Naive RAG
A basic form of RAG system employing a simple retrieve-concatenate-generate pipeline. While capable of providing fundamental functionality, it exhibits notable shortcomings in precision and retrieval quality.
Timeline (last 90 days)
The typical steps of a Naive RAG pipeline are: chunk the documents, create embeddings, take top-k by cosine similarity, stuff them into the prompt, and let the model answer.
In Anthropic's tests, the failure rate of Naive RAG (the correct text chunk not in the top 20) was 5.7%, dropping to 3.7% with contextual embeddings, 2.9% with contextual BM25, and 1.9% with reranking added.
The Naive RAG workflow is: convert the user query into a vector, perform a single vector similarity search, retrieve the most relevant chunks, and then feed the context together with the original question into an LLM to generate an answer.
The advantages of Naive RAG lie in its simplicity of implementation and fast response, making it suitable for prototype validation and rapid demonstrations.
The shortcomings of Naive RAG include: single vector retrieval tends to overlook exact keyword matching, there is no metadata filtering, and the lack of reranking results in insufficiently refined relevance ordering.
Naive RAG is suitable for early validation and small knowledge base projects, while Advanced Hybrid RAG is suitable to introduce when entering the production stage with high requirements for accuracy and traceability.
All Facts (6)
The typical steps of a Naive RAG pipeline are: chunk the documents, create embeddings, take top-k by cosine similarity, stuff them into the prompt, and let the model answer.
50%UnverifiedIn Anthropic's tests, the failure rate of Naive RAG (the correct text chunk not in the top 20) was 5.7%, dropping to 3.7% with contextual embeddings, 2.9% with contextual BM25, and 1.9% with reranking added.
50%UnverifiedThe Naive RAG workflow is: convert the user query into a vector, perform a single vector similarity search, retrieve the most relevant chunks, and then feed the context together with the original question into an LLM to generate an answer.
50%UnverifiedThe advantages of Naive RAG lie in its simplicity of implementation and fast response, making it suitable for prototype validation and rapid demonstrations.
50%UnverifiedThe shortcomings of Naive RAG include: single vector retrieval tends to overlook exact keyword matching, there is no metadata filtering, and the lack of reranking results in insufficiently refined relevance ordering.
50%UnverifiedNaive RAG is suitable for early validation and small knowledge base projects, while Advanced Hybrid RAG is suitable to introduce when entering the production stage with high requirements for accuracy and traceability.
50%