Next Word Predictor

Final Project Submission - Data Science Capstone

Hereby I present a prototype tool to facilitate the typing of text to the users of computers or mobile phones.

• This is done by providing a prediction of the next word.
• The algorithm is based on the “Stupid Backoff” model. http://www.aclweb.org/anthology/D07-1090.pdf.

In general terms:

• In any language, each collection of n words (n-grams) has an occurrence rate, which can be identified with its probability of appearing in a text. There are collections of n words that appear more frequently than others, specially due to the grammar structure.

• The same applies for all words that are used after specific combinations of words.

• The “Stupid Backoff” back-off model provides estimations based on the sampling of a large number of texts, and it makes estimations based on a parsimonious process, through which it is looking for matches for n-grams of inferior length, until finding a coincidence.

You can access the application in the following link:

https://francisco-garcia.shinyapps.io/predictor/

This application runs on Shinyapps.

Shinyapps is a platform for interactive web apps built with the Shiny R package.

Please, enter a text with three or four words and push the button “Submit”