Find answers from the community

Updated 4 months ago

@Logan M sounds like you have some

At a glance
sounds like you have some insight? I work in biology and my fellow scientists keep talking about knowledge graphs and the only use I can see for it is better RAG answers, and even then its use seems rather limited there
L
S
9 comments
ngl I think knowledge graphs are not really prime-time ready. Maybe if you have one already existing, you can use an LLM to query your knowledge graph with templates or text-to-cypher. But letting an LLM traverse a graph is not really feasible (slow and expensive)

If you don't have an existing knowledge graph, its not really feasible to let an LLM build one (also slow an expesnsive, and error prone)

In my opinion, vector search + keyword (or similar, like sparse vector) search + reranking will get you much further and scales much better. Add on top a little bit of query understanding (routing, question generation), and you are golden.
this is all my personal ๐ŸŒถ๏ธ opinion. Maybe someday they will get there, but lots of work to do
Is building a knowledge graph really that hard? I feel like they are very difficult to build
they can be hard -- takes a lot of manual curation, heuristics, etc. to have a good graph if you are just trying to create a graph from unstructured data
I agree, vector search and a good internet search should give you your answer most of the time.
People seem to underestimate the time and expertise needed to construct these graphs.
Reminds me of this meme
lol its very true
Add a reply
Sign up and join the conversation on Discord