Graph RAG finds what's similar. We should aim for what's relevant

Posted by hjeffery 2 hours ago

6611Original

Comments

Comment by DylanScott 1 hour ago

I have come across many benchmark results of the rag solution, all claiming to be superior.

Comment by KelvinKings 1 hour ago

That’s why you have to tryem out, most don’t mention the ones that are better, or are simply built for jsut one test.

Comment by MirandaWei 1 hour ago

[dead]

Comment by wanderwall 1 hour ago

Is the construction cost of Graph RAG really that high? Why is the query speed of Graph RAG so slow? Is Graph RAG difficult to maintain in actual projects?

Comment by Ringowester 1 hour ago

Token spending is a problem. But you’ll have to use graph rag if you wanna use rag.

Comment by hogrider666 1 hour ago

Async memory writes aren’t the actual problem.

Comment by hogrider666 1 hour ago

I just came across this on a WeChat official account a couple of days ago, really fresh idea.

Comment by cole_code81 1 hour ago

used Pinecone for vector search, quite glad to see focus on relevance with Graph RAG. Heard of many memory engines these days, hope it actually improves context retrieval.

Comment by salkahfi 1 hour ago

60+ points but still 2 karma? @dang

Comment by caseyCrow31 1 hour ago

is the relevance scoring based on graph embeddings or just keyword matching?

Comment by wosunheizhu 1 hour ago

This project doesn't even use keyword matching.

Comment by hjeffery 1 hour ago

There is, in preprocessing, for coreference

Comment by dk970 1 hour ago

[dead]

Comment by hjeffery 2 hours ago

[dead]

Comment by LebronJordan 2 hours ago

[dead]

Comment by Featbeat 2 hours ago

[dead]

Comment by PeterChengs 2 hours ago

[dead]

Comment by alexzhu1 2 hours ago

[dead]

Comment by morgan-rust-60 1 hour ago

[dead]

Comment by nohandsgamer 1 hour ago

[dead]

Comment by goldenhourhhh 1 hour ago

[dead]