polarity

polarity

Detect the polarity of text, based on AFINN-111.

Uses wooorm/afinn-111.

Installation

NPM:

$ npm install polarity

Component.js:

$ component install wooorm/polarity

Usage

var polarity = require('polarity');

polarity(['some', 'positive', 'happy', 'cats']); /* {
 *   polarity: 5,
 *   positive: [ 'positive', 'happy' ],
 *   negative: []
 * }
 */

polarity(['darn', 'self-deluded', 'abandoned', 'dogs']); /* {
 *   polarity: -4,
 *   positive: [],
 *   negative: [ 'self-deluded', 'abandoned' ]
 * }
 */

polarity(['some', 'positive', 'happy', 'cats'], {
  'cats' : Infinity
}); /* {
 *   polarity: Infinity,
 *   positive: [ 'positive', 'happy', 'cats' ],
 *   negative: []
 * }
 */

polarity.inject({
    'dogs' : -Infinity /* What?! */
});

polarity(['darn', 'self-deluded', 'abandoned', 'dogs']); /* {
 *   polarity: -Infinity,
 *   positive: [],
 *   negative: [ 'self-deluded', 'abandoned', 'dogs' ]
 * }
 */

API

polarity(values[string]?, inject?)

Get a polarity result from given values, optionally with one time injections.

polarity does not tokenise values. There are better tokenisers around (hint hint, parse-latin). However, the following will work pretty good:

function tokenize(value) {
    return value.toLowerCase().replace(/[^-a-z0-9 ]/g, '').split(' ');
}

polarity.inject(inject{word: weight})

Replace AFINN weights with your own.

polarity.afinn

Direct access to afinn-111.

Supported words

polarity supports all AFINN-111 words/phrases. afinn-111 has more information on what words and phrases are supported.

Other polarity detection libraries

• thisandagain/sentiment — Slower (but pretty good nonetheless), has tokeniser, async support;
• mileszim/sediment — Slower, has tokeniser.
• thinkroth/Sentimental — Slower, has tokeniser.

Benchmark

Run the benchmark yourself:

$ npm run install-benchmark # Just once of course.
$ npm run benchmark

On a MacBook Air, it detects the polarity of 168,820 tweets per second. This is with a manual tokenisation step (like the others do) taken into account. Without the tokenisation step, polarity is of course much faster.

         benchmarks * 20 tweets (10 pos, 10 neg)
  8,441 op/s » polarity -- this module
  7,708 op/s » sentiment
  3,950 op/s » sediment
  2,604 op/s » Sentimental

License

MIT