When you perform machine learning for a document, a new machine-learning-extraction subfolder is created in the batch class folder on the Transact server. This subfolder contains JSON files with machine learning data for each document type and index field.

For example, if another user learns the same index field under the same document type and batch class using a Web Service, the application saves the JSON file for that learning as well.

The next time you extract data from the document, the system compares the anchors around the extracted value with anchors saved in all existing JSON files. The value with the highest confidence is shown as the extraction result. The JSON files are then merged into one file containing the latest learning information.

This feature helps ensure that all machine learning results are saved and used during further data extraction. We recommend that you refrain from editing the JSON files manually.