How do I match reference with near_text or hybrid

Support
1

Hi all,

I currently have a problem using near_text with reference
I want to match my query with the reference chunk I have for example

response = building_collection.query.near_text(
                query="Green garden city",
                target_vector="buildingDetails", // namedVector in chunk class
                filters=Filter.all_of(filter_array) if len(filter_array) > 0 else None,
                limit=limit,
                offset=offset,
                return_references=[
                    QueryReference(
                        include_vector=True,
                        link_on="hasChunks"
                    ),
                ],
            )

this is my class definition

def create_buildings_vectordb_schema(client: WeaviateClient, logger: LoggerInterface) -> None:
    collection_name = BUILDINGS_COLLECTION_NAME    
    if not client.collections.exists(collection_name):
        new_collection = client.collections.create(
            name=collection_name,
            properties=[
                wvc.config.Property(
                    name="buildingTitle",
                    data_type=wvc.config.DataType.TEXT,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
                wvc.config.Property(
                    name="buildingAddress",
                    data_type=wvc.config.DataType.TEXT,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
                wvc.config.Property(
                    name="buildingDescription",
                    data_type=wvc.config.DataType.TEXT,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
                wvc.config.Property(
                    name="housingPrice",
                    data_type=wvc.config.DataType.NUMBER,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
                wvc.config.Property(
                    name="ownerName",
                    data_type=wvc.config.DataType.TEXT,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
                wvc.config.Property(
                    name="ownerEmail",
                    data_type=wvc.config.DataType.TEXT,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
                wvc.config.Property(
                    name="ownerWhatsapp",
                    data_type=wvc.config.DataType.TEXT,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
                wvc.config.Property(
                    name="ownerPhoneNumber",
                    data_type=wvc.config.DataType.TEXT,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
                wvc.config.Property(
                    name="imageURL",
                    data_type=wvc.config.DataType.TEXT,
                    vectorize_property_name=False,
                    skip_vectorization=True
                ),
            ]
        )
        
        logger.log_info(f"Successfully create collection: {new_collection}")

this is my class chunk definition

def create_building_chunks_vectordb_schema(client: WeaviateClient, logger: LoggerInterface) -> None:
    collection_name = BUILDING_CHUNKS_COLLECTION_NAME  
    if not client.collections.exists(collection_name):
        new_collection = client.collections.create(
            name=collection_name,
            vectorizer_config=define_transformers(),
            generative_config=define_generative(),
            properties=[
                wvc.config.Property(
                    name="chunk",
                    data_type=wvc.config.DataType.TEXT,
                    tokenization=wvc.config.Tokenization.WORD,
                ),
            ],
        )
        
        # the parent class collection
        building_collection = client.collections.get(BUILDINGS_COLLECTION_NAME)
        building_collection.config.add_reference(
            wvc.config.ReferenceProperty(
                name="hasChunks",
                target_collection=BUILDING_CHUNKS_COLLECTION_NAME
            )
        )
        logger.log_info(f"Successfully create collection: {new_collection}")

I have one chunk that has value of “This building located near Green garden city”, but the query result is not even similar at all

I actually tried to do the query from the chunk collection, it works but I need 2 or more reference similarity so I can perform query like, “Near Green garden and has maid cafe in it”

im sorry if this is confusing but I hope I get some enlightenment thanks

2

hi @Fakhri_Prayatna_Putr !!! Welcome to our community :hugs:

Have you tried hybrid search instead?

On those cases, I believe a hybrid search with fit better, and you do want some similarity search, but have a key word of interest too.

Let me know if this helps!

3

so we can’t query the reference? cause that would be awesome if i can query the reference aswell instead of only having filter by reference, and also by for example if the references got similarity match, it also give the total similarity score to the parent class based on how similar the reference queried

i actually also querying it by hybrid but i think that’s not the case

4

hi is there any way for this?