Query Topic Classification
When you log inferences with a user_query field, we will automatically run query topic classification against every logged inference.
Configure your Topics
You can configure your topics in your Settings (opens in a new tab) page.
Currently, we will categorize each user query into a single topic label.
So it is best to configure non-overlapping topics.
Example Topics
- Product Question
- Refunds
- Order Status
- Exchanges
- Complaints
- Product Availability
- Shipping
Explore your data, segmented by topic
When you apply a topic filter on the Observe (opens in a new tab) page or the Analytics (opens in a new tab) page, the usage and performance metrics will also update to only include inferences labeled with the selected topic.
This can help you narrow down to only look at inferences that were categorized into a certain topic.
Example: you may find that your model's pass rate is 85% for Product Questions, but only 68% for queries about Refunds.
Example: you may discover that your model's average response time is 1.3s for queries about Shipping, but 5.4s for queries about Product Availability.