Create a document type for each type you aim to recognize in this project. This section provides an overview of all the document types used in your project and allows you to create new, add existing, edit and delete document types.
Document types overview
The table has 2 columns:
- The name of the document type
- The number of entities for each document type
You can edit a document type by clicking on the row. This will open the panel on the right side where you can specify how the document type should be recognized.
Document type details
Recognition type
You can choose to use an A.I. model, set a regular expression (this allows you to choose keywords which in case matched, change the document type) or not automatically recognize the document type.
When defining a Regular Expression document type, keep in mind that Duco Adaptive IDP automatically adds spaces around all punctuation.
Page management
You can enable or disable the page management functionality per document type.
GenAI Instructions
You can also specify extra extraction instructions for the annotation-less model if the model doesn't give you good predictions.
GenAI Instructions
For each document type, you can provide GenAI instructions (prompts) to enhance document classification when using annotationless, zero-shot models. These prompts help improve model performance by offering unique identifiers for different document types. This is especially interesting when you are processing documents that are hard to distinguish.
The content for each prompt should be specific to your documents and use case. Here are some examples:
- This is a formal confirmation document that codifies the terms and conditions of the transaction between parties.
- This financial document should include the broker's details, client's information, invoice number, date, description of services provided, fees or commission, terms of payment, and any applicable taxes.
- A legal document that's issued by a carrier to a shipper detailing the type, quantity and destination of the goods being carried.
When writing prompts, focus on providing details that help the zero-shot model accurately identify the type of document. You can always refine or update the prompt as needed.
Add new document type
You can create a new document type or use an existing one by clicking the "+ Create" button.
