DEV Community

Alain Airom
Alain Airom

Posted on

Information Extraction (Tests on a Beta API) from Docling

#docling #data #dataextraction #datapreparation

An early test on information extraction API from Docling

Introduction

Docling provides the essential capability of extracting information — structured data — from unstructured documents. This process is highly configurable: the user defines the desired data schema, or “template,” which can be provided either as a simple dictionary or as a precise Pydantic model. Docling then processes the document and returns the extracted information as standardized output, cleanly organized by page for easy consumption and integration into downstream applications.

Disclaimer: Please note that the Docling API is currently in Beta and may be subject to changes as development progresses.

Docling- Information Extraction Capabilities

This document serves as an introduction to the Docling library, specifically detailing the functionality demonstrated in the accompanying docling_extractor.py Python application.

The Python code provided is a runnable translation of a sample Docling notebook, demonstrating how to execute information extraction in a programmatic way.

➡️ The aim of this code is to extract the information from the following invoice 🧾

To begin, ensure your development environment is properly prepared by installing the necessary dependencies. The adapted code, incorporating standard best practices and guardrails for robust execution, is provided below.

python3 -m venv venv
source venv/bin/activate

pip install --upgrade pip

pip install 'docling[vlm]'
Enter fullscreen mode Exit fullscreen mode

And the main code ⤵️

import os
import json
from typing import Optional
from pydantic import BaseModel, Field
from rich import print

# --- Setup and Initialization ---

# NOTE: The Docling library must be installed before running this script:
# Run: pip install 'docling[vlm]'

# 1. Define the output directory and ensure it exists, as requested.
output_dir = "./output"
os.makedirs(output_dir, exist_ok=True)
print(f"Output directory '{output_dir}' ensured to exist.")

# 2. Define Pydantic models for structured data extraction.
class Contact(BaseModel):
    """Represents contact details for the sender or receiver."""
    name: Optional[str] = Field(default=None, examples=["Smith"])
    address: str = Field(default="123 Main St", examples=["456 Elm St"])
    postal_code: str = Field(default="12345", examples=["67890"])
    city: str = Field(default="Anytown", examples=["Othertown"])
    country: Optional[str] = Field(default=None, examples=["Canada"])

class Invoice(BaseModel):
    """A simple model matching the initial extraction template."""
    bill_no: str
    total: float

class ExtendedInvoice(BaseModel):
    """A complex model for deeper, nested extraction."""
    bill_no: str = Field(
        examples=["A123", "5414"]
    )  # provides examples to guide the model
    total: float = Field(
        default=10, examples=[20]
    )  # provides default and examples
    garden_work_hours: int = Field(default=1, examples=[2])
    sender: Contact = Field(default=Contact(), examples=[Contact()])
    receiver: Contact = Field(default=Contact(), examples=[Contact()])

# --- Docling Core Logic ---

# We use the remote URL provided in the original request as the source document.
# (To use a local file from "./input", you would change this variable.)
file_path = (
    "https://upload.wikimedia.org/wikipedia/commons/9/9f/Swiss_QR-Bill_example.jpg"
)

from docling.datamodel.base_models import InputFormat
from docling.document_extractor import DocumentExtractor

print(f"\n--- Starting Document Extraction from: {file_path} ---")

# Initialize the DocumentExtractor
extractor = DocumentExtractor(allowed_formats=[InputFormat.IMAGE, InputFormat.PDF])


# --- 1. Extraction using JSON string template ---
print("\n[bold yellow]--- 1. Extracting with JSON string template ---[/bold yellow]")
result_json_str = extractor.extract(
    source=file_path,
    template='{"bill_no": "string", "total": "float"}',
)
print("Extracted Pages Data:")
print(result_json_str.pages[0].extracted_data)


# --- 2. Extraction using Python dict template ---
print("\n[bold yellow]--- 2. Extracting with Python dict template ---[/bold yellow]")
result_dict = extractor.extract(
    source=file_path,
    template={
        "bill_no": "string",
        "total": "float",
    },
)
print("Extracted Pages Data:")
print(result_dict.pages[0].extracted_data)


# --- 3. Extraction using simple Pydantic model (Invoice) ---
print("\n[bold yellow]--- 3. Extracting with simple Pydantic model (Invoice) ---[/bold yellow]")
result_invoice = extractor.extract(
    source=file_path,
    template=Invoice,
)
print("Extracted Pages Data:")
print(result_invoice.pages[0].extracted_data)


# --- 4. Extraction using complex Pydantic model (ExtendedInvoice) ---
print("\n[bold yellow]--- 4. Extracting with complex Pydantic model (ExtendedInvoice) ---[/bold yellow]")
result_extended = extractor.extract(
    source=file_path,
    template=ExtendedInvoice,
)
# Print the extracted data (raw dictionary from the Docling result)
print("Extracted Pages Data:")
print(result_extended.pages[0].extracted_data)


