Using feedback documents from applications to improve training
The model can learn from the corrections that are made in the live document processing applications by the business users. When you are adding document types and training the model, you can activate that option.
About this task
You can enable the use of feedback documents and select the percentage of corrected documents to use in retraining when you create or edit your documents types. For more information, see Establishing document types.
After you deploy your project, develop a document processing application, and deploy this application to IBM Business Automation Navigator, the business user is able to add documents and review any classification or data extraction issues if needed. The documents that are manually corrected in the application are the only ones that can be used as feedback documents in the model retraining.
Each day, the corrected documents are pulled back into Document Processing Designer, which you are notified of when you open your extraction model. You can choose to click Train model in the Teach model tab. In the dialog that opens, select Feedback documents to retrain the model with the new documents that were corrected in your production environment, which increases the model accuracy. However, the more feedback documents are used, the more time it takes to train the model. A maximum of 1000 feedback documents can be used during each retraining.
If Feedback documents is enabled on the document types settings and feedback loop documents for classification are available, when you enter the Classification model page a feedback prompt is displayed. Go to and click Train. On the Confirm training dialog, enable Feedback documents to include the feedback loop in classification training.