Classification Model Overview

Automatically attribute a class to a given document, from a list of classes you define yourself.

Use Cases

Process a single document sent in a single file. The classifier looks at all pages of the document in order to identify its type.

Some common examples:

• You have multiple types of files in your workflow input, with different business rules
• You want to identify the region or language of documents

A file sent to the Classification Model may have any number of pages, within limits.

If there is a high possibility of having multiple documents within the same file, use:

• Split Model Overview ⇒ multiple documents in the same file
• Crop Model Overview ⇒ multiple documents on the same page

Create a Classification Model

Classification models are always custom, there are no templates available in the Catalog. This keeps Classification models flexible for different documents and workflows.

Each Classification model gets its own unique model ID when you create it.

1. To create a Classification model, click on Models, and then click on Create your document AI model.
2. Scroll to the Document Utilities section, click on Classify.
3. A pop-up will appear, allowing you to enter the classes you want. Most of the time, you'll use one possible document type per class. For example, if the files you are processing contain invoices, receipts, and driving licenses, set the classes as: INVOICES, IDENTITY DOCUMENTS, CONTRACTS. Classes are returned exactly as defined: including upper or lower case, and any spaces.

Add the class OTHER if you need the model to identify documents that are not one of the explicitly defined classes.

1. Once ready, click on Create Utility to create your Classification Model. This step will also generate the model's unique ID.
2. You can now use the Live Test tab to process documents, and the Utility Configuration to update your classes.

Your utility is now available in your Models tab.

Here is a step-by-step tutorial that shows you how to properly create a Classify Utility :

Technical Considerations

Models requires at least two classes defined.

Class names may be in most languages and writing systems, but cannot exceed 128 characters.

Having more than a few dozen classes will yield unexpected results, and is not recommended. If you need classes based on vendor/customer names, product codes, phone numbers, etc, you should use a text field in an Extraction model instead.

Class names will be returned exactly as defined on the platform in the return, spaces and capitalization included.

If the class names are changed on the platform, the change in the API return will be immediate for all new files sent.

Integration

Once your Classification model is created and tested, integration documentation is provided in the "Documentation" page, or here: Classification Quick Start.