Swiftkey Keyboard Predictive Text Model

Regis O'Connor
March 18, 2017

Executive Summary

Swiftkey's mission is to enhance the interaction between people and technology. Swiftkey's hallmark predictive technology learns from a personal writing style, getting smarter over time.

The objective of this project is to develop a predictive text model and demo app for use on cell phones that is fast, accurate and energy efficient.

A backoff model created with a sample of blog text successfully predicted results between 1.3% and 2.6% of the cases studied. Although these figures seem small they represent odds that are about 2000 times better than chance.

Next steps in the mobil and app development should explore:

  • tweaking the model to employ smoothing techniques that improve accuracy, speed and energy efficiency,
  • user testing to find ways to enhance the experience.

Methodology

Model Overview

alt text

Click here to review the code: http://rpubs.com/regiso/258900

The Swiftkey Shiny App

alt txt

The Shiny App simulates the user experience, especially the responsiveness and speed of the model.

Click here to play with the Shiny App: https://regisoconnor.shinyapps.io/Final_Shiny_App/