QDRANT doc store with sparse retriever model like BM_25

[sinaayyy]

Hello,

I want to know if it's possible to use a sparse model retriever model with Qdrant ?

If it's not possible i want to know if i can join muliple document store because i want use 2 types of retievers : dense and sparse.
I wonder if i can use in a pipeline for example memorydocstore for sparse BM-25 retriever, Qdrant for dense retriever and connect it in the pipeline

[julian-risch]

@sinaayyy Yes, the QdrantDocumentStore supports BM25Retrievers, so you can do dense and sparse retrieval with this DocumentStore. You can find an example pipeline with a QdrantDocumentStore and a BM25Retriever here: https://github.com/qdrant/qdrant-haystack/blob/master/tests/integration/qdrant_haystack/document_stores/pipeline/test_haystack_pipeline.py
Using a dense retriever and a sparse retriever in combination is often referred to as hybrid search. This article might be interesting to you: https://qdrant.tech/articles/hybrid-search/
For hybrid retrieval with Haystack, we have a tutorial here: https://haystack.deepset.ai/tutorials/26_hybrid_retrieval
You can use the same DocumentStore and don't need to create separate ones for dense and sparse retrieval. However, even the latter would work if you switch to a DocumentStore that doesn't support both.