Like operator performance

Hello,

In one of our Weaviate-based projects we perform a where query with the Like operator for some fields (title, text) over a large set of objects. As the amount of objects stored in Weaviate for this particular class grows, this query becomes very slow (15s-20s).

There is a note in the documentation specifically for that, but maybe there are some alternatives to this search method? Basically it needs to be a full-text search (or even simple substring-matched search), but it needs to be able to match parts of words as well. We tried BM25 and Hybrid search but they don’t seem to properly match parts of the words, only the full words.

Hi @evenfrost - what version of Weaviate are you running?

We made some changes with 1.19 that should speed up filtering, so I wonder if you can give that a go as a first try.

Hi @jphwang ,

We’re currently running semitechnologies/weaviate:1.19.6 image on production (this is where the slowdown happened).

Hi @evenfrost, how many objects in your store ?

@IvankoPo around 100,000.

Hello all, any update on this?