Hybrid search capability

[thusithaC]

Is there a plan to introduce Hybrid search (using bm25 or similar algorithm). If not in the pipeline, can you please give some pointers in how best to implement it using the existing patterns within this library?

Thanks!

[averikitsch]

Hi @thusithaC , we are planning updates to support Hybrid Search (for dense and sparse vectors). I currently do not have a timeline for support. Currently, you could do this using the BM25Retriever and EnsembleRetriever, but the BM25Retriever will use sparse vectors stored in-memory:

from langchain_community.retrievers import BM25Retriever
from langchain.retrievers import EnsembleRetriever


store = await AlloyDBVectorStore.create(...)
# store.add_documents(docs)
retriever = store.as_retriever()
bm25_retriever = BM25Retriever.from_documents(docs)

ensemble_retriever = EnsembleRetriever(
    retrievers=[bm25_retriever retriever], weights=[0.5, 0.5]
)