Semantic Confidence - Intent Recognition by Confidence Band in ConvyAI
Overview
The Semantic Confidence page categorizes recognized intents by their confidence score, grouped into five bands. This analysis helps identify where an AI Model is performing well and where it needs improvement.
Navigate to Monitor → Semantic Confidence.
Confidence bands
| Band | Score range |
|---|---|
| No match | 0 |
| Very low | 1 – 40 |
| Low | 41 – 60 |
| Medium | 61 – 85 |
| High | 86 – 100 |
Search criteria
| Criterion | Description |
|---|---|
| Models | Select the AI Model engine to analyze. Defaults to all models. |
| Intents | Optionally filter by a specific intent. If Models is set to all, the Intents filter is also set to all. |
| Time frame | The period to analyze. Defaults to the current day. |
Actions
- Search - Execute the query.
- Save - Save the current search criteria under a name for future reuse.
Result panel
Each confidence band row shows the count of recognized sentences in that range. Click the lens icon in the Score column to expand a second table listing all individual sentences that fall within that confidence band.
For each sentence in the expanded table, click the eye icon in the Detail column to view:
- The bot that processed the conversation.
- The user involved.
- The complete conversation history.
Use this data to improve your AI Model
The confidence band breakdown provides a direct signal for model tuning:
- Sentences in the Very low and Low bands represent misclassified or uncertain recognitions - candidates for additional training examples.
- A high volume in the No match band indicates intents that are missing or under-represented in the model configuration.