Unstructured and complex documents user guide

The validation data provided in the Measure tab can help identify where adjustments are needed to the taxonomy, meaning the extraction schema. Iterating on instructions to generate improved predictions that match the annotations is critical to ensuring optimal model performance. This means you could add additional relevant context to instructions to capture instances of fields or field groups that the model missed. Make sure the schema is comprehensive and model performance aligns with the data extraction needs of your organization.

For more details, check Reviewing predictions.