Labelset-Based

Classifies documents using Label Sets to determine the best matching document type.

Remarks

The Labelset Based classification method analyzes the input document by extracting and scoring label matches for all Document Types in scope. It selects the best match based on label presence and scoring, enabling highly accurate classification for semi-structured documents where field labels are consistent but layouts may vary.

This method requires that a Labeling Behavior is configured on the parent Content Model or Content Category, and that each Document Type defines a Label Set describing its expected field labels.

How It Works

• For each Document Type in scope, the method loads its Label Set and attempts to match its labels against the document's text.
• The best-matching Document Type is selected as the classification result, based on label coverage and scoring.
• Classification rules defined on Document Types (such as positive/negative extractors and page count constraints) are enforced and can supplement labelset-based logic.

Usage Guidance

• Use this method for document sets where field labels are reliable indicators of type, even if layouts differ.
• Ensure each Document Type has a well-defined Label Set for optimal results.
• Adjust the 'Prescan Threshold' property to optimize performance for large or complex document sets.
• For outlier cases (such as unlabeled data), override extraction logic at the field level as needed.

For more information, see the documentation for Label Sets, Labeling Behavior, and Document Types.

Properties

Name	Type		Description
Prescan Threshold	Double	►	The percentage of labelset words that must be present in the document to trigger full labelset extraction. ConverterPercent Converter Default0% Min0% Max100% The 'Prescan Threshold' property enables a performance optimization for labelset-based classification. When set to a value greater than 0%, the system performs a quick prescan of the document's text to determine if enough labelset words are present before running full extraction for each Label Set. 0% (default): All labelsets are fully executed for every document, ensuring maximum accuracy but potentially increasing processing time for large or complex sets. Greater than 0%: Only labelsets where at least the specified percentage of words are found in the document are executed. This can significantly improve performance by skipping unlikely matches. Example: If set to 0.5 (50%), a labelset will only be fully executed if at least half of its words are found in the document text. Adjust this value to balance classification speed and accuracy for your solution. Lower values increase thoroughness, while higher values improve performance by reducing unnecessary processing.

Used By

Content Model