What is UiPath Document Understanding?
🎯 Learning objectives
- Describe the UiPath Document Understanding overview and product architecture
- Identify different document types and when to use each processing methodology
- Explain taxonomy — document types, classification, and field definitions
- Describe the end-to-end DU processing pipeline: Digitize → Classify → Extract → Validate → Export
- Explain confidence scores and their role in human-in-the-loop validation
UiPath Document Understanding overview
UiPath Document Understanding (DU) is a product within the IXP platform that enables robots to read, understand, and extract data from business documents — just like a human would — but at scale and speed.
Key capabilities of Document Understanding
- Digitization: Convert physical or image documents to machine-readable text using OCR engines
- Classification: Automatically identify what type of document it is (invoice, purchase order, tax form)
- Data Extraction: Pull specific fields and values from the document
- Human Validation: Route low-confidence results to humans for review via Action Center
- Continuous Learning: Use human corrections to retrain and improve models over time
Document types
| Type | Description | Examples | Best Approach |
|---|---|---|---|
| Structured | Fixed layout, consistent field positions | Tax forms (1040), government forms, driver's licenses | Form Extractor, RegEx |
| Semi-Structured | Same information but different layouts across vendors | Invoices, receipts, purchase orders | ML Extractor (OOTB or custom) |
Document processing methodologies
UiPath DU supports multiple extraction methodologies — each suited to a different document complexity level:
🔧 Rule-based (template-based) extraction
Relies on fixed rules, templates, or regular expressions. Best for structured documents with consistent layouts.
- RegEx Extractor: Pattern matching for fields like dates, phone numbers, IDs
- Form Extractor: Uses anchor-based approach to locate fields relative to known text
- Forms AI: Combines ML with form understanding for both structured and some semi-structured
Advantages: High accuracy for known documents.
🤖 Model-based (ML) extraction
Uses Machine Learning models trained on labeled document examples. Best for semi-structured documents.
- Out-of-the-Box ML Extractors: Pre-trained models for invoices, receipts — ready to use immediately
- Modern Projects (Active Learning): New experience combining annotation, training, and deployment in one UI
Key Insight: ML models learn from characteristics of data — position on the page, correlation of text, and the value itself. They are NOT a database of layouts — they truly understand the document.
⚡ Hybrid extraction
Combines rule-based and model-based approaches. Best for documents that mix structured sections (tables, checkboxes) with semi-structured content.
📍 Continue this module
The rest of this module is split into three short parts so it's easier to read and navigate: