Reconcile Extraction Results

What is reconciliation?

Reconciliation is the manual task of confirming or updating whether Appian extracted data from a piece of content correctly. Users compare the data that was extracted to an image of the uploaded document. DocCenter leverages reconciliation data to calculate important model performance metrics such as extraction accuracy, precision, recall, and F-1 score.

How a reconciliation task is assigned

Reconciliation may occur while you're refining a model by testing it with sample content or when the model runs in production. During testing, the reconciliation task is automatically assigned to the person who initiated the test.

How to reconcile extraction

Users review extracted data on the reconciliation form. Here, they can update any data that was extracted incorrectly.

Reconciliation tasks can be accessed as a related action on the extraction instance.

When you run through your document extraction process, most of the work occurs in the background. It's only after Appian extracts data from a document that DocCenter generates a task for a user to validate the data.

Complete the reconciliation task

To complete the reconciliation task, you compare the data that was extracted to an image of the uploaded document. You can use the information that displays in the document preview to update any incorrect or missing information.

For XLSX files, the document preview displays a basic representation of the source spreadsheet so you can compare extracted values against the original cells. Note that source file formatting such as color and font styles is not retained in the preview.

To complete the reconciliation task:

1. On the left side of the page, review the information in the fields. Use the document preview on the right to verify the accuracy of the data.

   • Note: To view where the information in the fields came from, click the icon before the field. The value is automatically highlighted in the document preview.

     

     Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

2. If any information is missing or incorrect, click the highlighter icon next to the field, and click and drag your cursor around the correct value in the document preview.

3. Perform additional reconciliation for tables, if present in the document preview. If any rows were missed during extraction, click Add Row to enter additional values.

4. After all fields are verified and populated, click SUBMIT.

Field validations and flags

As you complete a reconciliation task, you may see additional information to help you review the content. Refer to the table below to understand what each type of message means, and how you should interpret it.

Reconcile large tables

Large tables in Excel files can contain many columns and rows. When you start a reconciliation task, DocCenter displays a preview of the rows and columns in a large table.

You can correct data in-line or by uploading a corrected Excel file, depending on your preferences. In-line updates may be easier when there are only a few changes to make, while editing in Excel may be easier for more substantial updates.

Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

To correct data in-line:

1. Go to the extraction instance you want to reconcile.
2. Click Reconcile Fields.
3. By default, the reconciliation task displays a preview of the extracted data as well as the original file. To change any extracted value, click VIEW FULL TABLE.

4. Double-click the cell you want to edit. For easy comparison, the cell is highlighted on both the Reconciled tab and the Source Data pane.

   

   Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

5. Edit the value as desired.
6. Repeat the process until you're finished.
7. Click SAVE to complete the reconciliation task.

To correct data using an Excel file:

1. Go to the extraction instance you want to reconcile.
2. Click Reconcile Fields.
3. By default, the reconciliation task displays a preview of the extracted data as well as the original file. To change any extracted value, click VIEW FULL TABLE.
4. On the Reconciled tab, click DOWNLOAD.

5. Choose whether to download the file as an XLSX or CSV file.

   

   Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

6. Open the downloaded file on your local computer and edit values as desired.
7. Click UPLOAD RECONCILED FILE and select the file containing your changes. DocCenter alerts you to structural mismatches between the reconciled file and extracted data, such as a different number of columns.
8. Review the data preview to confirm the corrected data looks accurate.
9. Click SAVE to complete the reconciliation task.

Create a custom reconciliation form

To enable your users to review and reconcile extracted field data, you'll want to set up an interface. This interface will be used later on in your process model.

1. In the DocCenter application, find the AIA_ExtractionInstance_RecordView_Summary interface and check the box to select it.
2. Click DUPLICATE in the toolbar.
3. Provide a Name and Description for your new interface, and select the folder in the application where you'd like to save it.
4. Click CREATE. The new interface object opens in a new tab.
5. In Design mode, find the Record Action component.
6. Click and drag the component to the interface, and drop it where you'd like it to appear.
7. In the component configuration, click ADD ACTION. The Record Action Item is added to the list of Actions.
8. Click Record Action Item to configure it.
9. In the Record Action field, search for AIA Extraction Instance.
10. In the list of actions, select reconcileField.
11. In the Record Identifier field, select ri!extractionInstance.id.
12. Click SAVE.
13. Finally, add a button component so the user can submit the user input task when reconciliation is complete.

For example, your interface may look like this:

Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

When integrated into your process model, this interface provides users with a chance to review the extracted fields. If they notice something that needs to be fixed, they can click the RECONCILE FIELDS record action. If they change any values, the updated data will be saved, and model metrics will reflect the changes as well.

Save reconciled data

Reconciled data is stored in key-value pairs, which then must be saved back into a record type with fixed fields. To assist with this, DocCenter provides an expression rule called rule!AIA_API_Extraction_ConvertInstanceToMap.

Use the following rule to convert the extraction instance into a data map with keys matching the model keys:

1
2
3
rule!AIA_API_Extraction_ConvertInstanceToMap(
  ExtractionInstance (required): <Extraction instance to be converted>
)

The resulting map can then be cast directly to the record type (if the keys match) or mapped field by field using a helper expression rule.

Feedback