I tried and failed to create one—but then LLMs weren't a thing!
At first, knowledge graphs (KGs) sounded strange—not the concept, but the process of building them.
I have tried to build a knowledge graph before and failed.
Graphs are arguably one of the best ways to represent complex relationships. They have many uses, such as referral programs and fraud detection. However, what really interested me was finding information.
I started using knowledge graphs to build better RAGs.
RAGs don't really want information graphs. They don't want a database at all. As long as you can extract the relevant information from the larger pool and transfer it to the LLM context, RAGs work.
You can build RAG with web search as its information retrieval strategy or use a vector store to take advantage of its semantic text search features.
When using a graph database to retrieve context information, we call it GraphRAG.
This is not a post about GraphRAGs (Maybe in a future post). This is about the construction of knowledge graphs themselves using LLMs. But it's worth saying…
