Skip to main content
danielb0016
danielb0016
March 10, 2025
Solved

Extract From Document AI Tool - Is there a way to view the confidence scores of data tables?

  • March 10, 2025
  • 1 reply
  • 0 views

I am using the Extract From Document AI tool to read data from a pdf. In addition to text fields, the tool is also extracting a data table, and I'm using the Reconcile Doc Extraction smart service to train the AI to find the data in the table. After running this a few times, the Extract From Document tool has started to bring in data extracted from the table as well.

Using Extract From Document, the confidence scores for the text fields are easily visible as an output of the node, but there doesn't seem to be a way to view confidence scores for the table or its text values within it. Are confidence scores used in data table extract, and is there a way to view them?

Best answer by louis.prensky

Hi [mention:166b8669ade94476adcde7fdafbc6545:e9ed411860ed4f2ba0265705b8793d05] , right now, confidence scores are only available for the text fields in the Skill. This is due to a difference in the underlying ML technology: while you are training a custom ML model to identify the text fields in the document, table fields are extracted based on matching logic between the column headers in the document and the fields you map those columns to in your data structure. As you reconcile documents, the Skill will store the various column headers that may pertain to the fields in your structured data type and use this info for automatic extraction. However, the underlying table model is not actually being trained on this data, and so it does not return a confidence score.

Please let me know if you have any follow up questions, or would like to talk through alternative options to meet your use case.

1 reply

louis.prensky
louis.prenskyAnswer
March 26, 2025

Hi [mention:166b8669ade94476adcde7fdafbc6545:e9ed411860ed4f2ba0265705b8793d05] , right now, confidence scores are only available for the text fields in the Skill. This is due to a difference in the underlying ML technology: while you are training a custom ML model to identify the text fields in the document, table fields are extracted based on matching logic between the column headers in the document and the fields you map those columns to in your data structure. As you reconcile documents, the Skill will store the various column headers that may pertain to the fields in your structured data type and use this info for automatic extraction. However, the underlying table model is not actually being trained on this data, and so it does not return a confidence score.

Please let me know if you have any follow up questions, or would like to talk through alternative options to meet your use case.

    Terms & ConditionsAccessibility statement