Taxonomy in Document Understanding
🎯 Learning objectives
- Explain what a taxonomy is and why it's the single source of truth for a DU project
- Describe the three-level hierarchy: Taxonomy → Document Types → Fields
- Identify the available field types and when to use each
- Apply best practices for building a well-structured taxonomy
Taxonomy is the metadata collection that guides the Document Understanding framework throughout its operations. It acts as a single source of truth — defining what documents to expect and exactly what data to extract from them.
Hierarchical structure
Taxonomy follows a three-level hierarchy:
- TaxonomyThe complete metadata collection for the project
- Document TypesLogical document categories (e.g., Invoices, Medical Records, Contracts), each with a unique ID. These enable both classification and extraction.
- FieldsIndividual data points to be extracted from a specific document type (e.g., Invoice Number, Due Date, Total Amount)
Field types available
| Field Type | Description | Example |
|---|---|---|
| Number | Numeric values with up to eight decimals | Quantity: 5, Unit Price: 12.50 |
| MonetaryQuantity | Currency amounts with currency codes | Total: USD 1,250.00 |
| Date | Temporal data — day, month, year; supports MSDN-compliant format strings | Invoice Date: 01/06/2025 |
| Name | Person names — given name, middle name, last name | Patient Name: John A. Smith |
| Address | Multi-line addresses with city, state, country, postal code | Billing Address |
| Set | Predefined value options (enumeration) | Status: Paid / Unpaid / Pending |
| Boolean | Yes / No values | Signed: Yes |
| Table | Tabular data with defined columns | Line Items table |
| FieldGroup | Related fields organised hierarchically | Vendor Info (Name + Address + VAT) |
Additional taxonomy features
- Exception Handling: Predefined "Report as Exception" reasons appear in the Validation Station, allowing users to flag documents with specific issues
- Language Support: Taxonomies can specify supported languages with language codes, and include an "Undetermined Language" (code
und) for edge cases - Storage: Taxonomies are stored as
taxonomy.jsonfiles in the DocumentProcessing folder within your UiPath Studio project - Taxonomy Manager: Managed through the built-in Taxonomy Manager wizard in Studio and loaded at runtime via the
Load Taxonomyactivity
💡 Tip
Best Practice: Define your taxonomy before building any model. A well-designed taxonomy saves significant rework later. Work with a Subject Matter Expert (SME) who understands both the business process and all edge cases in the documents.
Building a good taxonomy
- Identify all document types in your process
- For each document type, list all data points that need to be extracted
- Assign the correct field type to each field (Number, Date, Table, etc.)
- Group related fields using FieldGroup for hierarchical data (e.g., line items)
- Define exception reasons for documents that cannot be processed
- Test with sample documents before investing in large-scale labelling