Wouldn’t Be Great If…

You could build a word prediction application customized to your needs?

NextWord is a new word prediction application that allows you to predict your next word based on state of the art prediction algorithms and a flexible system to tune its performance!

Using SwiftKey Data & Natural Language Processing

Behind and intuitive user interface, NextWord uses twitter, blog and news article data provided by SwiftKey. The configurable build system samples, processes, analyzes, creates, and evaluates the prediction model.

Using advanced natural language processing techniques such as Modified Kneser-Nye smoothing, NextWord delivers accuracy and blazing fast performance.

Performance Characteristics

More Than Just An Application

There is an entire automated workflow of natural language processing to tune a prediction model to fit your needs:

• Sampling & pre-processing & smoothing The Shiny applications uses quad-, tri-, and bigrams from a 20% sampling of a corpus of articles, blog posts and tweets. Preprocessing removes profanity, punctuation, numbers and foreign characters.
  
• Configurable tuning parameters Such as low frequency ngram thresholds provide flexibility to turn model performance and trade-off model size vs. accuracy.
  
• Testing and performance reporting Provides First-word and In-top-3-word acccuracy calculations as well as mean query time to compare and contrast model configurations.
  
• Self documenting model archive Each time a model is built and saved, a csv file is updated with the configuration parameters and performance statistics for future reference.

Make Predicting Your Next Word Fast and Easy!

Shiny Application: https://rougeone.shinyapps.io/nextword/

Code Repo: https://github.com/wayneheller/DataScienceSpecializationCapstone

Pitch Deck (github): https://wayneheller.github.io/DataScienceSpecializationCapstone/

Pitch Deck (rpubs): http://rpubs.com/wgheller/nextword