- Overview
- Getting started
- Building models
- Consuming models
- Model Details
- Public endpoints
- 1040 - document type
- 1040 Schedule C - document type
- 1040 Schedule D - document type
- 1040 Schedule E - document type
- 1040x - document type
- 3949a - document type
- 4506T - document type
- 709 - document type
- 941x - document type
- 9465 - document type
- ACORD125 - document type
- ACORD126 - document type
- ACORD131 - document type
- ACORD140 - document type
- ACORD25 - document type
- Bank Statements - document type
- Bills Of Lading - document type
- Certificate of Incorporation - document type
- Certificate of Origin - document type
- Checks - document type
- Children Product Certificate - document type
- CMS 1500 - document type
- EU Declaration of Conformity - document type
- Financial Statements - document type
- FM1003 - document type
- I9 - document type
- ID Cards - document type
- Invoices - document type
- Invoices Australia - document type
- Invoices China - document type
- Invoices Hebrew - document type
- Invoices India - document type
- Invoices Japan - document type
- Invoices Shipping - document type
- Packing Lists - document type
- Payslips - document type
- Passports - document type
- Purchase Orders - document type
- Receipts - document type
- Receipts Japan - document type
- Remittance Advices - document type
- UB04 - document type
- Utility Bills - document type
- Vehicle Titles - document type
- W2 - document type
- W9 - document type
- Supported languages
- Insights dashboards
- Data and security
- Licensing
- How to
- Troubleshooting
Document Understanding Modern Projects User Guide
Annotate documents
After successfully creating your project and uploading your documents to a specific document type, they are automatically pre-annotated. This is done using a combination of generative and specialized models, based on the document type's schema. The schema clearly defines the fields you want to extract from a particular document type. To find the document type's schema, go to the Annotation page and check the Fields section.
Predictions are indicated with underlines on the text within the document and they can't be deleted. If they are incorrect and cannot be matched to a particular field, you can ignore them. During the training process, only confirmed fields are used for training, while the underlines are not taken into account.
As you continue to add more annotations, the prediction underlines should progressively align with your input. There may be a few inconsistencies between underlines and user-annotated fields at the beginning. However, as you make more annotations and the model improves, the underlines should line up more precisely with the user-supplied data.
In the following image, the Shipping Address has been incorrectly predicted to include the person's name.
To fix this, you only need to confirm the Shipping Address. It's not necessary to remove the underlined text related to the name. As you continue with your annotation and correct such errors, the occasions when the underlined text doesn't align with the confirmed field should decrease.
You can change the extractions view mode using the Extractions view menu. To access this, select the three-dot icon ⁝ on the right side of the document type name and select Extractions view.
Filtered values consist of predictions, which are read-only, and annotations, which can be edited by the user.
- Merge by column: model predictions are displayed in columns that do not have any annotations. Choose this for smaller tables where you can view and validate the whole column.
- Merge by row: model predictions are displayed in rows that do not have any annotations. Choose this when you have larger tables and you want to validate row by row.
- Only confirmed: only extracted values for user confirmed annotations are displayed.
- Only predicted: only display model predictions. Updated automatically when model is retrained and is not editable.
- Show side panel: display the panel on the left side with annotation fields.
- Show table: display the table annotation panel.
After all documents are uploaded and predicted, your goal is to either validate or modify the pre-annotated fields. For a document where all fields are accurately predicted, select Confirm to approve all fields at once. A document, once confirmed, will be signified with a green shield symbol in the document list.
If a document is only partially confirmed, it will be marked with an empty shield symbol in the document list. This symbolizes that the annotation process for this particular document is In Progress. Your end aim should be to make sure that all documents are Confirmed.
- Prediction is correct and should be validated.
- Prediction is not correct and the field is present on the document.
- Prediction is not correct and the field is missing from the document.
- There is no prediction.
If the prediction is incorrect, select the correct text from the document and the appropriate field from the dropdown, then select Confirm.
When working with tables, you can choose to ignore incorrectly predicted values. These values will not be used for model training, and the retrained model will learn to avoid predicting them in future iterations.
You can change the document type settings from the Annotate view.
To do so, select the three-dot icon ⁝ on the right side of the document type name and select Settings.
- Base model: Dataset size estimations used in the Recommended Actions depend on the base model used to train. Using the most similar base model to your Document Type will reduce the amount of annotation work required.
- Number of languages: Dataset size estimation used in the Recommended Actions depend on the number of languages in the dataset. More languages generally require annotating more data.
