Skip to content

Knowledge Graph Extraction with Graphistry

Extract knowledge graphs from unstructured text using LLM-powered entity recognition and visualize with Graphistry.

Level: Advanced | Time: 30 minutes | VRAM Required: GPU 0: 5-8 GB, GPU 1: 2-4 GB

Overview

This notebook demonstrates how to extract knowledge graphs from unstructured text using LLM-powered entity recognition and visualize them with Graphistry on a split-GPU architecture.

Key Concepts

  • LLM-based entity extraction from documents
  • Relationship detection between entities
  • Graph construction with nodes (entities) and edges (relationships)
  • Graphistry visualization with interactive exploration
  • GPU acceleration using RAPIDS for large graphs
  • Split-GPU architecture (LLM on GPU 0, Graphistry on GPU 1)

Use Cases

  • Academic paper analysis
  • Legal document processing
  • News article relationship mapping
  • Scientific literature mining

Workflow

# 1. Extract entities and relationships using LLM
response = client.chat.create(
    messages=[{"role": "user", "content": f"Extract entities from: {text}"}]
)

# 2. Build graph
entities_df = pd.DataFrame(entities)
relationships_df = pd.DataFrame(relationships)

# 3. Visualize with Graphistry
g = graphistry.bind(source='from', destination='to', node='entity')
g.edges(relationships_df).nodes(entities_df).plot()

Open in Kaggle

Kaggle