Billy Jackson's Next Word Prediction App

Created by Billy Jackson
June 11, 2017

Product Summary

Goal: The purpose of this app is to take an English phrase as an input (multiple words) and predict the next word.

Features:

  • This app returns the three top choices for the next word.
  • This app comes with the option to filter profanity if the user prefers.
  • Uses a Backoff technique to find the best prediction if an exact given phrase is not found.

How to Use it

Easy!

Simply type any phrase into the textbox and the app returns the top 3 predictions for the next word in the phrase.

Background

To get a varied sense of the English language, this model generates its predictions from aggregating and analyzing three datasets containing lines of text from approximately:

  • 1 million blog articles
  • 900,000 news articles
  • 2.4 million tweets

The datasets were created on July 22, 2014 and provided by the Johns Hopkins University Data Science Specialization offered through Coursera.

The raw datasets can be downloaded here

Quantifying Performance

The performance of this app was put to the challenge against a common benchmark script used by many other students taking the capstone.

The model takes on average about 0.2 seconds to find its prediction. The accuracy of this app against the sample data turned out very strong with result rates of:

alt text