Pure-Js TextRank Algorithm

Implementing the TextRank algorithm in pure JavaScript for text summarization.

Overview

TextRank, an unsupervised graph-based ranking algorithm, is adept at condensing text. This implementation draws inspiration from Google's PageRank, applying it to a graph composed of sentences as nodes, each weighted by saliency. Edges represent sentence similarities, forming the basis for the ranking process.

Customizable Parameters

This implementation offers flexibility through four key parameters:

• encoding: Choose between two sentence considerations: "as_it_is" or "bag_of_words".
• distance: Four distance metrics available: "jaccard", "custom", "dice_coefficient", and "overlap".
• alfa: A pivotal factor in the PageRank algorithm, influencing ranking computation. (The so-called dumping factor)
• pr_iterations: Control the number of PageRank iterations for precision.

Additionally, you can determine the number of top-sentences to extract by providing a function to the summarize() method. For instance:

tr.summarize(sentences => Math.floor(sentences.length * 0.1));

Example

Available also in ./index.js

const TextRank = require('./TextRank');

const encodings = [
    'bag_of_word',
    'as_they_are'
]

const distances = [
    'jaccard', 
    'custom', 
    'overlap',
    'dice_coefficient'
]

const tr = new TextRank({
    encoding: encodings[0],
    distance: distances[0],
    alfa: 0.85,
    pr_iteration : 100
});

tr.load('./test.txt');

const summary = tr.summarize();

console.log(summary);