The main objective was to develop and deploy a Shiny app that is able to do seamless word prediction, real-time.

The data for this project came from Swiftkey and for deployed app parts of news and twitter datasets were used.

Final deployment consists of 2 versions: one as per course specifications and other with more user-friendly response of three predicted words and a way to build naturally a full sentences.

For each set of words last three words are used:
- Prior to prediction, preparation of 2-gram, 3-gram and 4-gram datasets is done
- Cleaning of input data is done to prepare input the same way training dataset is cleaned (profanity, stopwords, various artifacts…)
- First, 4-gram data is looked up for match of all three words
- If no match is found, 3-gram and eventually 2-gram is looked up

App is deployed as R Shiny app on:
https://duxan.shinyapps.io/DS_capstone

This presentation's original location: http://rpubs.com/duxan/DS_capstone