What is the difference between vector RAG and your RAG?

[mjyoon-official]

Even though it uses Neo4j as a graph database and build a knowledge graph,
it still perform vector embeddings and rely primarily on embedding-based search.

If that is the case, I would not use Neo4j but rather more specialized vector databases like Milvus or Pinecone.

Can this still be considered a GraphRAG?

[stellasia]

Hi @mjyoon-official ,

This package provides several retrievers: the VectorRetriever is a pure similarity based retriever, that only uses the Neo4j vector index and indeed doesn't add any context compared to other vector stores. On top of that , it also has VectorCypherRetriever and Text2CypherRetriever in which you can traverse the graph:

• the VectorCypherRetriever starts with a similarity search query to find relevant nodes, but they run a user-defined Cypher query to extend the context based on the connected nodes
• the Text2CypherRetriever is a pure Cypher/graph retriever, where no vector similarity is involved.

Hope that helps.