Skip to main content
After you create a Document skill, follow these steps to train and publish it.
To open Document skill settings, click the settings icon to the right of the skill name in the Skill Designer.
1

Upload training and test documents

Navigate to the Documents tab in the Skill Designer and click Upload documents — in the center of the designer, on the toolbar, or in the Actions pane. Each file must contain a single document image.
While documents are uploading, a progress indicator appears at the top of the Skill Designer (to the right of the bookmarks). To preview a document, click the icon to the left of its name; drag the preview window’s left border to resize it up to 80% of the browser width.
2

Label fields in your training documents

Navigate to the Editor tab — either by clicking the tab name, or by selecting one or more documents from the list and clicking Label Fields and Create Business Rules in the Actions pane. Label every field you want to extract. You can also add validation rules and skill parameters here.
3

Train the skill and review accuracy

Click Train in the Actions pane. When training finishes, the Train button shows Completed. To stop training, click Cancel under the Train button. Review extraction accuracy and correct any errors.
4

Iterate

Repeat the previous steps until you are satisfied with the results.
5

Publish to the Skill Catalog

Publish your skill to the Skill Catalog to make it available for use.
6

(Optional) Derive a skill

You can derive new skills from your published skill. Derived skills inherit all rules and fields from their base skill and can be seamlessly updated to the latest base version while preserving your changes.

​
Process structured documents

ABBYY Vantage offers a machine learning mode for processing structured documents — documents where field locations are the same on every instance. Examples include questionnaires, application forms, and tax return forms. This mode handles forms that have multiple variants, such as IRS Form 1040 for different years, where the set and location of fields differ slightly between variants. Each variant is a separate structured document, and you must upload a blank form for each.
1

Enable fixed-form documents

Create a new Document skill and turn on the Fixed-form documents toggle.
Fixed-form documents toggle in Document skill settings

Fixed-form documents toggle

2

Upload a blank form for each variant

Navigate to the Blank Form tab and click Upload Blank Form — in the center of the designer, on the toolbar, or in the Actions pane. If you don’t have a blank form, upload a completed form and mark it as a blank form.One skill can handle up to 10 variants of one form (for example, IRS Form 1040 for different years).
3

Label fields

Label the fields you want to extract.
4

Eliminate field background (if needed)

In the field settings, enable Eliminate field background for fields whose background may affect recognition.
5

Train the skill

In the Actions pane, click Train.
6

Test with completed documents

Click the Test Set tab and upload completed test documents. Confirm that all fields are labeled correctly on each document. If any field locations don’t match an uploaded blank form, add a blank form for that variant.
7

Review test results

In the Actions pane, test your skill. When the operation completes, review the results. If you are not satisfied, adjust the labeling and train again.
8

Publish to the Skill Catalog

Publish your skill to the Skill Catalog to make it available for use.
Tabs in the Skill Designer for a fixed-form skill

Tabs displayed in the Skill Designer for a skill that processes structured documents

If you edit your skill in Advanced Designer and combine the Forms activity with other activities, the Fixed-form documents toggle is disabled in Vantage and the skill can no longer be edited outside of Advanced Designer.

​
Switch between structured and semi-structured

If you later decide your documents are better treated as semi-structured:
  1. Open Document skill settings.
  2. Turn off the Fixed-form documents toggle. All labeled fields are preserved.
  3. Retrain the skill.

​
Work with tables and repeating groups

When processing structured documents, Vantage can handle tables and repeating groups if:
  • The maximum number of table rows or group instances is known in advance.
  • The boundaries of the table or group are fixed.
Label every row that may appear across all variants of the form. Only rows with data appear in processing results; empty rows are ignored.
Only tables with text values are supported. If your table has columns with checkboxes or barcodes, use a repeating group instead.
If the number of rows or instances in a group is not known in advance, use another Vantage technology — see Process structured documents in Advanced Designer.

​
Configure recognition languages

When processing a document, Vantage selects a processing language from the list of languages enabled on the skill. By default, new skills have English, French, German, and Spanish enabled. To modify the list:
  1. Open Document skill settings.
  2. Select the languages you need. The list is sorted alphabetically, with the currently selected languages pinned to the top. At least one language must be selected.
  3. Click Save to keep your changes, or Cancel to discard them.
The number of selected languages may affect document processing speed. Restrict the list to the languages you actually expect in your documents.
You can change the recognition languages after a skill has been trained. To apply the change, upload the documents again and retrain the skill.

​
Configure Online learning

Online learning collects processed documents into a training set and continues training the skill using those documents. It is available for Document skills and Classification skills. Document skills support two Online learning modes: To use Collect only, open Document skill settings and select the mode.
Collect only mode in Document skill settings

Collect only mode

To turn Online learning on or off for a skill, see Enable Online learning.