spec and Description for model of the skill
• Step 2: checking the search results of the API query
  • API search result: model's thought process to call an API. Inferring the request variables (parameters) required for the API call
  • Action: API method. Only supports requests_get and requests_post
  • Action input: URL and parameters inferred by the model. Input fields differ depending on the action
    • If requests_get, action input field: API URL/parameter
    • If requests_post, action input field: API URL/parameter (key, value)
      • Click [Add field] to create an additional blank field
      • Click the icon to delete the field
  • Call result: API call result
• Step 3: organizing content based on the call result
  • Thought: model's thought process to organize the content of the call
  • Result: result of model creation

Final response

In this step, the final response is created and reviewed. After the creation of the final response, click [Task completed] to complete data collection.

Creating call options

By creating call options, you can fix a certain value for the header and body parameters for the API spec operation ID.

To create a call option, follow these steps:

1. While running data collection, click [Call option] of the Setting menu on the right side of the screen.
   • The call option window appears.
2. Enter the call option in JSON type in the call value input area.
   • Create it referring to the example format at the top of the screen.
   • For more information on call value input, see Detailed call option.
3. Once the call value input is completed, click [Apply].
   • If it fails to be applied, an error message is displayed. Re-edit the call value.

Detailed call option

It is information on the fields that can be written in the call option.
The call options are applied in the order of operations field, baseOperation field, and then the user query.