Capstone Project Presentation

This application is designed to enhance user typing experience on mobile devices.

• Typing on small screens can be challenging as most of the times is with one hand
• An efficient prediction application will improve the overall user experience

• The model should operate with limited resources and should not impact device performance

• Shiny Application Link - https://enframy.shinyapps.io/NextWordPrediction/

The Stupid Backoff algorithm used by Google was considered for this application.

• It creates a smoothing approximation for very large n-grams.
• Gives up the idea of trying to make the language model a true probability distribution
• Instead, if a higher-order N-gram has a zero count, the model backoff to a lower order N-gram
• Weights are assigned by a fixed (context-independent) amount

• Available on Shiny on https://enframy.shinyapps.io/NextWordPrediction/
• Go to “Next Word Prediction” tab and start typing on the text box presented
• The prediction algorith will display 3 recommendations based on sample collected
• Click the “About This Application” tab for references and more details about the model

• Improve efficiency by using a larger corpora with frequent resampling
• Reach more users by including additional languages
• Validate additional models to improve accuracy

As part of the Coursera Capstone project in cooperation with SwiftKey, the requirements requested include:

• Analyze a Corpora of text from different sources (news, blog, and Twitter) and create a sample
• Tokenize the sample and create n-grams, find frequencies of common n-grams needed for the predictive model
  
• Build a model that takes any input from the user and allow to predict the next word
• Create and deploy a Shiny application to allow users get a prediction of the next word as they type