How to pass a filtering query to an AzureVectorIndexRetriever

[AsmaaMHadir]

I use Azure Index as my vector database, and use the AzureSearchVectorStoreRetriever as my retriever in my Langgraph rag chain.
I have a requirement of retrieving chunks belonging to a specific document, i.e having a specific source metadata field. I attempt to pass a filter query using the following line :
azureIndex = AzureVectorIndex() retriever = azureIndex.get_retriever( search_kwargs={ "filter": { "source": "source_doc" } } )
However, I get this error: TypeError: AzureVectorIndex.get_retriever() got an unexpected keyword argument 'search_kwargs'

I also attempt to use an Odata filtering query to the retriever but I constantly get an invalid query. Can you give me a sample correct Odata query to filter the retrieval and only run a retrieval/similarity search over documents from a specified source doc? how do I correctly implement this filtering?

retriever = AzureAISearchRetriever( content_key="content", top_k=5, index_name=index_name, service_name=os.getenv("AZUREAI_SERVICE_NAME"), api_key=os.getenv("VECTOR_STORE_PASSWORD"), filter="search.in(source,'')" )

[dosubot[bot]]

To correctly implement a filtering query in AzureVectorIndexRetriever for retrieving chunks from a specific document using a source metadata field, you can use the pre_filter parameter in the similarity_search method. Here's a sample OData query for this purpose:

query = "What did the president say about Ketanji Brown Jackson"
docs = vectorstore.similarity_search(
    query,
    pre_filter={"metadata.source": {"$eq": "specific_document_name.txt"}},
)

This query filters the results to only include chunks from the document with the specified name in the metadata.source field ^[1].

Regarding the error you encountered with AzureVectorIndex.get_retriever(), it seems that the method does not accept search_kwargs as a parameter. You might need to check the method's documentation or implementation to see the correct parameters it accepts. Unfortunately, I couldn't find specific details about the available parameters for this method in the repository.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}

[AsmaaMHadir]

What worked for me is to use
vector_store.similarity_search(query,filters='field eq 'value'')
Such that vector_store is an AzureSearch object