Knowledge Graph-based RAG

KG-based retrieval-augmented generation (RAG) process

The KG-based retrieval-augmented generation (RAG) process begins when a user submits a query through the chatbot. The query text is first processed by an LLM to extract relevant entities. These entities are then used to query the Neo4j knowledge graph using a Cypher query to match entities, retrieving the associated relationship tuples. The original query is augmented by combining it with the extracted entities and relationships. This augmented query, containing the user’s input and the retrieved context, is passed to the LLM to generate a final response, as illustrated in the following slide deck:

Below, we implement the chatbot we designed above. Our chatbot will generate responses based on the context retrieved from the knowledge graph stored in Neo4j. The implementation follows the same steps shown in the above illustration, so it will be easy to follow. We’ll explain the new code and skip the details of the code we've already covered in previous lessons, which we’re reusing here.

Implementing a graph RAG-powered chatbot

We are already familiar with most of the concepts, including using an LLM to construct a knowledge graph. Now, we’ll extend this to extract entities from a user’s query and use them to generate responses with additional context provided by the knowledge graph.

We also know how to use Cypher queries to store and retrieve information from a Neo4j database. In this implementation, we’ll use Cypher queries to retrieve relationships relevant to the user’s query, augmenting it with knowledge from the graph.

We have broken down our chatbot application into three files (chat_app.py ...