Performing semantic searches - ELSER

[sinhapunit (Punit)]

I am new to Elastic and I am sure my issue is very basic, if someone could give me high level pointers (or detailed instructions if they wish) it would be very helpful.

I have about 120,000 documents and each document has a property called "tx"

I want to perform semantic searches against the tx property of the document.

My questions

1. how do I setup the index and the mappings
2. how do I structure the documents for ingestion, does it need any additional property to support ELSER
3. after ingesting documents do I need to reindex?
4. how do I structure my query.

Thanks

[dadoonet (David Pilato)]

Welcome!

I think that a good starting page would be this one: Tutorial: semantic search with ELSER | Elasticsearch Guide [8.12] | Elastic

[iulia (Iulia Feroli)]

What language are you using? We have more content and examples in python - check out the notebooks here on searchlabs

To answer your specifics:

• If you already have your data in an index, you can create a new "target" index and use a pipeline to re-index all your documents with the new embedding values stored in a sparse-vector field.
• Alternatively, if your data comes from a different source, you can directly index it using ELSER enhancements in one go during the initial ingestion.
• For queries, you will be using the text_expansion to perform semantic search:

"text_expansion": {
                    "ml.tokens": {
                        "model_id":".elser_model_1",
                        "model_text":"looking for a specific document in here"
                    }
                },

Good luck!