LLMGraphTransformer

🧠 LLMGraphTransformer

LLMGraphTransformer is a Python library designed to extract structured knowledge graphs from unstructured text using LLMs. It allows users to define schemas for nodes and relationships, ensuring that the extracted graph follows a strict format. 🔗📊

🚀 Installation

Install LLMGraphTransformer from PyPI:

pip install LLMGraphTransformer

🛠️ Usage

📥 Importing the Required Modules

from LLMGraphTransformer import LLMGraphTransformer
from LLMGraphTransformer.schema import NodeSchema, RelationshipSchema
from langchain_openai import ChatOpenAI
from langchain_core.documents import Document


from dotenv import load_dotenv
import os
load_dotenv(".env")  

🏗️ Defining the Schema

🏷️ Node Schemas

Node schemas define the types of entities that can be extracted from the text. Each node has:

• A type (e.g., "Person", "Organization")
• A list of properties that store additional information (e.g., "name", "birth_year")
• An optional description to describe the node type

📌 Example:

node_schemas = [
    NodeSchema("Person", ["name", "birth_year", "death_year", "nationalitie", "profession"], "Represents an individual"),
    NodeSchema("Organization", ["name", "founding_year", "industrie"], "Represents a group, company, or institution"),
    NodeSchema("Location", ["name"], "Represents a geographical area such as a city, country, or region"),
    NodeSchema("Award", ["name", "field"], "Represents an honor, prize, or recognition")
]

🔗 Relationship Schemas

Relationship schemas define the allowed connections between entities. Each relationship has:

• A source node type
• A target node type
• A relationship type
• A list of optional properties (e.g., "year")

📌 Example:

relationship_schemas = [
    RelationshipSchema("Person", "SPOUSE_OF", "Person"),
    RelationshipSchema("Person", "MEMBER_OF", "Organization", ["start_year", "end_year", "year"]),
    RelationshipSchema("Person", "AWARDED", "Award", ["year"]),
    RelationshipSchema("Person", "LOCATED_IN", "Location"),
    RelationshipSchema("Organization", "LOCATED_IN", "Location")
]

⚙️ Defining Additional Instructions

You can specify additional rules for extraction:

additional_instructions="""- all names must be extracted as uppercase"""

📜 Defining the Input Text

Provide the text from which the knowledge graph should be extracted:

text="""Marie Curie, born in 1867, was a Polish and naturalised-French physicist and chemist who conducted pioneering research on radioactivity.
She was the first woman to win a Nobel Prize, the first person to win a Nobel Prize twice, and the only person to win a Nobel Prize in two scientific fields.
Her husband, Pierre Curie, was a co-winner of her first Nobel Prize, making them the first-ever married couple to win the Nobel Prize and launching the Curie family legacy of five Nobel Prizes.
She was, in 1906, the first woman to become a professor at the University of Paris."""

🤖 Initializing the LLM Model

Use OpenAI's API (or a compatible model) to process the text:

api_key = os.getenv("API_KEY")
base_url = os.getenv("BASE_URL")
model_name = os.getenv("MODEL_NAME")

llm = ChatOpenAI(
    api_key=api_key,
    base_url=base_url,
    model=model_name,
    temperature=0,
)

🔄 Initializing the Transformer

Create an instance of LLMGraphTransformer:

llm_transformer = LLMGraphTransformer(
    llm=llm,
    allowed_nodes=node_schemas,
    allowed_relationships=relationship_schemas,
    additional_instructions=additional_instructions
)

🔍 Converting Text to a Knowledge Graph

Process the text into a structured knowledge graph:

document = Document(page_content=text)
graph_document = llm_transformer.convert_to_graph_document(document)

print(f"Nodes: {graph_document.nodes}")
print(f"Relationships: {graph_document.relationships}")

📊 Output Format

The extracted knowledge graph will be represented in JSON format with nodes and relationships:

{
  "nodes": [
    {
      "id": "Marie Curie",
      "type": "Person",
      "properties": {
        "name": "Marie Curie",
        "birth_year": "1867",
        "nationalitie": ["Polish", "French"],
        "profession": ["physicist", "chemist"]
      }
    },
    ...
  ],
  "relationships": [
    {
      "source": "Marie Curie",
      "target": "Pierre Curie",
      "type": "SPOUSE_OF"
    },
    ...
  ]
}

📜 License

This project is licensed under the MIT License.

🤝 Contributing

Pull requests and feature suggestions are welcome! Open an issue for bug reports or improvements. 🚀