Perform an extraction benchmark

Use the procedure in this topic to perform an extraction benchmark.

To generate extraction benchmarks, you need to open your golden files or a benchmark document set in the Documents window.

After running an extraction benchmark, you might want to change your extraction configuration settings. The following are common mistakes you want to avoid:

• Lowering the minimum confidence thresholds just to get more green valid fields. You may end up missing invalid data as a result.

• Increasing the minimum confidence thresholds just to eliminate red incorrect valid fields. This likely does not fix the problem that is causing the incorrect valid fields, only masks it. These types of errors may still occur in production.

• Modifying recognition profile settings to improve a specific field. This may adversely affect other locator methods and fields.

Instead, you should concentrate on the following:

• Confirm the recognition engine you are using is the most effective for your documents.

• Improve your classification results so they do not negatively affect your extraction results.

• Improve the regular expressions used by a Format Locator.

• Review and update zone settings such as image cleanup, anchoring and registration for an Advanced Zone Locator.

• Review and improve any scripts used at both project and class level.

• Verify that all databases are up-to-date and that your queries return the correct data.

• Review and improve Table Locator settings for regular and trainable tables.

• Update any other locator-specific settings that could improve extraction results.