A Vector Embedding Index Search Guide

This guide will show how run a search against a vector embedding index.

Create an index

Create a vector embedding index if it does not already exist. In this example we will create an index for the sentence embedding stored in the `model2vec_embeddings` field.

`model2vec_embeddings`

// embeddings indexes
CREATE VECTOR INDEX pubmed_document_model2vec_embeddings_index IF NOT EXISTS
    FOR (n:pubmed_document)
    ON n.model2vec_embeddings
    OPTIONS { indexConfig: {
        `vector.dimensions`: 256,
        `vector.similarity_function`: 'cosine'
    }
};

CREATE VECTOR INDEX drug_model2vec_embeddings_index IF NOT EXISTS
    FOR (n:drug)
    ON n.model2vec_embeddings
    OPTIONS { indexConfig: {
        `vector.dimensions`: 256,
        `vector.similarity_function`: 'cosine'
    }
};

CREATE VECTOR INDEX disease_model2vec_embeddings_index IF NOT EXISTS
    FOR (n:disease)
    ON n.model2vec_embeddings
    OPTIONS { indexConfig: {
        `vector.dimensions`: 256,
        `vector.similarity_function`: 'cosine'
    }
};

CREATE VECTOR INDEX gene_protein_model2vec_embeddings_index IF NOT EXISTS
    FOR (n:gene_protein)
    ON n.model2vec_embeddings
    OPTIONS { indexConfig: {
        `vector.dimensions`: 256,
        `vector.similarity_function`: 'cosine'
    }
};

CREATE VECTOR INDEX anatomy_model2vec_embeddings_index IF NOT EXISTS
    FOR (n:anatomy)
    ON n.model2vec_embeddings
    OPTIONS { indexConfig: {
        `vector.dimensions`: 256,
        `vector.similarity_function`: 'cosine'
    }
};

Create vector index is not exists

Show vector indexes

Create a vector embedding index if it does not already exist. In this example we will create an index for the sentence embedding stored in the `model2vec_embeddings` field.

Show vector indexes if not exists. Look for `online` status and not `populating`

SHOW VECTOR INDEXES;

Show Vector Indexes

Query Similar Embeddings

Query similar embeddings with the `queryNodes` command. The `model2vec_embeddings` is a merged embedding generated from the article `title` and `keywords` fields.

In this example we query for a "graft verse host disease" article.

MATCH (n:pubmed_document { pmcid: '11759061' })
CALL db.index.vector.queryNodes('pubmed_document_model2vec_embeddings_index', 25, n.model2vec_embeddings)
YIELD node AS article, score
RETURN score, article.pmcid AS pmcid, article.keywords as keywords, article.title AS title;

graft-vs-host-disease