Intents Confusion Report

Intents Confusion Report

Understand how to use & read the Intents Confusion Report
Last Updated June 16, 2022
Table of Contents

You can read the following report horizontally, vertically, or diagonally.
Each one gives you different insights/results about your intents.

Intents Confusion Report

ℹ️ The diagonal intents on the chart are the correctly classified samples.
ℹ️ The horizontal and vertical ones are confused with other intents.

2 reasons why you might misclassify a specific intent:

1st reason: the intent samples might be confused with other intents’ samples. 

What to do if the samples are confused with other intents:

  1. Check the quality of the intents (if you wrote them correctly).
  2. Check the quantity of the intents (if you added sufficient samples).
  3. Check the two intents and see if you need to merge them into one.
  4. Check the samples added to the intent if they belong there or not.

2nd reason: the model is unable to detect if this is “an intent” or not. It’s self-confused.

What to do if the model is self-confused:

  1. Go back to the intent samples page.
  2. Add more samples to the intent.

ℹ️ In the left menu, you’ll see what intents are confused if you cannot read the graph correctly.

Want to learn more?
Was this article helpful?