Feedback-based validation for bridged learning instances

This topic explains how validation feedback works for learning instances that are imported into Document Automation using IQ Bot - DA Bridge package.

Note: Feedback-based improvement is not a replacement for IQ Bot training (bridged learning instance). It allows continuous learning from user validation feedback in the production environment. For handling additional workloads like IQ Bot training, we recommend building learning instances natively in Document Automation rather than moving them from IQ Bot.

By default, the bridged learning instances only extract data through the training that is available at the time of import from IQ Bot into Document Automation. So, the extraction in Document Automation will natively use IQ Bot "As-Is". In addition, you can enable the validation feedback and apply it on documents being processed in Document Automation by selecting the Improve accuracy using validation option.


Feedback validation in DA for bridged instances

How validation feedback works?

Scenario 1:

When user processes a document on a bridged IQ Bot learning instance, some documents are sent to the validator for manual validation. In this scenario, users can leverage the capabilities of Improve accuracy using validation option to perform feedback-based extraction on such documents. The advantage of this feature in Document Automation against IQ Bot is that you need not spend time on training your learning instance, simply validating your documents will improve extraction for sub-sequent runs on similar documents.

Scenario 2:

For example, user had trained a learning instance of certain vendor, such as A and B. However, for new vendor documents such as X, they are not getting extracted with available training. So, these documents are sent to the validator and IQ Bot bridged learning instance will extract these documents with empty field values. In this scenario, users can apply validation feedback on these new vendor documents from validator and avoid reprocessing for all sub-sequent similar documents.

Example

Consider an example of bridged learning instance that contains the form fields as Invoice Number, Invoice Date and Description, Price, and Quantity as table fields. The below table lists all the scenarios to explain how validation works for this learning instance.
IQ Bot training (Bridged learning instance) Validation Feedback Actions Extraction results
Form fields Table fields Form fields Table fields
IQ Bot training IQ Bot training No feedback given No feedback given IQ Bot training extracts all fields
IQ Bot training IQ Bot training Feedback applied on all fields No feedback given
  • Form fields - Validation feedback
  • Table fields - IQ Bot training
IQ Bot training IQ Bot training Feedback applied on Invoice Number which is populating as blank for the processed document while Invoice Date is populated correctly. No feedback given
  • Form fields - Validation feedback applicable only on Invoice Number while Invoice Date will be through IQ Bot training
  • Table fields - IQ Bot training
IQ Bot training IQ Bot training Invoice Number which is populated but remapped to a different value during validation. Invoice Date is populated correctly. No feedback given
  • Form fields - Document will be processed based on the validation. However, all future similar document's Invoice Number, Invoice Date will be processed through IQ Bot training. Remapping done will not override the IQ Bot training for Invoice Number.
  • Table fields - IQ Bot training
IQ Bot training IQ Bot training No feedback given Feedback applied on all fields
  • Form fields - IQ Bot training
  • Table fields - Validation feedback
IQ Bot training IQ Bot training No feedback given Feedback applied on Price, which is populating as blank for the processed document while other fields are populated correctly.
  • Form fields - IQ Bot training
  • Table fields - Validation feedback will be generated for all the fields such as Description, Price, and Quantity based on validation done in Document Automation. Thereafter, this feedback will power extraction for all table fields on subsequent documents.
Note: In this scenario, the best practice is to perform the validation at least 2 times for the feedback to be applied effectively.
IQ Bot training IQ Bot training No feedback given Price which is populated but remapped to a different value during validation. All other fields are populated correctly.
  • Form fields - IQ Bot training
  • Table fields - Document will be processed based on validation. However, all the future similar document's Description, Price, and Quantity will be processed through IQ Bot training. Remapping done on Price will not override IQ Bot training for subsequent documents.
IQ Bot training IQ Bot training Feedback applied on all fields Feedback applied on all fields
  • Form fields - Validation feedback
  • Table fields - Validation feedback
Note: Validation feedback based extraction might function differently versus IQ Bot training based extraction. It is advisable to test and validate typical document workload to assess the extraction outcomes.

Language support

Currently, learning instances that are imported into Document Automation using IQ Bot - DA Bridge package supports the following languages for feedback based validation:
  • Dutch
  • English
  • French
  • German
  • Italian
  • Portuguese
  • Spanish