The Adaptive Search Queries feature allows users to test and define specific search queries for documents with similar layouts within a single learning instance, enhancing data extraction accuracy for varied document formats.

A single set of search queries is useful in learning instances that process documents that have uniform layouts. However, such queries might not work as expected when processing documents with diverse layouts, even when the documents belong to a particular document type such as invoices and receipts.

The Adaptive Search Queries feature provides the capability for users to test different search queries for sets of documents with similar layouts within a single learning instance. For example, if you are testing data extraction for invoices that include the following different sets of documents:

  • Invoices that include details of purchase for a single product.
  • Invoices that include details of purchase for multiple products.

In such a scenario, a single search query might not work well to extract the required data. Using the Adaptive Search Queries feature, you can define specific search queries that work well for each set of documents.

In test mode, a unique cluster ID is generated for a set of documents with similar layout and content. You can define and test search queries for form and table fields for such documents.

Using the Adaptive Search Queries feature, you can perform the following:

  • Test and validate search queries at the cluster level.
  • Define different search queries for the same field across multiple clusters within a single learning instance.
  • Promote validated cluster-specific queries to production for robust and consistent extraction performance.
  • Minimize manual query tuning and improve extraction accuracy across varied document formats from within a single, unified learning instance.

Benefits

  • Improved accuracy: Increases extraction success rates across varied layouts within the same learning instance.
  • Reduced manual tuning: Minimizes repetitive trial-and-error in defining the required search queries.
  • Streamlined navigation: Eliminates back-and-forth adjustments by refining queries at the cluster level.
  • Traceability: Allows to track search query changes using version history. See View version history of a learning instance.

Availability

  • The Document Extraction package version must be 3.38.8 or later.
  • The Adaptive Search Queries feature is available only in test mode. See Test learning instances.
  • The Adaptive Search Queries feature is not supported for Unstructured document type.
  • User must have the Validator user role along with the Train learning instance groups permission. See Document Automation users.
  • Learning instances must use a provider in the Generative AI-driven data extraction option for data extraction.