Online learning
Online learning improves extraction results. If a document has poor extraction results, you can provide the correct data and then mark that document for online learning. When Kofax TotalAgility processes subsequent documents, the new trained documents help minimize similar extraction errors.
Kofax TotalAgility extracts document contents and stores the content in meaningful structures. Online learning improves the ability of trainable locators to use a dynamic knowledge base to find and extract content during production. Dynamically trained content is added to that knowledge base. After the solution integrator has reviewed the training data, it is published to a static knowledge base. This static knowledge base is a binary file which is stored with the other project files. The dynamic knowledge base for online learning is stored in the data repository.
When a document is selected for online learning, a training algorithm searches the document's meta-data and extracts information meaningful for the trainable locators.
You can mark a document for online learning when:
-
A document is not classified correctly.
-
You have not already marked a document for specific online learning , extraction is poor, and this document is a typical example of its class.
- Automatic Marking
-
If a document group has both online learning and automatic online learning enabled, certain fields can automatically mark a document for online learning. These fields will have the "Monitor for Automatic Online Learning" property enabled.
If the operator selects a recognition result for a monitored field that is different than the one chosen by the extraction process, online learning is automatically enabled. If the operator types new text, it means the recognition completely failed, and online learning cannot be used for this field. Consequently, it will not be automatically enabled despite the property setting.
If the Online Learning button is visible, the operator may decide to cancel, or unmark the document after it was enabled by a monitored field.