News for hackers. Stuff that matters.

A comment on Do we think about vector dbs wrong?. Parent comment

Re:Do we think about vector dbs wrong? by J_Shelby_J on Tuesday September 05, 2023 @03:08PM (#37396307)
Speaking of tools, I want to add keyword search to my CEQ project. I’m currently just using pinecone for retrieval, but I’m not always getting results for specific niche technical terms. Do you have any advice?
[ Parent | on HN ]
Re:Do we think about vector dbs wrong? by aslPine on Tuesday September 05, 2023 @03:33PM (#37396624)
If you want to add true keyword search (i.e. traditional BM25 matching-type search capabilities), you should try hybrid search. LMK if you have issues! https://docs.pinecone.io/docs/hybrid-search
[ Parent | on HN ]
Re:Do we think about vector dbs wrong? by aslPine on Tuesday September 05, 2023 @03:32PM (#37396605)
Definitely!

Have you tried adding whatever terms you want retrieved back as 'metadata' attached to your vectors?

https://docs.pinecone.io/docs/metadata-filtering

[ Parent | on HN ]
Re:Do we think about vector dbs wrong? by J_Shelby_J on Thursday September 07, 2023 @06:06PM (#37426246)
Oh my god. I didn’t realize you could use the $in clause like that. I already have the text content indexed so this will work nicely. Embarrassing, but thank you!
[ Parent | on HN ]