Why knowledge graphs may not beat standard RAG
are expected to help connect related facts that use different words. For example, one law may ban stealing, while another may allow taking food in a true emergency.
can miss that link because the words do not match. may find the connection, but the needed text chunk can still fall outside the top-k window and never reach the answer.
sound like a stronger way to store and update relationships, but they do not always perform better than other RAG methods. The weak point may be that the LLM fails to extract enough complex relationships during indexing, or that the cannot bridge the wording gap when storing and searching information.
Key points
- can miss related rules when the wording is different.
- can still fail if the right chunk falls outside the top-k window.
- may not automatically outperform simpler RAG methods.
- Poor relationship during indexing can limit graph quality.
- The value for agents depends on whether accuracy improves while token use drops.