# --- 5. Pydantic Validation and Final Output ---
# Validate the extracted data against the Pydantic model for type checking and structure
invoice = ExtendedInvoice.model_validate(result_extended.pages[0].extracted_data)

print("\n[bold green]--- Pydantic Validation Result (Model Instance) ---[/bold green]")
print(invoice)

# Print specific fields using the validated model
print("\n[bold cyan]--- Formatted Output from Model ---[/bold cyan]")
print(
    f"Invoice [bold magenta]#{invoice.bill_no}[/bold magenta] was sent by [bold magenta]{invoice.sender.name}[/bold magenta] "
    f"to [bold magenta]{invoice.receiver.name}[/bold magenta] at {invoice.sender.address}."
)

# --- 6. Write Result to Output Folder ---
output_file = os.path.join(output_dir, "extracted_invoice.json")

# Convert the Pydantic model instance to a JSON dictionary
# We use .model_dump_json() and then load/dump to ensure a clean, formatted JSON file
json_data = json.loads(invoice.model_dump_json(indent=2))
with open(output_file, 'w') as f:
    json.dump(json_data, f, indent=2)

print(f"\n[bold green]Successfully saved structured output to:[/bold green] [yellow]{output_file}[/yellow]")
Enter fullscreen mode Exit fullscreen mode

The output ⬇️


python docling_extractor.py
Output directory './output' ensured to exist.

--- Starting Document Extraction from: https://upload.wikimedia.org/wikipedia/commons/9/9f/Swiss_QR-Bill_example.jpg ---

--- 1. Extracting with JSON string template ---
/Users/alainairom/Devs/docling-informationextraction/venv/lib/python3.12/site-packages/docling/document_extractor.py:143: UserWarning: The extract API is currently experimental and may change without prior notice.
Only PDF and image formats are supported.
  return next(all_res)
The following generation flags are not valid and may be ignored: ['temperature', 'top_p', 'top_k']. Set `TRANSFORMERS_VERBOSITY=info` for more details.
Extracted Pages Data:
{'bill_no': '3139', 'total': 3949.75}

--- 2. Extracting with Python dict template ---
Extracted Pages Data:
{'bill_no': '3139', 'total': 3949.75}

--- 3. Extracting with simple Pydantic model (Invoice) ---
Extracted Pages Data:
{'bill_no': '3139', 'total': 3949.75}

--- 4. Extracting with complex Pydantic model (ExtendedInvoice) ---
Extracted Pages Data:
{
    'bill_no': '3139',
    'total': 3949.75,
    'garden_work_hours': 28,
    'sender': {'name': 'Robert Schneider', 'address': 'Rue du Lac 1268', 'postal_code': '2501', 'city': 'Biel', 'country': 'Switzerland'},
    'receiver': {'name': 'Pia Rutschmann', 'address': 'Marktgasse 28', 'postal_code': '9400', 'city': 'Rorschach', 'country': 'Switzerland'}
}

--- Pydantic Validation Result (Model Instance) ---
ExtendedInvoice(
    bill_no='3139',
    total=3949.75,
    garden_work_hours=28,
    sender=Contact(name='Robert Schneider', address='Rue du Lac 1268', postal_code='2501', city='Biel', country='Switzerland'),
    receiver=Contact(name='Pia Rutschmann', address='Marktgasse 28', postal_code='9400', city='Rorschach', country='Switzerland')
)

--- Formatted Output from Model ---
Invoice #3139 was sent by Robert Schneider to Pia Rutschmann at Rue du Lac 1268.

Successfully saved structured output to: ./output/extracted_invoice.json
-> Cannot close object, library is destroyed. This may cause a memory leak!
-> Cannot close object, library is destroyed. This may cause a memory leak!
-> Cannot close object, library is destroyed. This may cause a memory leak!
-> Cannot close object, library is destroyed. This may cause a memory leak!
Enter fullscreen mode Exit fullscreen mode

With the Json generated 📄

{
  "bill_no": "3139",
  "total": 3949.75,
  "garden_work_hours": 28,
  "sender": {
    "name": "Robert Schneider",
    "address": "Rue du Lac 1268",
    "postal_code": "2501",
    "city": "Biel",
    "country": "Switzerland"
  },
  "receiver": {
    "name": "Pia Rutschmann",
    "address": "Marktgasse 28",
    "postal_code": "9400",
    "city": "Rorschach",
    "country": "Switzerland"
  }
}
Enter fullscreen mode Exit fullscreen mode

Conclusion

In conclusion, we’ve explored Docling’s robust capabilities within the domain of data extraction and preparation. This powerful tool leverages flexible templates and Pydantic models to efficiently convert unstructured documents into structured, page-organized data. The sample provides a comprehensive introductory context and guiding through its core functionality.

Links

Top comments (0)