Previewing Intent Classification and Improvements

Purpose

Testing your intent classification ensures your agent understands user queries accurately before going live. The preview feature simulates real-world interactions, allowing you to assess and refine the agent's intent recognition.

How to Use the Preview Feature

Access Preview:
- In the Intent CMS, click the "Preview" button located in the header.
Input Your Utterance:
- Type a user query into the "Utterance" field of the preview modal. Use phrases that represent what users might say.
Send the Query:
- Click "Send" to process the utterance using the intent classification method you've selected.
Review the Response:
- LLM Classification (if enabled):
  - Displays the intent triggered by the LLM.
- NLU Classification:
  - Shows up to 10 intents triggered by the NLU, sorted by confidence score.
- Note: Only intents used in your agent will appear in the results.
Provide Feedback:
- Click the thumbs up or thumbs down icon to indicate if the classification was correct.
- If you select thumbs down, you can specify the intended intent.
Refine and Iterate:
- Based on the results, adjust your intents, utterances, or settings to improve accuracy.
- Use "Re-use last utterance" to retest with the same input after making changes.

Testing Strategies

Test with Variety:
- Use a wide range of utterances, including different phrasings, slang, and common misspellings.
Edge Cases:
- Include ambiguous or complex queries to test how the agent handles them.
Test Different Settings:
- Experiment with model settings in the preview (e.g., AI model, temperature).
- Changes in the preview are temporary and do not affect the agent globally.
Monitor Confidence Levels:
- Low confidence scores may indicate the need for more training data or clearer intent definitions.

Applying Successful Changes

If satisfied with the preview results, manually update the global settings in the Intent CMS to apply them to your agent.