Table Identification Quality Assurance

Table Identification QA tasks are specific to Semi-structured documents with tables. Table Identification QA tasks are part of the Identification QA task queue in Tasks > Perform Tasks.

Note that Field Identification QA tasks and MVTs are also loaded into the Identification QA task queue. To learn more about Field Identification QA tasks and MVTs, see Field Identification Quality Assurance and Model Validation Tasks.

To enable Table Identification QA, you need to:

  1. Go to Flows.

  2. Click on a flow.

  3. Enable Table Identification Quality Assurance in the left-hand sidebar.

  4. Click Save.

Table Identification QA Sampling Rate

Documents that are matched to Semi-structured layouts with tables are eligible for QA sampling. The QA sampling rate can be configured separately for each flow by following the below steps:

  1. Go to Flows.

  2. Click on a flow.

  3. Edit the value for Transcription QA Sample Rate.

  4. Click Save.

Note that Table Identification QA samples on a document level. For example, if you have submitted 10 documents with tables and your sampling rate is 10%, only 1 document will be sent for QA.

Table Identification QA Task

Table Identification QA tasks allow users to add, confirm, and fix cell predictions. For nested tables, Table Identification QA tasks include both the child and the parent tables. You first need to complete Table Identification QA for your child table, and then complete Table Identification QA for your parent table.

Table Identification QA tasks are the same as Table ID Supervision tasks. To learn more about these tasks, see Table ID Supervision.

If the Layout Is Incorrect

If the page displayed in the QA task does not appear to have been matched to the correct layout, follow these steps:

  1. Go to the Document Details dropdown in the right-hand sidebar and expand it.

  2. Click the Mark Layout Variation Incorrect button.

If the page is marked as incorrect:

  • The task will terminate, and the user will be taken to the next Identification QA task that exists. If one does not, they will be taken back to the dashboard.

  • None of the cells on the respective document will be scored for accuracy.

  • The document will not be automatically marked as "No Layout Found".

It is imperative that these pages are caught in QA because we do not want these pages to be included in a Table Locator Model Training job. If those inaccurate pages are included in a training job, it may lead to poor model performance.

Prediction Scenarios

If the layout variation is correct, the following scenarios are possible when completing Table Identification QA tasks:

  1. All cells are accurately predicted.

    1. Complete the task.

      • If you are reviewing a regular table, click the Submit Document (CMD + Enter) button.

      • If you are reviewing a nested table’s child cells, click the Continue to next table (CMD + Enter) button. You will then be asked to review the parent table cells. Once you finish reviewing the parent table cells, click the Submit Document (CMD + Enter) button.

      • Some of the cells are accurately predicted and the rest are not identified/ predicted.

  1. Adjust the cell predictions.

  2. Manually draw the cells that were not identified.

  3. Complete the task.

    • If you are reviewing a regular table, click the Submit Document (CMD + Enter) button.

    • If you are reviewing a nested table’s child cells, click the Continue to next table (CMD + Enter) button. You will then be asked to review the parent table cells. Once you finish reviewing the parent table cells, click the Submit Document (CMD + Enter) button.

    • No cells are predicted.

  1. Manually identify a template row.

  2. Review the predictions made based on your template row.

  3. Adjust the cell predictions if necessary.

  4. Complete the task.

    • If you are reviewing a regular table, click the Submit Document (CMD + Enter) button.

    • If you are reviewing a nested table’s child cells, click the Continue to next table (CMD + Enter) button. You will then be asked to review the parent table cells. Once you finish reviewing the parent table cells, click the Submit Document (CMD + Enter) button.

To complete Table Identification QA tasks, follow the step-by-step guidelines from Table ID Supervision. All available actions for Table ID Supervision are also available for Table ID QA.