Create a custom extraction model using Standard Forms

Create a Standard Forms extraction model in the Control Room.

Prerequisites

  • Ensure your Control Room has the Document Automation (Number of pages) product license. For more details on licenses, see Understanding licenses in Document Automation.
  • You must have the AAE_IQ Bot Admin or AAE_IQ Bot Services roles to complete this task
  • Identify between five and fifteen training documents
  • The maximum limit of Standard Forms document type is 157286400 bytes.
  • For limitations about using custom neural models, see Document Intelligence custom neural model.

Procedure

  1. In the Control Room, navigate to AI > Document Automation, and click Create model.
    Note: You will not be able to see the Create Model option on the learning instances page to create models when you use your own license (BYOL). See Build and train a custom extraction model.
    The Projects window opens in a new tab.
  2. Click Create project.
  3. Enter a name for the project, click Browse to upload documents to train the extraction model, and click Create.
  4. Define the field tags:
    1. Click the plus (+) icon on the upper-right corner of the screen and select Field.
    2. Enter a name for the tag and select Enter. For example, Invoice Number or Invoice Date.
    Selecting a field value region
    Note: Each field tag and its corresponding location tag is assigned a unique color so that it is easy to associate them when distinguishing or validating data visually. Clicking a field tag will highlight the corresponding location tag on the document.
  5. Define the table tags:
    1. Click the plus (+) icon on the upper-right corner of the screen and select Table.
      Adding a table field
    2. Enter a name for the table and click Create.
      Adding a table name
    3. For each column in the table, select the column name and perform one of the following actions:
      • Rename column: Select this option to rename the column and select Enter to rename.
      • Insert column: Select this option to insert a column after the selected column.
      • Delete column: Select this option to delete the selected column.
    4. Update the column details.
    5. Optional: Click the plus (+) icon to add more rows.
    6. Click the close (x) button on the upper-right corner of the screen to exit the table view.
    Enter the column name in the green-outlined box
  6. Click the Region option on the upper-left corner of the screen and define the location of the data for each field:
    1. For the form fields, highlight the location of the data to extract and select the corresponding field name from the list.
      Adding a field name
    2. For each cell in the table, highlight the location of the data to extract, then select the cell. Click the close (x) button on the upper-right corner of the screen to exit the table view.
      Selecting a table field value region
    3. Select the next document and repeat the steps to define the locations of the form and table fields.
  7. Click the Train icon and enter a name for the model.
    Train model icon
  8. Select one of the following models:
    • Template: Custom template is an easy-to-train document model that accurately extracts labeled key-value pairs, selection marks, tables, regions, and signatures from documents. This model is typically used to extract data from structured documents that have consistent layout and defined visual templates. See Document Intelligence custom template model.
    • Neural: Custom neural model is a deep learned model type that combines layout and language features to accurately extract labeled fields from documents. This model is more suitable to extract data from semi-structured and unstructured documents. See Document Intelligence custom neural model.
      Important: As the custom neural model uses deep learning technology, it might take some time for completing the training for the model. Therefore, you might not see the model immediately on the custom models page. We recommend that you wait for the training to be complete and then select this model for use.
  9. Click Train.
  10. Click the Analyze icon, click Browse for file, and select a document, and click Run analysis to test the extraction model.
    Note: As of Automation 360 v.29, you can create and train new models in Azure AI Document Intelligence v.3.0 only. If you are in Azure AI Document Intelligence v.2.1, then you must update to Azure AI Document Intelligence v.3.0 by following the below steps:
    1. Open Project.
    2. Click the Train icon.
    3. Click Train. The model is created in the Azure AI Document Intelligence v.3.0.

Next steps

Create a learning instance for standard forms
Note: You cannot edit an extraction model after associating it with a learning instance.