Created
January 26, 2019 10:58
-
-
Save tomericco/14b5ceac90d6eed6f9ba6cb5305f8fab to your computer and use it in GitHub Desktop.
Cosine similarity implementation in JS
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
const str1 = 'This is an example to test cosine similarity between two strings'; | |
const str2 = 'This example is testing cosine similatiry for given two strings'; | |
// | |
// Preprocess strings and combine words to a unique collection | |
// | |
const str1Words = str1.trim().split(' ').map(omitPunctuations).map(toLowercase); | |
const str2Words = str2.trim().split(' ').map(omitPunctuations).map(toLowercase); | |
const allWordsUnique = Array.from(new Set(str1Words.concat(str2Words))); | |
// | |
// Calculate IF-IDF algorithm vectors | |
// | |
const str1Vector = calcTfIdfVectorForDoc(str1Words, [str2Words], allWordsUnique); | |
const str2Vector = calcTfIdfVectorForDoc(str2Words, [str1Words], allWordsUnique); | |
// | |
// Main | |
// | |
console.log('Cosine similarity', cosineSimilarity(str1Vector, str2Vector)); | |
// | |
// Main function | |
// | |
function cosineSimilarity(vec1, vec2) { | |
const dotProduct = vec1.map((val, i) => val * vec2[i]).reduce((accum, curr) => accum + curr, 0); | |
const vec1Size = calcVectorSize(vec1); | |
const vec2Size = calcVectorSize(vec2); | |
return dotProduct / (vec1Size * vec2Size); | |
}; | |
// | |
// tf-idf algorithm implementation (https://en.wikipedia.org/wiki/Tf%E2%80%93idf) | |
// | |
function calcTfIdfVectorForDoc(doc, otherDocs, allWordsSet) { | |
return Array.from(allWordsSet).map(word => { | |
return tf(word, doc) * idf(word, doc, otherDocs); | |
}); | |
}; | |
function tf(word, doc) { | |
const wordOccurences = doc.filter(w => w === word).length; | |
return wordOccurences / doc.length; | |
}; | |
function idf(word, doc, otherDocs) { | |
const docsContainingWord = [doc].concat(otherDocs).filter(doc => { | |
return !!doc.find(w => w === word); | |
}); | |
return (1 + otherDocs.length) / docsContainingWord.length; | |
}; | |
// | |
// Helper functions | |
// | |
function omitPunctuations(word) { | |
return word.replace(/[\!\.\,\?\-\?]/gi, ''); | |
}; | |
function toLowercase(word) { | |
return word.toLowerCase(); | |
}; | |
function calcVectorSize(vec) { | |
return Math.sqrt(vec.reduce((accum, curr) => accum + Math.pow(curr, 2), 0)); | |
}; |
Sure, Gabriel. I'd be honored :)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Thanks for sharing that! I used your implementation of cosine similarity to measure image similarity in my recent project for the CS50 course. It worked very well. Here's a link to the repo https://github.com/gabriel-aleixo/cs50-final-project. Thanks!