documentEmbedding

R2026b

Document embedding model to map documents to vectors

Since R2024a

expand all in page

    Description

    A document embedding maps documents to real vectors.

    The vectors attempt to capture the semantic content of the full document, so similar documents have similar vectors. The document can be a sentence, a paragraph, or a longer text.

    Creation

    Create a document embedding from a pretrained embedding using documentEmbedding.

    Syntax

    emb = documentEmbedding
    emb = documentEmbedding(Model=modelName)

    Description

    emb = documentEmbedding returns a document embedding using the all-MiniLM-L6-v2 sentence transformers model.

    This function requires Deep Learning Toolbox™. This syntax requires the Text Analytics Toolbox™ Model for all-MiniLM-L6-v2 Network support package.

    example

    emb = documentEmbedding(Model=modelName) returns the document embedding model specified by the Model name-value argument.

    Input Arguments

    expand all

    Document embedding model, specified as one of these values:

    • "all-MiniLM-L6-v2"— Sentence transformer model with six self-attention layers. This model outputs a 1-by-384 embedding vector. This option requires the Text Analytics Toolbox Model for all-MiniLM-L6-v2 Network support package.

    • "all-MiniLM-L12-v2"— Sentence transformer model with twelve self-attention layers. This model outputs a 1-by-384 embedding vector. This option requires the Text Analytics Toolbox Model for all-MiniLM-L12-v2 Network support package.

    If the required support package is not installed, then the function provides a download link.

    This argument sets the ModelName property.

    Data Types: string

    Properties

    expand all

    Since R2026b

    This property is read-only after object creation.

    Document embedding model, specified as one of these values:

    • "all-MiniLM-L6-v2"— Sentence transformer model with six self-attention layers. This model outputs a 1-by-384 embedding vector. This option requires the Text Analytics Toolbox Model for all-MiniLM-L6-v2 Network support package.

    • "all-MiniLM-L12-v2"— Sentence transformer model with twelve self-attention layers. This model outputs a 1-by-384 embedding vector. This option requires the Text Analytics Toolbox Model for all-MiniLM-L12-v2 Network support package.

    Data Types: string

    Since R2026b

    This property is read-only.

    Length of embedding vectors, specified as 384.

    In general, the length of the embedding vectors depends on the embedding model. For both of the two models that the documentEmbedding function supports, the length of the embedding vectors is 384.

    Object Functions

    embedMap document to embedding vector

    Examples

    collapse all

    This example uses:

    Open Live Script

    Load the pretrained document embedding all-MiniLM-L6-v2 using the documentEmbedding function. This model requires the Text Analytics Toolbox™ Model for all-MiniLM-L6-v2 Network support package. If this support package is not installed, then the function provides a download link.

    emb = documentEmbedding;

    Create an array of input documents.

    documents = [
    "the quick brown fox jumped over the lazy dog"
    "the fast brown fox jumped over the lazy dog"
    "the lazy dog sat there and did nothing"];

    Map the input documents to vectors using the embed function.

    embeddedDocuments = embed(emb,documents);

    To estimate how similar the documents are, compute the pairwise cosine similarities using cosineSimilarity.

    similarities = cosineSimilarity(embeddedDocuments)
    similarities = 3×3

    1.0000    0.9840    0.5505
    0.9840    1.0000    0.5524
    0.5505    0.5524    1.0000

    References

    [1] Reimers, Nils, and Iryna Gurevych. "Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks" Preprint, submitted August 27, 2019. https://doi.org/10.48550/arXiv.1908.10084.

    Version History

    Introduced in R2024a

    expand all

    See Also

    | | | |

    Topics