wink-embeddings-sg-100d

100-dimensional English word embeddings for wink-nlp

This pre-trained 100-dimensional English word embedding set is specifically optimized for winkNLP. This package (~110MB download, ~310MB installed) includes embeddings for over 350K English words. Boost accuracy in semantic similarity, text classification, and more – even in the browser.

Getting Started

Prerequisite

It requires Node.js version 16.0.0 or above and winkNLP version 2.1.0 or above.

Installation

The model must be installed along with the wink-nlp and the wink-eng-lite-web-model:

# Install wink-nlp
npm install wink-nlp --save
# Install wink-eng-lite-web-model
npm install wink-eng-lite-web-model --save
# Install wink-embeddings-sg-100d
npm install wink-embeddings-sg-100d --save

Example

The code below computes the similarities between all pairs of four sentences by obtaining their vectors and using cosine similarity:

// Load wink-nlp package.
const winkNLP = require( 'wink-nlp' );
// Load english language model.
const model = require( 'wink-eng-lite-web-model' );
// Load word embeddings.
const vectors = require( 'wink-embeddings-sg-100d' );
// Load similarity utility.
const similarity = require( 'wink-nlp/utilities/similarity.js' );

// Use only tokenization and sentence boundary detection pipe.
const nlp = winkNLP( model, [ 'sbd' ], vectors );
// Obtain "its" helper to extract item properties.
const its = nlp.its;
// Obtain "as" reducer helper to reduce a collection.
const as = nlp.as;
// The following text contains 4-sentences, where the first 
// two and the last two have high similarity.
const text = `The cat rested on the carpet. The kitten slept on the rug.
The table was in the drawing room. The desk was in the study room.`;
// This will hold the array of vectors for each sentence.
const v = [];
// This will hold sentence pairs and their similarities.
const similarities = [];
// Run the nlp pipe.
const doc = nlp.readDoc( text );
// Compute each sentence's embedding and fill in "v[i]".
// Only use words and ignore stop words.
doc
  .sentences()
  .each( ( s, k ) => {
    v[ k ] = s
      .tokens()
      .filter( (t) => (t.out(its.type) === 'word' && !t.out(its.stopWordFlag)))
      .out(its.value, as.vector);  
  })
// Compute & save similarity for all the pairs.
for ( let i = 0; i < v.length; i += 1 ) {
  for ( let j = 0; j < v.length; j += 1 ) {
      similarities.push( 
        {sentence1: `${i+1}. ${doc.sentences().itemAt( i ).out()}`, sentence2: `${j+1}. ${doc.sentences().itemAt( j ).out()}`, similarity: `${similarity.vector.cosine( v[ i ], v[ j ] ).toFixed( 2 )}`}
      );
  }
}

The output i.e. the similarities array, of the above example is visually illustrated below:

Explore more in the "How to use word embedding with winkNLP?" Observable notebook.

Need Help?

If you spot a bug and the same has not yet been reported, raise a new issue.

About winkJS

WinkJS is a family of open source packages for Natural Language Processing, Machine Learning, and Statistical Analysis in NodeJS. The code is thoroughly documented for easy human comprehension and has a test coverage of ~100% for reliability to build production grade solutions.

Copyright & License

Wink NLP is copyright 2017-24 GRAYPE Systems Private Limited.

It is licensed under the terms of the MIT License.