Test field extraction
Use the procedure in this topic to test field extraction for any document type. The same test documents are used for both classification and extraction testing.
Testing field extraction ensures that your fields return the best results possible during production. If any formatters or validators are applied to a field, those are considered within the extraction results, so you know what operators will see during production.
Training documents have a plain thumbnail image, and test documents have a test button on the thumbnail image.
Before you begin
Before you test field extraction, verify that the following steps are complete:
-
You have at least one configurable or predefined document type.
-
Extraction is enabled.
-
Extraction is configured for one or more fields, or you have predefined document types.
Procedure
- Select the Fields menu.
- Select a document type.
-
Select a test document and select
Test.
Extraction is performed, and the field extraction results are displayed under the Fields tab. The extraction result takes both formatting and validation into account, if configured. The color of the extraction data circle does not indicate validation errors. It is possible to have a green extraction data circle and an invalid result that fails validation.
The extraction data circle has one of the following results. The tooltip for each icon includes extraction details for a field, including original value, and any validation messages, if applicable. The tooltip for the field on the Document Viewer also includes the formatted value.
Icon
Description
No value is found during testing.
A field is extracted with high quality and confidence.
A field is extracted with low quality and confidence.
-
If a field is not extracted as expected, compare the test document with a training document and perform one or more of
the following:
-
If the field label differs on both documents, add the new label to the list of keywords on the training document field.
-
If the layout of the document is not consistent, ensure that the same field is added to multiple test documents in each location.
Also, consider adding a document with this layout as a training document for this document type. Train your project and then test the extraction again.
-
If the extracted value is not as expected, check that the field is configured with the appropriate type and formatter, and has the appropriate length and character restrictions.
-
Review any validators to ensure that they are not too restrictive. Any field that does not meet the validation rules is flagged as invalid and requires review by a validation operator during production.
-
- After updating your field settings, Test again.
- Repeat steps 2 to 5 for other test documents and document types until your extraction results meet your requirements.
Next steps
-
Configure your import channels.
-
Optional. Publish your solution.
More info