Library for calculate vector space model using cosine similarity.. for now i test this library using indonesian languange. so i didn't testing it with english data. but i was added english lemmatize document. but maybe i will change it to porter algorithm

How to install

1npm install vector-space-model-similarity --save

How to use

first we import our function

1import { VSM } from 'vector-space-model-similarity

next we define documents, it's an array

1const documents = ["rumah saya penuh makanan", "saya suka makan nasi", "nasi berawal dari beras"] // define our variable

next we call VSM Class and define our object from VSM class

1const document = new VSM(documents); // define our object of VSM

1const idf = document.getIdfVectorized(); // return an array ob object, the key is tokenize our documents and the value is the 
2
3const query = new VSM(["sistem cerdas"], idf); // we define our object again, it's for query. and we pass our idf constant variable
4
5const cosine = Cosine(query.getPowWeightVectorized()[0], document.getPowWeightVectorized()); // calculating cosine similarity

descriptions

example of vsm calculating using excel.

VSM

1import { VSM } from 'vector-space-model-similarity

VSM is a class that extends from Tfidf class. VSM has one constructor and in the constructor it has two parameter. the first parameter is an important parameter and the second is optional. it's the parameter

1documents: string[], idfVector:any[] = []

documents represented about our document, and idfVector is the idf from our vector of IDF number. idfVector is important if you want to search data from query. you must pass idfVector from the idf you got from documents before. to get idfVector use this function.

getIdfVectorized will return this array. but not array of number, it's array of object. the key is the word and the value is the IDF value

1getIdfVectorized() // <-- this is method from TFIDF Class.

getWeightVectorized() will return idf value. and the return is an multidimension array

1getWeightVectorized() // <-- return weight of documents

getPowWeightVectorized() will return Exponent of IDF from the documents

1getPowWeightVectorized() // <-- return weight of documents

Methods	Returned
getIdfVectorized	any[]
getWeightVectorized	any[][]
getPowWeightVectorized	any[][]

Cosine

when you was got documents and query vector idf you can use this function

1import { Cosine } from 'vector-space-model-similarity

Cosine library has two parameters. the first paramter is a query, and the second is a documents

1queries:any[], documents:any[][] // <=== the parameters
2
3number[] // <=== the return

after you get exponent of document idf from getPowWeightVectorized() you can use this function

query is single dimension of array, and documents is a multi dimension of array. becasue getPowWeightVectorized() return multidimension array and the query parameter required singledimension of array you must pass the first index of your array. e.g :

1
2const document = new VSM([
3    "sistem cerdas adalah kumpulan elemen",
4    "adalah kumpulan elemen yang saling berinteraksi",
5    "Sistem berinteraksi untuk mencapai tujuan"
6]);
7
8const idf = document.getIdfVectorized();
9
10const query = new VSM(["sistem cerdas"], idf);
11
12const cosine = Cosine(query.getPowWeightVectorized()[0], document.getPowWeightVectorized()); // output : [ 4.457087767265072, 0, 0.4853443577859814 ]
13

props

all function you can import from this package

Methods	Returned	Descriptions
Tokenize	string[]	To get token from the string. eg : "The sun and the moon" => ["the", "sun", "and", "the", "moon"]
Stemming	string	currently I'm using lemmatization algorithm. and maybe I will change to porter or other stemming or lemma algorithm. it's a function to get base word. and support english or indonesian. eg : memakan => makan, berlari => lari
stopword	string[]	to remove not important word. eg : ["saya", "suka", "dia"] => ["suka"]
stopword	string	to remove not important word. eg : ["saya", "suka", "dia"] => ["suka"]