When you log inferences with a user_query field, we can automatically run query topic classification against every logged inference.

Athina segments your data based on these query topics to show you granular analytics and comparisons.

Alternatively, you may log the topic manually. Learn more

Configure Query Topics

You can configure your topics in your Settings 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 page or the Analytics 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.