Natural Language Processing Toolkit for node.js

This module covers some basic nlp principles and implementations. The main focus is performance. When we deal with sample or training data in nlp, we quickly run out of memory. Therefore every implementation in this module is written as stream to only hold that data in memory that is currently processed at any step.

Install

npm install nlp-toolkit

Example

Frequency distribution of words in texts. Tokenize, remove stopwords, stem words, count words. Traditionally those steps happen sequentially. But we do not need to tokenize the whole text before removing stopwords.

var nlp = require('nlp-toolkit');
var fs = require('fs');
var es = require('event-stream');
 
fs.createReadStream('./pride_prejudice.txt')
.pipe(es.split())
.pipe(nlp.tokenizer())
.pipe(nlp.stopwords())
.pipe(nlp.stemmer())
.pipe(nlp.frequency())
.on('data', function (freqDist) {
  console.log(freqDist.slice(0, 10));
})
.on('error', function (err) {
  console.error(err);
});

Modules

Tokenizer

.pipe(nlp.tokenizer(options))

options:

Tokenizer also work in a non-stream context:

var tokens = nlp.tokenizer(string, options);

Stopwords

.pipe(nlp.stopwords(options))

options:

additionalWords:

Supported languages: da, de, en, es, fi, fr, hu, it, nl, no, pt, ro, ru, se, tr.

Stopwords also work in a non-stream context:

nlp.stopwords(sentence, options)
.then(function (tokens) {}})
.catch(function (err) { console.error(err); });

Stemmer

.pipe(nlp.stemmer(options))

options:

Supported languages: da, de, en, es, fi, fr, hu, it, nl, no, pt, ro, ru, se, tr.

Stopwords also work in a non-stream context:

var tokens = nlp.stemmer(sentence, options);

This module uses the stemmer implementation of Snowball-Stemmer.

Frequency Distribution

.pipe(nlp.frequency())