Knowledge graphs and Neo4j

The big idea

A knowledge graph stores facts as things and the links between them. Instead of a sentence like "James is Emily's father", you store two nodes (James, Emily) and one relationship (James → Father → Emily).

Because the links are stored directly, you can follow them: from Emily to her father, from her father to his parents, and so on.

Three building blocks

Talking to Neo4j: Cypher in two minutes

Neo4j is a graph database. You ask it questions in Cypher, which draws the pattern you want with brackets and arrows:

• (a) is a node, -[r]-> is a relationship.
• MATCH describes the pattern to find, RETURN says what to show.

MATCH (a:GraphNode)-[r:RELATES_TO]->(b:GraphNode)
RETURN a.name, r.relationship, b.name
LIMIT 10

Read it as: find any node linked to another node, and show both names and the link's label.

Where it lives in GraphRAG Chats

Every graph you draw is saved in Neo4j like this:

• One SavedGraph node holds the graph's name.
• It links to each of your boxes with HAS_NODE. Each box is a GraphNode with a name, type, description and properties.
• Your arrows are RELATES_TO relationships, with your label stored as relationship.

The Cypher lives in backend/src/backend/neo4j_client.py.

Practice: build and query your own graph

1. On the graphs page, click + New graph and name it My team.

2. Add four nodes: three people and one project. Use Add node for the first one and the + on a node for the rest. Give each a type (Person, Project) and a one-line description.

3. Connect them with labelled arrows, such as Works on and Manages.

4. Click Save graph.

5. In the Neo4j Browser, run:

   cypherCopy

   MATCH (g:SavedGraph {name: 'My team'})-[:HAS_NODE]->(a)-[r:RELATES_TO]->(b)
   RETURN a.name, r.relationship, b.name
   

You are done when Neo4j lists the same links you drew on the canvas.

Check yourself

1. Name the three building blocks of a graph.
2. In Cypher, what does -[r]-> stand for?
3. Where does the app store the label you type on an arrow?