Testing the extraction model
After you create and review the model, you can test the model with additional sample documents and correct issues if needed.
Procedure
To test the trained model:
- From the main page in the Designer, on Extraction model, click Open.
- In the Test model tab, test your data extraction model by
uploading new samples that correspond to the document types in your model. The test of the model happens automatically when you upload samples.
To test both the classification and extraction models, you can upload documents in the Any Document Type tab.
- Review results to see whether any issues occurred for your document types. Select a
document to see its extraction results:
- The Results for all fields table lists all fields for the
document type and shows whether each field was found in the sample and, if so, what the confidence
in the match was. For fields that were found in the document, you can expand the row for that field
to see the list of all potential key-value pair (KVP) matches. If more than one KVP is listed, the
first one is the one that the model selected. For each KVP in this expanded row, you can see the
extracted data for each field, next to the corresponding enriched value.
If validation issues occur during testing, they are flagged next to the relevant value. You can directly edit the value and click Rerun validation to take your changes into account. Only the top-ranked value can be edited, for any type of field.
For the subsequent KVPs of a table or composite field, you can click View table data or View subfield results to open the data view and see the results. To avoid confusion, validator failures are not displayed on the data view for those subsequent KVPs.
You can also use the toolbar at the top of the field list to filter fields. This toolbar is available only for the main field list, not for the table data view or composite subfields data view.
- For composite fields, click View and edit subfield results to open the subfield data view.
- For table fields where one or more matches were found, click View and edit
table data to view the data that was found in each row for the matching table, and
correct the values if necessary.
If a table field or composite field is marked as
not in document
, you cannot add table field data or composite data. If a simple field, a subfield of a composite field, or a cell of a table is not found in the document, you can edit its value and rerun the validation.Note: If a field has multiple validators, there might be multiple validator failures. In this case, what is indicated next to the field is the issue with the highest severity for the top-ranked value. - When you are done reviewing the results for a sample document, click Back to return to the list of results.
- The Results for all fields table lists all fields for the
document type and shows whether each field was found in the sample and, if so, what the confidence
in the match was. For fields that were found in the document, you can expand the row for that field
to see the list of all potential key-value pair (KVP) matches. If more than one KVP is listed, the
first one is the one that the model selected. For each KVP in this expanded row, you can see the
extracted data for each field, next to the corresponding enriched value.
- When you are finished reviewing the Test model results, click Done.