Graphs and RAG

Graph RAG with LangChain and Neo4j

Adam Cowley

Manager, Developer Education at Neo4j

Meet your instructor

Manager, Developer Education at Neo4j
Developer Education through Neo4j GraphAcademy
20+ years of software development experience
10+ years Neo4j experience

Adam's picture

¹ https://graphacademy.neo4j.com

The R in RAG

An application diagram with a User input being added to a prompt that is sent to an LLM

The R in RAG

The prompt requires context that comes from documents and data sources

The R in RAG

This data is often converted into vector embeddings

Where does semantic search fail?

Vectors work well for

Fuzzy or Open-Ended questions

Vectors are ineffective for

Highly Specific or Fact-Based Questions
Numerical or Exact-Match Queries
Logical Queries

What does Paul Graham think about Generative AI?

How many Generative AI Startups has Paul Graham invested in?

Vectors vs. knowledge graphs

Text	Vector Embedding
He's just a city boy born	`[0.12, -0.34, 0.56, 0.78, ..., -0.91]`
born and raised in South Detroit	`[0.22, 0.45, -0.67, 0.11, ..., 0.33]`
He 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]`

A graph diagram showing the entities mentioned in the lyrics for Don't stop believing. A small town boy, with BORN and RAISED relationships to South Detroit

Vectors vs. knowledge graphs

Text	Vector Embedding
He's just a city boy born	`[0.12, -0.34, 0.56, 0.78, ..., -0.91]`
born and raised in South Detroit	`[0.22, 0.45, -0.67, 0.11, ..., 0.33]`
He 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]`

The nodes are a person: Small town boy, a location: South Detroit, a train: Midnight Train and a location: Anywhere

Vectors vs. knowledge graphs

Text	Vector Embedding
He's just a city boy born	`[0.12, -0.34, 0.56, 0.78, ..., -0.91]`
born and raised in South Detroit	`[0.22, 0.45, -0.67, 0.11, ..., 0.33]`
He 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]`

The relationships BORN and RAISED go from the person to the location. The GOING train goes from the Midnight Train node to the Anywhere node

Knowledge graphs and Neo4j

Neo4j is the world's leading graph database
Flexible, schema-optional
langchain-neo4j LangChain integration
LCEL: LangChain Expression Language

The neo4j and langchain logos

¹ https://db-engines.com/en/ranking/graph%20dbms

Nodes represent things

from langchain_neo4j.graphs.graph_document \
  import Node


book = Node(

    type="Book",

    id=f"9781098127107",

    properties={
        "title": "Building Knowledge Graphs"
    }

)

A book node in a graph diagram with properties for ID and Title

Relationships connect nodes

from langchain_neo4j.graphs.graph_document \
  import Node, Relationship

book = Node(type="Book", id=f"9781098127107")

jesus = Node(type="Person", id="dr-jesus-barrasa")
jim = Node(type="Person", id="chief-scientist")

for author in [jesus, jim]:
    relationship = Relationship(









    )

Two WRITTEN_BY relationships going from a Book node to two Person nodes that represent the authors

Relationships connect nodes

from langchain_neo4j.graphs.graph_document \
  import Node, Relationship

book = Node(type="Book", id=f"9781098127107")

jesus = Node(type="Person", id="dr-jesus-barrasa")
jim = Node(type="Person", id="chief-scientist")

for author in [jesus, jim]:
    relationship = Relationship(

        source=book,







    )

The source node of the WRITTEN_BY relationship is the Book node

Relationships connect nodes

from langchain_neo4j.graphs.graph_document \
  import Node, Relationship

book = Node(type="Book", id=f"9781098127107")

jesus = Node(type="Person", id="dr-jesus-barrasa")
jim = Node(type="Person", id="chief-scientist")

for author in [jesus, jim]:
    relationship = Relationship(

        source=book,

        target=author,





    )

The destination node for a WRITTEN_IN relationship is a Person

Relationships connect nodes

from langchain_neo4j.graphs.graph_document \
  import Node, Relationship

book = Node(type="Book", id=f"9781098127107")

jesus = Node(type="Person", id="dr-jesus-barrasa")
jim = Node(type="Person", id="chief-scientist")

for author in [jesus, jim]:
    relationship = Relationship(

        source=book,

        target=author,

        type="WRITTEN_BY"

        properties=dict(...)

    )

The node diagram including the Book and two WRITTEN_BY relationships to authors to demonstrate that nodes can have many relationships.

Connecting to Neo4j

from langchain_neo4j import Neo4jGraph


graph = Neo4jGraph(
    url=NEO4J_URI, 
    username=NEO4J_USERNAME, 
    password=NEO4J_PASSWORD
)

¹ https://neo4j.com/docs/getting-started/

Saving graph documents

from langchain_neo4j.graphs.graph_document import GraphDocument


# Wrapper for Nodes and Relationships
doc = GraphDocument(

  nodes=[book, jesus, jim],

  relationships=[jesus_wrote_book, jim_wrote_book]

)


# Save nodes and relationships to the database
graph.add_graph_documents([graph_document])

So, what is GraphRAG?

The previous diagram with context from documents and their embeddings

So, what is GraphRAG?

The previous diagram with context from documents with an added knowledge graph showing nodes and relationships derived from the documents

Let's practice!

Graph RAG with LangChain and Neo4j