What's the 'text_node_property' in Neo4jVector?

Neo4j Graph Platform Graph Algorithms/Graph Data Science
1

I am following this graph rag tutorial using Neo4j, and have question to the Neo4jVector's API:

https://api.python.langchain.com/en/latest/vectorstores/langchain_community.vectorstores.neo4j_vector.Neo4jVector.html

when creating a vector index:

vector_index = Neo4jVector.from_existing_graph(
        embedding,
        url=uri,
        username=user,
        password=password,
        index_name='vectorIndex',
        node_label="Product",
        text_node_properties=['name', 'description'],
        embedding_node_property='embedding',
    )

And after creating the index, to load the index:

vector_index = Neo4jVector.from_existing_index(
        embedding,
        url=uri,
        username=user,
        password=password,
        index_name='vectorIndex',
        text_node_property = 'name'
    )

My question is, when in loading the index, why is it a single 'text_node_property', not a list of properties of text when comparing this field to text_node_properties=['name', 'description'],?

2

So you can send multiple concatenated properties to the LLM to create the embedding, which is why you can create the vector index (first code block) with a list of properties. However, when you query the vector index, you can only pull an embedding from a single property (e.g. you couldn't pull multiple embeddings - one for name and one for description). Your second code block to load the vector index should probably look like below because when you create the index, you set the embedding_node_property to save the embedding value to a property called embedding:

vector_index = Neo4jVector.from_existing_index(
        embedding,
        url=uri,
        username=user,
        password=password,
        index_name='vectorIndex',
        text_node_property = 'embedding'
    )