📄 ClassifyText (JS)

Use machine learning to classify text using n-grams and cosine similarity.

Minimal library that can be used both in the browser and in Node.js, that allows you to train a model with a large amount of text samples (and corresponding labels), and then use this model to quickly predict one or more appropriate labels for new text samples.

Installation

Using npm

npm install ml-classify-text

Using yarn

yarn add ml-classify-text

Getting started

Import as an ES6 module

import Classifier from 'ml-classify-text'

Import as a CommonJS module

const { Classifier } = require('ml-classify-text')

Basic usage

Setting up a new Classifier instance

const classifier = new Classifier()

Training a model

let positive = [
    'This is great, so cool!',
    'Wow, I love it!',
    'It really is amazing',
]

let negative = [
    'This is really bad',
    'I hate it with a passion',
    'Just terrible!',
]

classifier.train(positive, 'positive')
classifier.train(negative, 'negative')

Getting a prediction

let predictions = classifier.predict('It sure is pretty great!')

if (predictions.length) {
	predictions.forEach(prediction => {
		console.log(`${prediction.label} (${prediction.confidence})`)
	})
} else {
	console.log('No predictions returned')
}

Returning:

positive (0.5423261445466404)

Advanced usage

Configuration

The following configuration options can be passed both directly to a new Model, or indirectly by passing it to the Classifier constructor.

Options

Using n-grams

The default behavior is to split up texts by single words (known as a bag of words, or unigrams).

This has a few limitations, since by ignoring the order of words, it's impossible to correctly match phrases and expressions.

In comes n-grams, which, when set to use more than one word per term, act like a sliding window that moves across the text — a continuous sequence of words of the specified amount, which can greatly improve the accuracy of predictions.

Example of using n-grams with a size of 2 (bigrams)

const classifier = new Classifier({
	nGramMin: 2,
	nGramMax: 2
})

let tokens = classifier.tokenize('I really dont like it')

console.log(tokens)

Returning:

{
    'i really': 1,
    'really dont': 1,
    'dont like': 1,
    'like it': 1
}

Serializing a model

After training a model with large sets of data, you'll want to store all this data, to allow you to simply set up a new model using this training data at another time, and quickly make predictions.

To do this, simply use the serialize method on your Model, and either save the data structure to a file, send it to a server, or store it in any other way you want.

let model = classifier.model

console.log(model.serialize())

Returning:

{
    nGramMin: 1,
    nGramMax: 1,
    vocabulary: [
    	'this',    'is',      'great',
    	'so',      'cool',    'wow',
    	'i',       'love',    'it',
    	'really',  'amazing', 'bad',
    	'hate',    'with',    'a',
    	'passion', 'just',    'terrible'
    ],
    data: {
        positive: {
            '0': 1, '1': 2, '2': 1,
            '3': 1, '4': 1, '5': 1,
            '6': 1, '7': 1, '8': 2,
            '9': 1, '10': 1
        },
        negative: {
            '0': 1, '1': 1, '6': 1,
            '8': 1, '9': 1, '11': 1,
            '12': 1, '13': 1, '14': 1,
            '15': 1, '16': 1, '17': 1
        }
    }
}

Documentation

• Classifier
• Model
• Vocabulary
• Prediction

Contributing

Read the contribution guidelines.

Changelog

Refer to the changelog for a full history of the project.

License

ClassifyText is licensed under the MIT license.