Text embedding guide for Web

The MediaPipe Text Embedder task lets you create a numeric representation of text data to capture its semantic meaning. These instructions show you how to use the Text Embedder for web and JavaScript apps.

For more information about the capabilities, models, and configuration options of this task, see the Overview.

Code example

The example code for Text Embedder provides a complete implementation of this task in JavaScript for your reference. This code helps you test this task and get started on building your own text embedding app. You can view, run, and edit the Text Embedder example code using just your web browser.

Setup

This section describes key steps for setting up your development environment and code projects specifically to use Text Embedder. For general information on setting up your development environment for using MediaPipe tasks, including platform version requirements, see the Setup guide for Web.

JavaScript packages

Text Embedder code is available through the @mediapipe/tasks-text package. You can find and download these libraries from links provided in the platform Setup guide.

You can install the required packages with the following code for local staging using the following command:

npm install @mediapipe/tasks-text

If you want to deploy to a server, you can use a content delivery network (CDN) service, such as jsDelivr, to add code directly to your HTML page, as follows:

<head>
  <script src="https://cdn.jsdelivr.net/npm/@mediapipe/tasks-text@latest/index.js"
    crossorigin="anonymous"></script>
</head>

Model

The MediaPipe Text Embedder task requires a trained model that is compatible with this task. For more information on available trained models for Text Embedder, see the task overview Models section.

Select and download a model, and then store it within your project directory:

<dev-project-root>/app/shared/models

Create the task

Use one of the Text Embedder createFrom...() functions to prepare the task for running inferences. You can use the createFromModelPath() function with a relative or absolute path to the trained model file. The code example below demonstrates using the createFromOptions() function. For more information on the available configuration options, see Configuration options.

The following code demonstrates how to build and configure this task:

async function createEmbedder() {
  const textFiles = await FilesetResolver.forTextTasks("https://cdn.jsdelivr.net/npm/@mediapipe/tasks-text@latest/wasm/");
  textEmbedder = await TextEmbedder.createFromOptions(
    textFiles,
    {
      baseOptions: {
        modelAssetPath: `https://storage.googleapis.com/mediapipe-tasks/text_embedder/universal_sentence_encoder.tflite`
      },
      quantize: true
    }
  );
}
createEmbedder();

Configuration options

This task has the following configuration options for Web and JavaScript applications:

Option Name Description Value Range Default Value
l2Normalize Whether to normalize the returned feature vector with L2 norm. Use this option only if the model does not already contain a native L2_NORMALIZATION TFLite Op. In most cases, this is already the case and L2 normalization is thus achieved through TFLite inference with no need for this option. Boolean False
quantize Whether the returned embedding should be quantized to bytes via scalar quantization. Embeddings are implicitly assumed to be unit-norm and therefore any dimension is guaranteed to have a value in [-1.0, 1.0]. Use the l2Normalize option if this is not the case. Boolean False

Prepare data

Text Embedder works with text (string) data. The task handles the data input preprocessing, including tokenization and tensor preprocessing. All preprocessing is handled within the embed function. There is no need for additional preprocessing of the input text beforehand.

const inputText = "The input text to be embedded.";

Run the task

The Text Embedder uses the embed function to trigger inferences. For text embedding, this means returning the embedding vectors for the input text.

The following code demonstrates how execute the processing with the task model.

// Wait to run the function until inner text is set
const embeddingResult = textEmbedder.embed(
  inputText
);

Handle and display results

The Text Embedder outputs a TextEmbedderResult that contains a list of embeddings (either floating-point or scalar-quantized) for the input text.

The following shows an example of the output data from this task:

TextEmbedderResult:
  Embedding #0 (sole embedding head):
    float_embedding: {0.2345f, 0.1234f, ..., 0.6789f}
    head_index: 0

You can compare the semantic similarity of two embeddings using the TextEmbedder.cosineSimilarity function. See the following code for an example.

// Compute cosine similarity.
const similarity = TextEmbedder.cosineSimilarity(
  embeddingResult.embeddings[0],
  otherEmbeddingResult.embeddings[0]);

The Text Embedder example code demonstrates how to display the embedder results returned from the task, see the code example for details.