Reconcile Classification Results

What is reconciliation?

Reconciliation is the manual task of confirming whether Appian classified content correctly, or changing the predicted category if it's incorrect. Reconciliation is important to validate whether the model's predictions are accurate before downstream decisions are made based on those predictions. For example, classification results might be used to initiate any number of other business processes, so it's important that the model classifies a document or email properly.

Classification also provides monitoring data, which can be used to inform prompt improvements in the future.

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 classification

Users review classified documents 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 classification instance.

Complete the reconciliation task

To complete the reconciliation task for a single document, you compare the chosen category to an image of the uploaded document:

1. On the left side of the page, review the selected category.
2. Use the document preview on the right to verify the accuracy of the selection.
3. If the model selected an incorrect category, click the correct category.
4. Click SUBMIT.

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

For models configured to identify and split packets into individual documents, the reconciliation task includes additional steps to verify and correct document splits and page ranges:

1. Review the original document in the preview pane on the right.
2. In the left panel, you'll see a list of all identified subdocuments with:
   • Start Page: The starting page number for this split.
   • End Page: The ending page number for this split.
   • Category: The document type assigned to this split
   • Classification Reasoning: The AI's reasoning for the category classification.
   • Images: Visual preview of the document pages
3. Verify each split by comparing the page ranges to the document preview:
   • Confirm that the start and end pages correctly identify each subdocument.
   • Check that the assigned document category matches the content
4. Make corrections as needed:
   • Edit page ranges: Click on the Start Page or End Page fields to modify the page numbers. You can specify individual pages or ranges, including out-of-order pages.
   • Edit document type: Click on the Category field to select a different category from the available options.
   • Create a new split: If the model missed a subdocument, click + Split New Document and specify the page range and document category.
   • Delete a split: If a split is incorrect, use the delete option to remove it.
5. The system will alert you if potential issues are detected; however, you can review the following recommendations before submitting:
   • All pages should be associated with a document category (unclassified pages are assigned to Other).
   • Page ranges should not overlap.
6. Click SUBMIT to save your corrections.

Feedback