Advanced Structures

Document Structure

import instructor from openai import OpenAI from pydantic import BaseModel, Field from typing import Optional, List client = instructor.from_openai(OpenAI()) class Section(BaseModel): heading: str content: str subsections: List["Section"] = Field(default_factory=list) Section.model_rebuild() class Document(BaseModel): title: str abstract: Optional[str] = None authors: List[str] = Field(default_factory=list) sections: List[Section] = Field(default_factory=list) keywords: List[str] = Field(default_factory=list) def extract_document_structure(text: str) -> Document: return client.chat.completions.create( model="gpt-4", messages=[ { "role": "system", "content": "Extract the structured representation of this document, including all sections and subsections." }, { "role": "user", "content": text } ], response_model=Document ) document_text = """ This paper explores the applications of machine learning in healthcare settings. machine learning, healthcare, AI, medical diagnosis Machine learning has shown promising results in healthcare applications. Healthcare has historically been slow to adopt new technologies. Data privacy and model interpretability remain significant challenges. We employed a mixed-methods approach. Our findings indicate a 30% improvement in diagnostic accuracy. These results have significant implications for clinical practice. Machine learning will continue to transform healthcare delivery. """ doc_structure = extract_document_structure(document_text) print(f"Title: {doc_structure.title}") print(f"Authors: {', '.join(doc_structure.authors)}") if doc_structure.abstract: print(f"Abstract: {doc_structure.abstract}") print(f"Keywords: {', '.join(doc_structure.keywords)}") print("\nSections:") for section in doc_structure.sections: print(f"- {section.heading}") for subsection in section.subsections: print(f" - {subsection.heading}")

Instructor can extract structured representations of documents, such as articles, papers, or reports. This approach helps convert unstructured text into structured data that can be analyzed and processed.

import instructor
from openai import OpenAI
from pydantic import BaseModel, Field
from typing import Optional, List

Initialize the client with instructor

client = instructor.from_openai(OpenAI())

Define section structure

class Section(BaseModel):
    heading: str
    content: str
    subsections: List["Section"] = Field(default_factory=list)

Important! For recursive models, we need to rebuild the model

Section.model_rebuild()

Define document structure

class Document(BaseModel):
    title: str
    abstract: Optional[str] = None
    authors: List[str] = Field(default_factory=list)
    sections: List[Section] = Field(default_factory=list)
    keywords: List[str] = Field(default_factory=list)

Extract document structure from text

def extract_document_structure(text: str) -> Document:
    return client.chat.completions.create(
        model="gpt-4",
        messages=[
            {
                "role": "system",
                "content": "Extract the structured representation of this document, including all sections and subsections."
            },
            {
                "role": "user",
                "content": text
            }
        ],
        response_model=Document
    )

Example usage

document_text = """

Abstract

This paper explores the applications of machine learning in healthcare settings.

Keywords

machine learning, healthcare, AI, medical diagnosis

Introduction

Machine learning has shown promising results in healthcare applications.

Background

Healthcare has historically been slow to adopt new technologies.

Current Challenges

Data privacy and model interpretability remain significant challenges.

Methods

We employed a mixed-methods approach.

Results

Our findings indicate a 30% improvement in diagnostic accuracy.

Discussion

These results have significant implications for clinical practice.

Conclusion

Machine learning will continue to transform healthcare delivery.
"""

doc_structure = extract_document_structure(document_text)

Print the document structure

print(f"Title: {doc_structure.title}")
print(f"Authors: {', '.join(doc_structure.authors)}")
if doc_structure.abstract:
    print(f"Abstract: {doc_structure.abstract}")
print(f"Keywords: {', '.join(doc_structure.keywords)}")
print("\nSections:")
for section in doc_structure.sections:
    print(f"- {section.heading}")
    for subsection in section.subsections:
        print(f"  - {subsection.heading}")

Running the Example

First, install Instructor and any dependencies

$ pip install instructor pydantic

Run the Python script

$ python document-structure.py

Further Information

Documentation link 1