Our goal was to create a simple and elegant app that could predict the next word in a sentence out of a meaningful input of words.

To provide this functionality we built a N-gram probabilistic language model and used Stupid Backoff algorithm to rank next-word candidates.

Our application, Your Next Word, achieves the following accuracy using an independent test set and generates predictions in less than 0.348s by using 69.5MB of disk space.

We built our app using N-gram probabilistic language model. Given our chosen corpus of (1) 899,288 blog posts, (2) 77,259 news articles, and (3) 2,360,148 tweets, we assigned probabilities to sequences of N words. We then estimated the probability of the last word of an N-gram given the previous (N-1) words.

To rank next-word candidates, we applied a mechanism called Stupid Backoff. The scoring function is defined by the following formula:

N-gram to use

• We chose quadgram for our app because there is only minimal improvement in prediction accuracy by using fivegram instead of quadgram.

Try YourNextWord at https://winniemaktse.shinyapps.io/YourNextWord!