What is GraphRAG?

Retrieval-Augmented Generation (RAG)

Retrieval-Augmented Generation (RAG) is a technique that improves the responses of LLMs by providing them with relevant, up-to-date information retrieved from external sources.

Retrieval-Augmented Generation (RAG) with Vectors

This typically involves converting text into vector embeddings that encodes the semantic meaning of the text, in a format that the user understands, and then using similarity search to find relevant information.

Vector-based RAG

Vectors work well for:

Contextual or Meaning Based Questions
Synonyms or Paraphrasing
Fuzzy or Vague queries
Broad or Open-Ended questions
Complex queries with multiple concepts

What does Paul Graham think about Generative AI?

Vectors are ineffective for:

Highly Specific or Fact-Based Questions
Numerical or Exact-Match Queries
Boolean or Logical Queries
Ambiguous or Unclear Queries without Context
Specialised Knowledge

How many Generative AI Startups has Paul Graham invested in?

Unstructured data as vectors

Text	Vector embedding
She’s just a small town girl	[0.12, -0.34, 0.56, 0.78, …, -0.91]
Living in a lonely world	[0.22, 0.45, -0.67, 0.11, …, 0.33]
She took the midnight train	[-0.55, 0.89, 0.12, -0.44, …, 0.67]
Going anywhere	[0.78, -0.23, 0.45, 0.91, …, -0.12]

Text

Vector embedding

She’s just a small town girl

[0.12, -0.34, 0.56, 0.78, …, -0.91]

Living in a lonely world

[0.22, 0.45, -0.67, 0.11, …, 0.33]

She took the midnight train

[-0.55, 0.89, 0.12, -0.44, …, 0.67]

Going anywhere

[0.78, -0.23, 0.45, 0.91, …, -0.12]

Unstructured data as a graph

Text	Vector embedding
She’s just a small town girl	[0.12, -0.34, 0.56, 0.78, …, -0.91]
Living in a lonely world	[0.22, 0.45, -0.67, 0.11, …, 0.33]
She took the midnight train	[-0.55, 0.89, 0.12, -0.44, …, 0.67]
Going anywhere	[0.78, -0.23, 0.45, 0.91, …, -0.12]

Text

Vector embedding

She’s just a small town girl

[0.12, -0.34, 0.56, 0.78, …, -0.91]

Living in a lonely world

[0.22, 0.45, -0.67, 0.11, …, 0.33]

She took the midnight train

[-0.55, 0.89, 0.12, -0.44, …, 0.67]

Going anywhere

[0.78, -0.23, 0.45, 0.91, …, -0.12]

Knowledge Graphs

GraphRAG involves creating a knowledge graph of nodes and relationships contained in unstructured data.

Nodes

Nodes represent things
Nodes are grouped by labels
Nodes are described by properties as key-value pairs

A boy with the description of City Boy
A Location with the name South Detroit.

Relationships

Any two nodes can be connected by a relationship
Each relationship has a type and a direction
Relationships have properties as key-value pairs

The City Boy took the Midnight Train going Anywhere.

Relationships describe how things are connected

Cypher

Cypher is a A Powerful & Expressive Query Language for querying graphs.

Steps of GraphRAG

The term Graph RAG encapsulates the process of extracting nodes and relationships from unstructured text, which sit along the vector embeddings in the knowledge graph.

The knowledge graph structure can be enriched with additional features derived from graph algorithms in the Neo4j Graph Data Science library, providing deeper insights into the data’s patterns and connections.

Then querying the resulting knowledge graph, sometimes in combination with vector search, to retrieve the necessary information for the task.

We will explore these points in more detail as we progress through the course.

Introduction to GraphRAG Workshop

Introduction to GraphRAG

Building Knowledge Graphs

Building GraphRAG Agents