Predict the nextWord!

Ryan Morton
06/20/2017

Introduction

  • Describe the Algorithm
  • Demostrate the App
  • Report Performance

The Algorithm

  • N-gram model using up to 5 word keys to predict the next word
  • Model holds 1,360,650 rows in its dictionary
  • Model utilizes data.table package to increase speed of searching the dictionary
  • Stupid backoff attempts highest n-gram first and backoffs to the next lowest down to 2-grams

Application Demo

  • Simple interface makes for ease of use

Performance of nextWord App

  • 20 seconds to load 1M+ dictionary
  • Returns results as you type
  • Test sample of 1,067 Twitter 6-grams yielded 33% correct predictions with the algorithm (your experience may differ)
  • Model is sensitive to training dictionary context
  • Give it a try!