Configuring clarification to improve response accuracy


When configured in the environment, an administrator can use the clarification to improve the accuracy of BMC HelixGPT responses. The clarification is triggered when a user's query is ambiguous and could match multiple documents from the configured data sources.

Static and dynamic clarifications are the two ways to configure clarification in BMC HelixGPT. To learn more, see Static clarification and Dynamic clarification

To configure clarification for prompt-based skills, use the Knowledge Clarification prompt, and to configure clarification for agent-based skills, use the Static Query Clarification tool. The Knowledge Clarification prompt uses both static and dynamic clarification methods to generate clarification options, whereas the Static Query Clarification tool uses only the static clarification method.

BMC HelixGPT supports both static and dynamic clarification methods, and each works independently. You can use either or both based on your use case. If both static and dynamic clarifications are configured, the system first runs the static clarification. If no matching record is found in the QueryClarification records, it then runs the dynamic clarification.

The following image displays a clarification question with options as clickable buttons:

Clarification_prompt.png

 Configuring static clarification

In static clarification, the administrators can configure record definitions as predefined questions and options. These predefined options, shown as clickable buttons, are displayed when additional context is needed to refine the user’s query. This is useful for common or expected ambiguities.

After the predefined clarification options are configured, you can configure the similarity threshold parameter in the tool configurations. The similarity threshold defines how closely a user query must match an intent or knowledge article. If the query does not meet the threshold, BMC HelixGPT asks for clarification.

Consider the following points while configuring the similarity threshold value:

  • You can configure the similarity threshold value for the Agent-based skill in either the skill or the tool configuration.
  • The default and the recommended threshold value is 0.9 in skill configuration; you can modify it as per your requirements. There is no default value configured in the tool configurations.
  • If the value is configured in both the skill and tool configurations, the parameter in the tool configuration takes precedence over that in the skill configuration.
  • Configuring the similarity threshold for the prompt-based skill is optional. 

Task 1: To configure the predefined clarification options

Important:

Modifications to records in QueryClarification may take up to one hour to be reflected in the system's functionality.

  1. Log in to BMC Helix Innovation Studio.
  2. On the Workspace tab, click HelixGPT Agent Studio.
  3. From the Records tab, select QuerClarification and click Edit data.
  4. In the Data editor (QueryClarification) window, click +New.
    Record_definition.png
  5. In the New record window, specify the following fields:
    Field nameDescription
    Clarification QuestionThe clarification response presented to the user.
    FacetA category or tag used to group or filter the clarification context.
    QuestionThe trigger phrase or keyword entered by the user that activates the clarification record.
    Options

    A JSON list of selectable clarification options presented to the user.

    For example:  ["option1", "option2"]

    StatusStatus of the record.
    Indicates whether the record is in an active or inactive state.
  6. Click Save.

Task 2: To configure the similarity threshold in the skill configuration

  1. Log in to the HelixGPT Agent Studio, select the application and the skill for which you want to add the clarification configuration.
  2. Go to the Configure tab and select the Configuration box.
    Skill_configuration.png
  3. Add the threshold value as a key-value pair. 
    For example:  clarificationQuestionThreshold: 5.
  4. Click Apply.
     

To configure the similarity threshold in the tool configuration

You can set the similarity threshold by updating the Static Query Clarification tool.
For more information, see Modifying the default Static Query Clarification tool settings.

Task 3: To add the clarification instructions to the agent instructions

For the agent-based skills to trigger the clarification, the agent instructions must be updated to instruct the agent to invoke the static clarification tool.

Click here to view the sample agent instructions to invoke the Static Query Clarification tool:

Static query clarification vs knowledge search:
When the user is asking for information that would typically come from knowledge articles, documentation, how‑to guides, troubleshooting steps, or policies, you MUST first call the QueryClarificationTool.
CRITICAL: You MUST call QueryClarificationTool for EVERY knowledge-seeking query, even if:
- The same or similar question was asked earlier in this conversation
- The user previously selected an option for a similar query
- You believe you know what the user wants based on conversation history
Treat each user message as independent for clarification purposes. Do NOT assume the user means the same thing as a previous question, even if the wording is identical.
Your first ACTION MUST be to call the QueryClarificationTool with the full user query.
If the tool returns a clarification question with a list of options (validation), you MUST:
- Ask the user to select exactly one of the provided options.
- Wait for the user's choice.
- Use the chosen option to refine the query and then call the KnowledgeTool.
If the QueryClarificationTool returns no clarification question (empty content), proceed directly to using the KnowledgeTool.

To enable clarification for the custom agents

  1. Add the Knowledge Tool and Static Query Clarification Tool to the custom agent.
  2. Add the clarification instructions to the agent instructions.

Configuring dynamic clarification

In dynamic clarification, the administrators can use the Knowledge Clarification prompt to dynamically generate clarification options based on the referenced documents from connected data sources. The prompt helps in scenarios where predefined questions are ambiguous.

To configure the Knowledge Clarification prompt

Use the knowledge Clarification prompt to configure clarification in prompt-based skills.

Related topic

Creating and managing skills

 

 

Tip: For faster searching, add an asterisk to the end of your partial query. Example: cert*

BMC HelixGPT 26.1