The Retrieval-Augmented Generation (RAG) Process
The RAG process typically involves three main steps:
-
Understanding the User Query
The system first interprets the user’s input or question to determine what information is needed.
-
Information Retrieval
A retriever searches external data sources (such as documents, databases, or knowledge graphs) to find relevant information based on the user’s query.
-
Response Generation
The retrieved information is inserted into the prompt, and the language model uses this context to generate a more accurate and relevant response.
The RAG Process
RAG systems can provide responses that are both contextually aware and grounded in real, up-to-date information.
If building a chatbot for a news agency, you could use RAG to pull real-time headlines or results from a news API.
When a user asks, "What’s the latest news on the Olympics?", the chatbot, can provide a current headline or summary from the most recent articles, ensuring the response is timely and accurate.
Grounding
The process of providing context to an LLM to improve the accuracy of its responses and reduce the likelihood of hallucinations is known as Grounding.
Data sources
The news agency chatbot could use the following data sources:
-
A news API to retrieve the latest articles or headlines.
-
A knowledge graph to understand the relationships between different news topics, such as how they relate to each other or their historical context. This would help the chatbot provide more in-depth and contextual responses.
-
A document database to store and retrieve articles, reports, or other textual data that can be used to answer user queries.
Learn more about knowledge graphs
You will learn more about knowledge graphs and their construction in the next module.
Graph-Enhanced Vector Search
A common approach to GraphRAG is to use a combination of vector search and graph traversal.
This allows for the retrieval of relevant documents based on semantic similarity, followed by a graph traversal to find related entities or concepts.
The high-level process is as follows:
-
A user submits a query.
-
The system uses a vector search to find nodes similar to the users query.
-
The graph is then traversed to find related nodes or entities.
-
The entities and relationships are added to the context for the LLM.
-
The related data could also be scored based on its relevance to the user query.
Text to Cypher
Text to Cypher is an alternative approach in GraphRAG that allows users to express their information needs in natural language, which is then automatically translated into Cypher queries.
You leverage the power of LLMs to interpret user intent and generate precise graph queries, enabling direct access to structured data and relationships within the graph.
You can use text to Cypher to turn users' queries into complex searches, aggregations, or traversals, making advanced graph querying more accessible and flexible.
Text to Cypher works by passing the user’s query and the graph schema to an LLM, which generates a Cypher query that can be executed against the graph database.
Exercise caution with LLM-generated queries
Caution should always be taken when executing LLM-generated Cypher queries, as they may not always be safe or efficient.
You are trusting the generation of Cypher to the LLM. It may generate invalid Cypher queries that could corrupt data in the graph or provide access to sensitive information.
In a production environment, you should ensure that access to data is limited, and sufficient security is in place to prevent malicious queries.
Lesson Summary
In this lesson, you learned about RAG and GraphRAG techniques, and how they can be used to enhance information retrieval.
In the next lesson, you will learn about knowledge graphs and how they represent real-world entities and their relationships.