Use Machine Learning to fix extraction errors

In the Validation queue (in Validator), IQ Bot can learn from manually corrected (by humans) field values, over a specific time period.

Prerequisites

IQ Bot can learn from the document field data pairs, for example, incorrect field values and manually corrected field values, and become more confident about an erroneous field value and what that value correction is. If IQ Bot reaches a 90%+ confidence level, it auto-corrects the erroneous field value.

If additional documents are uploaded in Production, IQ Bot auto-corrects errors, skips the Validation queue, and counts the documents as straight-through processing (STP).

Sometimes, IQ Bot does not reach the 90%+ confidence level. For example, if incorrect data and manually corrected data differ in length by more than four characters, the manually-corrected data is not provided to the Machine Learning (ML) model.
Note: The Validation queue auto-correction does not apply to check box extraction.

Procedure

  1. In the Validation queue, when correcting a field value error, select from the available options.
  2. Choose to not select a suggestion, but instead enter a value, or click to another field.
  3. Choose to enter a value in a field to fix the error, and save your corrections, IQ Bot saves your manually-corrected field value.
  4. If you make multiple corrections for the same fields with the same value, the confidence of IQ Bot might increase. If the confidence level reaches 90%; the system auto-corrects the field value in real-time with your suggested value (when you click into the cell).
    Note: For uploaded documents, if all failed fields are auto-corrected, the document skips the Validator, and instead, exports directly to CSV, counting as Dashboard STP.