Implement information extraction solutions 5.2 Extract content from documents

A Content Understanding classifier routes incoming files to extraction analyzers.

Study guide skill: Extract information by using multimodal pipelines that combine OCR, layout analysis, and field extraction

Question B5-024

A Content Understanding classifier routes incoming files to extraction analyzers. Invoices arrive in six visual layouts from different suppliers, and the team defined six categories named Invoice_Layout1 to Invoice_Layout6. Classification is inconsistent. What should you do?

Select one answer.

  • A Add another category each time a supplier introduces a new layout
  • B Make the categories differ by which checkboxes are ticked on each layout
  • C Replace them with one Invoice category whose description gives its semantic meaning and key content markers ✓ Correct answer
  • D Rename the categories with non-ASCII supplier names so they're more distinctive

Answer: C

Content Understanding handles visual template variations within a semantic category, so you should use one Invoice category for all invoice variations instead of one per template. Good category definitions give distinguishing context such as semantic meaning and key content markers. The guidance also says to avoid categories that differ only in checkbox values and to use only ASCII characters in category names.

Deep dive

How it works

A Content Understanding classifier routes files by semantic category, and it already handles visual variation. The best-practices page says it automatically handles visual template variations within semantic categories. So the rule is: use semantic categories, not visual templates. Don't create separate categories for files of the same semantic type with different layouts.

The page's own example is this question: use one Invoice category for all invoice variations, rather than Invoice_Template_A and Invoice_Template_B. Six near-identical categories compete with each other, which is why classification is inconsistent.

A good category definition:

  • uses common titles, such as "Annual Financial Report";
  • uses only ASCII characters in category names;
  • provides distinguishing context: semantic meaning, key content markers, or distinctive layouts;
  • adds an "Other" category if you need to catch outliers.

Then route the single Invoice category to your invoice extraction analyzer, which deals with supplier-specific layouts.

The other options, and when you'd use them

  • A new category per new layout: That's the anti-pattern. Don't create separate categories for the same semantic type with different visual layouts.
  • Differ by which checkboxes are ticked: The guidance says to avoid categories that differ only in checkbox values.
  • Non-ASCII supplier names: Category names should use only ASCII characters.

Exam tip

Content Understanding classification: one semantic category per document type (one Invoice), described by meaning and key markers. Not per template, not checkbox-only, ASCII names.

References

Last reviewed:

Ready to test yourself? The quiz asks this and every other AI-103 question, with instant feedback. Practice this in a quiz →

All Information extraction questions · AI-103 study guide

Thanks for your feedback!