• In this project a word prediction prototype app has been built that would predict the next word that they "most likely" want to type.
• The app might be used in very limited resource devices and so an optimised training set of 6-grams data is used from the twitter, news and blog corpuses.
• Goal was to have smallest application with faster performance to predict next word with reasonable accuracy.

• The data was read into R using data.table::fread for faster reading of huge files.
• Using text2vec package in R, the text was cleaned, tokenized, split into one to six grams and stored with the count of their occurances. This text2vec package is used coz it is flexible, fast and memory efficient.
• Smoothing was done using "STUPID BACKOFF" algorithm, which worked out good for large corpuses and calculated score using the frequency of the words.
• The query when entered is searched for matching word in the data bases and atmost three predicted words are listed in descending order of their frequency or their score.

• The app reads in the data tables for each n-gram along with its score. Setkeyv index is used for faster access
• Takes input in the provided text box and reactively predicts the next words and lists below. The whole sentence is displayed, once the next word is typed by the user.

The Link to the shiny app is :

Next Word Predictor Prototype App