Next Word Prediction

An NLP-powered text prediction application

Data Science Capstone

Built using:

• R
• Quanteda
• Shiny
• SwiftKey English corpus

Objective: Predict the most likely next word from a phrase entered by the user.

Why Next Word Prediction?

Typing assistance can improve the speed and convenience of text input.

Our objective

• Accept a phrase from the user
• Analyze the preceding words
• Predict the most likely next word
• Return the prediction through an interactive Shiny application

Training Data

The model uses the English SwiftKey corpus:

• Blogs
• News
• Twitter

A 1% sample was used to balance model coverage and application performance.

N-gram Language Model

The prediction engine uses a backoff strategy:

4-gram
   ↓
3-gram
   ↓
2-gram
   ↓
Unigram