Skip to content

AI Glossary · Concepts

What is Knowledge Graph?

Knowledge Graph

A knowledge graph stores information as entities and the relationships between them, rather than as loose documents or vectors. For AI, it gives a model structured, connected context it can traverse — which is one answer to grounding and hallucination.

Also known as: knowledge graphs

· Updated · Chain of Thought

RAG & Retrieval

  • 2 explainers
  • 6 episodes
  • 2 guest clips
  • 2 sources

Most retrieval today is vector-based: chunk documents, embed them, fetch the nearest. A knowledge graph takes a different shape — it represents data as nodes (entities like people, products, accounts) and edges (the relationships between them). That structure lets a system answer questions that depend on connections, not just similarity, and traverse from one fact to related ones.

For grounding AI, graphs and vectors are complementary. Vector retrieval is great at “find me text that’s about this”; a graph is great at “what’s connected to this, and how.” Using a graph as the retrieval layer is often called graph RAG, and some systems combine it with vector search. Microsoft Research’s GraphRAG paper describes a well-known version: a model extracts an entity graph from the documents, summarizes each cluster of closely related entities, and answers from those summaries rather than from retrieved chunks. (The released GraphRAG library also has a local search mode that finds starting entities by vector similarity.) Its authors built it for whole-corpus questions such as “what are the main themes in this dataset?”, which retrieving a handful of similar chunks can’t answer.

Knowledge graph: multi-hop reasoning follows relationships across data sets Two multi-hop paths from Neo4j's Sudhir Hasbe in episode 55. Anti-money laundering jumps from a legal entity to an account to a transaction. Fraud ring detection goes from a person to their phone, to another person linked to that phone, to that person's account, to a transaction, to a mule account. Each node can live in a different system; the graph stores the links between them. Below: vector search finds records that look similar to the question, while a graph follows the stated relationships from one record to the next. Multi-hop: follow the links, not the likeness Two paths Neo4j’s Sudhir Hasbe walked through in episode 55. Each hop can cross into another system. ANTI-MONEY LAUNDERING · 2 HOPS Legal entity Account Transaction hop 1 hop 2 FRAUD RING DETECTION · 5 HOPS Person Phone Linked person Account Transaction Mule account The question “is this a fraud ring?” depends on the chain, not on any one record. VECTOR SEARCH Finds records that look similar to the question. KNOWLEDGE GRAPH Follows the stated relationships from one record to the next. The two are often paired: find a starting point by similarity, then traverse from it.
Both paths are Sudhir Hasbe’s examples of multi-hop reasoning from episode 55. The answer depends on the chain of links, which is what a graph stores. Download the image

Hear it from the guest

“If you look at anti-money laundering, it's going to jump from the legal entity to account to transaction. So multi-hop reasoning is nothing but ability for you to go jump between all these disparate data sets that are in different systems and trying to rationalize between their relationships and make decisions.”
Sudhir Hasbe, Neo4j EP 55 · around 30:39 · read the transcript
Watch on YouTube
“Knowledge graph is all about entities and relationships. Now context graph is building on top of a knowledge graph.”
Jitender Aswani, Starburst EP 66 · around 47:40 · read the transcript
Watch on YouTube

Quotes lightly edited to remove filler words.

Sources

Go deeper

  • Neo4j: MATCH patterns in Cypher docs

    Trace how relationship types and direction determine the returned rows. The examples use a sample Neo4j graph; matching stored facts does not verify their accuracy.

From the conversation

GraphRAGSemantic SearchVector DatabaseAgentic RAG