Neural Network Word Prediction

Start by typing your sentence or phrase into the text box.

While the box is selected, every time you press spacebar or enter; the model will return the next most likely word or character.

If the word you are looking for doesn't appear start typing the word.

Note this model only recognizes about 20,000 individual tokens.

The words will appear both in a menu below the box and in the Word Cloud.

If you do not want to filter your text then uncheck the Filter Words box.

This model is capable of guessing three writing styles: BLOG, TWITTER or NEWS

Writing style guess

Filter Words

Word Cloud


This Project

This project was designed by Swiftkey as part of the Coursera Data Science Capstone, which was designed by Johns Hopkins University.

The goal of the project was to create a web application that highlights a word prediction algorithm and to build a short slide presentation to pitch the app.

The Model

This word prediction model runs on a recurrent neural network (RNN) built in the R-programming language. Rather then explicitly programming a set of language rules, a RNN was implemented in order to learn and generalize patterns that naturally occur in English.

The type of RNN used for this project is a long short-term memory (LSTM).

When text is entered into the typing field it is sent to a server containing R-code, which runs the text through the LSTM model. A table of tokens is reordered according to the probability of each token being the next word or character.

For a more in depth look, follow the links in the Resources tab.


I have been programming for a little over a year. I originally started in R, but I have had to expand my lexicon to include HTML, JavaScript, and Bash in order to complete this project. Admittedly, this is one of my first attempts with front end web development, so I consider this application more proof of concept rather than fully developed.