cmpstr/src/algorithms/cosine.js

lightweight npm package to calculate string similarity

/**
 * Cosine Similarity
 * CmpStr module
 * 
 * Cosine similarity is a measure how similar two vectors are. It's often used
 * in text analysis to compare texts based on the words they contain.
 * 
 * @author Paul Köhler (komed3)
 * @license MIT
 */


'use strict';

/**
 * private helper function
 * get term frequency from string
 * @private
 * 
 * @param {String} str string
 * @param {String} delimiter term delimiter
 * @returns {Object} term frequency
 */
const _termFreq = ( str, delimiter ) => {

    let freq = {};

    str.split( delimiter ).forEach( ( term ) => {

        freq[ term ] = ( freq[ term ] || 0 ) + 1;

    } );

    return freq;

};

/**
 * module exports
 * @public
 * 
 * @param {String} a string a
 * @param {String} b string b
 * @param {Object} options having {
 *   @param {String} [delimiter=' '] term delimiter
 * }
 * @returns {Number} similarity score (0..1)
 */

module.exports = ( a, b, { delimiter = ' ' } = {} ) => {

    /* step 1: count the frequency of chars per string */

    let termsA = _termFreq( a, delimiter ),
        termsB = _termFreq( b, delimiter );

    let allTerms = new Set ( [
        ...Object.keys( termsA ),
        ...Object.keys( termsB )
    ] );

    /* step 2: calculate the dot product */

    let dotProduct = [ ...allTerms ].reduce(
        ( sum, char ) => sum + ( termsA[ char ] || 0 ) * ( termsB[ char ] || 0 ),
        0
    );

    /* step 3: calculate the vector magnitudes */

    let magnitudeA = Math.sqrt( [ ...allTerms ].reduce(
        ( sum, char ) => sum + ( termsA[ char ] || 0 ) ** 2,
        0
    ) );

    let magnitudeB = Math.sqrt( [ ...allTerms ].reduce(
        ( sum, char ) => sum + ( termsB[ char ] || 0 ) ** 2,
        0
    ) );

    /* step 4: calculate Cosine similarity */

    return magnitudeA && magnitudeB
        ? dotProduct / ( magnitudeA * magnitudeB )
        : 0;

};