Versions (relevant - OpenSearch/Dashboard/Server OS/Browser):
opensearch 2.15.0
npm i @opensearch-project/opensearch => “@opensearch-project/opensearch”: “^3.0.0-beta.3”
Describe the issue:
So i’ve been making the jump from elasticsearch to opensearch to get easier access to some of the more advanced ML features. I’ve been integrating text embeddings for the last few days and playing around with neural searches with different models and text embeddings, and while I’ve had some pretty good results on things like semantic search on all sorts of different fields, my ultimate goal is to be able to send in a query that looks something like this:
“show me videos from 2023” or “show me results from sept 2006”, and have opensearch return the appropriate results based off a property date like this:
mappings: {
properties: {
date: {
type: “date”,
format: “strict_date_optional_time||epoch_millis”
},
this is an example of the value the date field has:
2024-07-29T19:00:00.000+00:00
when i do neural search for any of the text fields, with a sentence transformer model, it seems to give me pretty decent semantic results. It can also return me hits that are FAIRLY accurate from the appropriate dates with queries like “show me results from 1965”, most of the returns will be 1965 hits. So perhaps I have achieved my outcome…
But i was wondering, before i went too far down this road, if this was an effective implementation for achieving that kindof conversational query, or perhaps if someone knows about more effective tools for this, whether it be the type of query or if different model types are more effective for this than sentence transformers.