We have close to 11 million objects in class
Article and each of those objects has a
Metadata as cross reference. Now when we try to do hybrid search on
Article class with filters on
hasMetadata and a vector for semantic search on
Article it is taking on an average 30 seconds with some queries taking upto 2 minutes. is there anyway we can speed up this?
Hi @vamsi - am I understanding that you’re filtering by cross-referenced properties? As in, properties of the cross-referenced objects?
I think that’s what’s slowing you down, because each cross-reference has to be resolved for filtering, but I will confirm that with the team.
If that is the case, is it possible to build in to the
Article class the properties that you want to search for?
Could you also share with us which Weaviate version are you using?
Yes. since there can be a lot of article objects with same metadata, we split into two separate classes.
The thing is there are close to 50 properties in metadata. We can’t finalize properties to search on because the use cases might change in future and we might need to filter on more properties.
Hey @jphwang did you get a chance to check with the team why filtering by cross-referenced properties is slow and if there are any possible workarounds/resolutions for users?