AI Builder Confidence Score and Classification

[Xinlin]

Hi, 

I have a Text Classification Model that will read incoming email and rate the email content with Confidence Score and Classification. The flow managed to run successfully but the result came out with two classification and pasted the inaccurate value back to my Dataverse table. 

Based on the data that I trained with, it should update my table with 'Need Further Clarification' but it updated with 'Pending' instead. Below shows the output of AI Model and data updated in Dataverse: 

[
  {
    "@odata.type": "#Microsoft.Dynamics.CRM.expando",
    "type": "Need Further Clarification",
    "score": 0.9958772
  },
  {
    "@odata.type": "#Microsoft.Dynamics.CRM.expando",
    "type": "Pending",
    "score": 0.77722
  }
]

Why is the AI Builder capturing two different value when it should be just one? Is there any way I can capture the score with higher value and update in my Dataverse table? I've also attached my flow below for better reference. 

Thank you in advance! 

[Antrod]

Hi @Xinlin ,

It's possible that several categories correspond to a text. If you want to just take the one with the higher confidence, here's an example of logic you could add in your flow

• Set an object variable called SelectedCategory which is empty by default
• When browsing all the returned categories by the AI model, you can add this logic:
  • If the variable SelectedCategory is empty, set it with {"category": "<current category value>", "confidence": <current category confidence>}
  • If the variable SelectedCategory is not empty, compare current confidence with the one in the variable: variables('SelectedCategory')['confidence']. If its higher, replace the SelectedCategory values with the ones from the current.
• Then you can insert the category from the variable into your Dataverse row, along with the other information

Hoper that helps.

[Xinlin]

Hi @Antrod ,

Thanks for replying! However, I don't really understand the below part of the solution as I think my variable wont be empty. Can you show me an example of it? Thank you. 

• If the variable SelectedCategory is empty, set it with {"category": "<current category value>", "confidence": <current category confidence>}
• If the variable SelectedCategory is not empty, compare current confidence with the one in the variable: variables('SelectedCategory')['confidence']. If its higher, replace the SelectedCategory values with the ones from the current.

[Antrod]

Hi @Xinlin ,

Here's an example of logic you could apply to get the category with the highest confidence score:

Instead of the Compose, use your logic to add data in Dataverse