Configure processing settings

Mode

Select whether you want to process individual documents or groups of documents. Select from:

• Document

• Folder (Default)

The value selected here is passed as a variable to every step in the solution process.

Recognition provider

Select from:

OmniPage:

Fast and ideal for machine-printed documents. (Default)

Tungsten:

Sends documents to the Azure AI Document Intelligence Service using a Tungsten volume. This performs raw text and data recognition. We recommend this provider for difficult-to-read hand printed documents, or forms with fields with bounding boxes.

When selected, the following additional setting is available.

Perform OCR on entire document at once:

When set to Yes, an entire document is sent to the Azure AI Document Intelligence Service in a single transaction. (Default: No)

When set to No, each page is sent to the Azure AI Document Intelligence Service separately.

Tungsten with Layout:

Sends documents to the Azure AI Document Intelligence Service using a Tungsten volume. This performs document recognition using the text hierarchy and structure. For example, table fields.

When selected, the following table detection settings are available:

Use for table detection:

When set to Yes, the Azure AI Document Intelligence Service performs table detection instead of Tungsten table detection.

The Azure AI Document Intelligence Service is more optimized to detect tables than the default Tungsten table detection.

When set to No, the default Tungsten table detection is used.

Perform OCR on entire document at once:

When set to Yes, an entire document is sent to the Azure AI Document Intelligence Service in a single transaction. (Default: No)

When set to No, each page is sent to the Azure AI Document Intelligence Service separately.

If you are processing multi-page tables, select this setting. Otherwise, each page of the table will be extracted as a separate table.

Additional recognition providers may be available in this list if configured in TotalAgility Advanced Studio. For more information about configuring recognition engines, see the Tungsten TotalAgility Advanced Studio Help.

If you select a different Azure AI Document Intelligence Service endpoint, both the Use for table detection and Perform OCR on entire document at once settings are available. The same performance issues apply.

Language of documents

Select the language so documents are recognized by the system with the best settings for that language. (Default: English)

Available only when the Recognition provider is set to OmniPage. The language of a document is detected automatically when using Tungsten, Tungsten with Layout, Clarity, or Azure AI Document Intelligence Service recognition providers.

Enable Trainable Separation

Available only when:

• Mode is set to Folder. When it is set to Document, only single documents are processed, so no separation is required.

• At least one of layout or content classification must be enabled. Generative AI classification does not require separation.

When set to Yes, trainable separation is enabled. (Default: No)

The documents provided to train classification are also used to train separation.

Classification

Enable classification accuracy calculation

Select to enable classification accuracy for your solution. (Default: No)

When enabled, you are able to perform a classification accuracy calculation for a solution and view detailed results in an interactive result matrix window. The window displays key metrics including F1 Score (overall accuracy), Precision, and Recall, along with a color-coded result matrix showing correct classifications (green), misclassifications (red), and unclassified documents.

For more information, see Classification accuracy.

Layout Classification

Enabled

Set to Yes to use layout classification to classify layout-based documents. (Default: Yes)

Layout classification requires 2 to 5 training documents.

Scope

Available only when Enabled is set to Yes.

Specify how layout classification is determined. Choose from the following values:

Classify first page only:

Select to use the first page of a document only for layout classification. If the first page cannot be classified the document remains unclassified. (Default)

Classify each page:

Select to process each page in sequence, of a multi-page document for layout classification. Classification ends when a page is classified. If no pages are classified the document remains unclassified.

Minimum confidence (%)

The minimum threshold for a layout classification result to be considered confident. (Default: 70)

Content Classification

Enabled

Set to Yes to use content classification to classify content-based documents. (Default: Yes)

Content classification requires 10 to 20 training documents.

Scope

Available only when Enabled is set to Yes.

Specify how content classification is determined. Choose from the following values:

Classify first page only:

Select to use the first page of a document only for content classification. If the first page cannot be classified the document remains unclassified. (Default)

Classify each page:

Select to process each page in sequence, of a multi-page document for content classification. Classification ends when a page is classified. If no pages are classified the document remains unclassified.

Classify all pages at once:

Select to merge the test of all pages so it is processed by the content classifier at once.

Minimum confidence (%)

The minimum threshold for a content classification result to be considered confident. (Default: 50)

Generative AI Classification

Enabled

Set to Yes to use generative AI classification when trainable classification files , documents are sent to an LLM when trainable classification fails. (Default: No)

Also available when creating a solution.

Generative AI provider

Select the AI provider to use in this solution for generative AI classification. (Default: <Default> Tungsten)

When you select the provider flagged with <Default> and the default provider in TotalAgility Advanced Studio changes, the change is automatically applied to your solution. However, if you select the provider with the same name as the default value but without the <Default> tag and the default changes in TotalAgility Advanced Studio, the provider is not updated in your solution.

If the selected provider no longer exists, a warning is displayed and you must select a valid provider to save these settings.

Page count

Available only when Enabled is set to Yes.

Specify how many pages, starting from the first page in a document are sent to the LLM for classification (Default: 1)

Minimum confidence (%)

The minimum threshold for a generative AI classification result to be considered confident. (Default: 90)

Run before other classifiers

Set to Yes to run generative AI classification before any other classifiers. (Default: No)

When cleared, other enabled classifiers are run first, based on the classification sequence.

Enable extraction accuracy calculation

Select to enable extraction accuracy for your solution. (Default: No)

When enabled, you are able to perform an extraction accuracy calculation for a solution or a document type within a solution.

For more information, see Extraction accuracy.

Generative AI provider

Select the AI provider to use in this solution for extraction. (Default: <Default> Tungsten)

When you select the provider flagged with <Default> and the default provider in TotalAgility Advanced Studio changes, the change is automatically applied to your solution. However, if you select the provider with the same name as the default value but without the <Default> tag and the default changes in TotalAgility Advanced Studio, the provider is not updated in your solution.

If the selected provider no longer exists, a warning is displayed and you must select a valid provider to save these settings.

List of fuzzy databases

Use the  Add fuzzy lookup source,  Edit fuzzy lookup source, and  Delete fuzzy lookup source to manage fuzzy lookup sources.

For more information, see Create a fuzzy lookup source.