Overview of an intelligent recommendation system in Neo4j GenAI ecosystem
Let us look at how recommendation systems that are built on LLM/RAG principles would function in the Neo4j GenAI ecosystem (Figure 7.1).
Figure 7.1 — Neo4j RAG recommendation architecture
We can leverage the features of these frameworks to build RAG applications backed by knowledge graphs. In this architecture, we are leveraging the Spring AI app to augment the graph to be able to provide more personal recommendations.
Also, for RAG, this architecture can leverage the vector indices as well as graph traversal to augment the response, to get the best of both worlds to get more accurate responses. This concept is called Graph RAG. Knowledge graphs can bring more accurate responses, rich context, and explainability for AI model interactions. Neo4j can integrate into LangChain4j and Spring AI to act as a vector store as well as a graph database to augment the LLM responses.
