@jrc03c/js-nlp-tools

Intro

This is a little set of JS natural language processing tools.

Installation

npm install --save @jrc03c/js-nlp-tools

Usage

import { Corpus, Document } from "@jrc03c/js-nlp-tools"
import fs from "node:fs"

const doc1 = new Document({
  name: "Frankenstein",
  raw: fs.readFileSync("path/to/frankenstein.txt", "utf8"),
})

const doc2 = new Document({
  name: "Pride & Prejudice",
  raw: fs.readFileSync("path/to/pride-and-prejudice.txt", "utf8"),
})

const doc3 = new Document({
  name: "Moby Dick",
  raw: fs.readFileSync("path/to/moby-dick.txt", "utf8"),
})

const corpus = new Corpus({ docs: [doc1, doc2, doc3] })

corpus.process().then(() => {
  console.log(corpus.computeTFIDFScore("Frankenstein", doc1))
})

API

Corpus

Methods

Corpus(data) (constructor)

Returns a new Corpus instance. Can optionally take a data argument, which is an object with properties corresponding to Corpus instance properties (e.g., docs).

computeIDFScore(word)

Returns the inverse document frequency score for a given word. Is computed as:

\text{IDF} = \text{log}(N / n_t)

Where:

• $N$ = the total number of documents in the corpus
• $n_t$ = the number of documents in which the word appears

computeTFScore(word, doc)

Returns the term frequency score for a given word and document. Is computed as:

\text{TF} = 0.5 + 0.5 \frac{f_{t, d}}{\text{max}_{\{t'∈d\}} f_{t',d}}

Where:

• $f_{t, d}$ = the number of times the word appears in the document
• $\text{max}_{{t'∈d}} f_{t',d}$ = the number of times the most frequently-occurring word appears in the document

computeTFIDFScore(word, doc)

Returns the tf-idf score for a given word and document. Is computed as the term frequency score multiplied by the inverse document frequency score.

process(progress)

Returns a Promise that resolves once all documents in the corpus have been processed. Can optionally take a callback function that is passed the progress through the documents as a value between 0 and 1.

Properties

docs

An array of Document instances.

hasBeenProcessed

A boolean indicating whether or not the instance's process method has been invoked (and completed).

Document

Methods

Document(data) (constructor)

Returns a new Document instance. Can optionally take a data object with properties corresponding to Document instance properties (e.g., wordCounts).

getWordCount(word)

Returns the number of times word (a string) appears in the document.

process()

Returns a Promise that resolves once the document has been processed (indexed).

Properties

hasBeenProcessed

A boolean representing whether or not the instance's process method has been invoked (and completed).

isCaseSensitive

A boolean representing whether or not case should matter when indexing words.

mostFrequentWord

A string representing the word that appears most frequently in the document.

name

A string representing the name of the document. If no name is assigned via the data object passed into the constructor, then a random string will be assigned as the document's name.

raw

A string representing the raw text on which the document is based.

totalWordCount

A non-negative integer representing the total number of words in the document.

wordCounts

A dictionary that maps words (as strings) to the numbers of times those words appear in the document (as non-negative integers).

Utility functions

clean(raw, shouldPreserveCase)

Given raw (a string) and optionally shouldPreserveCase (a boolean), returns a copy of raw in which all punctuation has been removed and all whitespace characters have been replaced with spaces. By default, shouldPreserveCase is false.

defineReadOnlyProperty(object, name, value)

Defines a read-only property called name on object with the value value. Returns object.

Note that any read-only properties defined this way will fail silently when new values are assigned to them. In other words, you won't be notified when any assignment attempts fail.