Vector query filters using Azure AI Search and Langchain support

[druva-raj]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from langchain_community.retrievers import AzureAISearchRetriever

retriever = AzureAISearchRetriever(
    content_key="HotelId", top_k=5, index_name="hotels-vector-quickstart", filter=f"HotelName eq 'Nordick''s Motel'"
)

Description

I'm trying to implement vectorFilterMode of Azure AI Search using AzureAISearchRetriever. However, the parameter vectorFilterMode is not accepted currently. Is there a workaround to achieve this?

Using the OData filter in the retriever will give us the top K relevant documents based on the filter condition. But the next retrieval-answer generation step takes VectorStore as input but not the top K documents.

System Info

%pip install --upgrade --quiet langchain-community
%pip install --upgrade --quiet langchain-openai
%pip install --upgrade --quiet azure-search-documents>=11.4
%pip install --upgrade --quiet azure-identity

[druva-raj]

Since the retriever method doesn't accept the vectorFieldMode parameter, I've opted to use the vector search methods by setting up a vector store with AzureSearch. How can I verify if the argument vector_filter_mode is being included in the actual query

vector_store = AzureSearch(
azure_search_endpoint=endpoint,
azure_search_key=key_credential,
index_name=index_name,
embedding_function=embeddings.embed_query,
fields=fields,
vector_filter_mode="preFilter",
semantic_configuration_name="default"
)

And, also would like to understand how this is different from a simple OData filter?