How to visualise a Knowledge Graph

Introduction

Hello everyone! Today's focus is on demonstrating how to visualize a knowledge graph that you've created using a large language model (LLM), which you might be using for Graph Retrieval-Augmented Generation (Graph RAG). Let's dive in!

Prerequisites

Before jumping into the visualization part, I assume you are already familiar with creating knowledge graphs using LangChain. If not, make sure to check out my previous video on knowledge graph creation using LangChain.

First, install the necessary libraries via pip:

pip install langchain google-generativeai langchain-core JsonRepair networkx

Libraries and Versions

LangChain: Specific version
Google Generative AI: Specific version if using the free API key
LangChain Core
JsonRepair
NetworkX: This will be particularly important for visualization.

Creating the Knowledge Graph

We will be using the Google Generative AI and LangChain libraries to set up the text for our knowledge graph.

Here is an example text about Marie Curie:

Marie Curie, born in 1867, was a physicist and chemist who was awarded the Nobel Prize.

We will transform this text into a knowledge graph using the convert_to_graph_documents function from the LangChain library.

Saving Relationships to a DataFrame

Save the relationships formed from the graph document into a DataFrame with columns: node1, node2, and relationship. Here's a basic skeleton of what the DataFrame might look like:

node1	node2	relationship
Marie Curie	1867	born in
Marie Curie	physicist	occupation
Marie Curie	Nobel Prize	awarded Nobel Prize

Visualizing the Knowledge Graph

Using NetworkX

We'll be utilizing the NetworkX library for graph visualization. Follow these steps:

Create a Directed Graph Object:
```
import networkx as nx
G = nx.DiGraph()
```
Add Nodes and Edges to the Graph:

Read each row in the DataFrame and add the nodes (node1, node2) and the relationship as the edge.
```
for index, row in df.iterrows():
    G.add_edge(row['node1'], row['node2'], relationship=row['relationship'])
```

Set Up Layout for Visualization:

import matplotlib.pyplot as plt
pos = nx.spring_layout(G)

Drawing the Graph:

edge_labels = nx.get_edge_attributes(G, 'relationship')

plt.figure(figsize=(12, 8))
nx.draw(G, pos, with_labels=True, node_color='skyblue', node_size=2000, edge_color='gray', font_size=15, width=2, arrows=True)

nx.draw_networkx_edge_labels(G, pos, edge_labels=edge_labels, font_color='red')
plt.title("Graph Visualization")
plt.show()

Conclusion

And that's how you can visualize a knowledge graph created using an LLM! We’ve covered the setup, conversion into a DataFrame, and visualization using NetworkX. I hope this tutorial helps you extend your work to build a Graph RAG and more.

Thank you for your time. The Graph RAG crash course is out now, which includes all the codes, a theory eBook, and links to video tutorials. Check out the link for the course in the description below.

Keywords

Knowledge Graph
LangChain
NetworkX
Large Language Model (LLM)
Text Visualization
Graph RAG
Google Generative AI
DataFrame
Directed Graph
Graph Visualization
Python

FAQ

1. What libraries do I need to install for visualizing a knowledge graph?

You need to install langchain, google-generativeai, langchain-core, JsonRepair, and networkx.

2. How do I create a knowledge graph using LangChain?

You can create a knowledge graph by using Google Generative AI and LangChain libraries, and the convert_to_graph_documents function.

3. How can I save the relationships in the knowledge graph?

Save the relationships into a DataFrame with columns node1, node2, and relationship.

4. What library is used for visualizing the graph?

We use NetworkX for visualizing the graph.

5. What is an edge in the context of NetworkX?

An edge in NetworkX has three components: the two nodes it connects and the relationship, which serves as the edge label or name.