Coursera Data Science Capstone - Final Project

This project corresponds to the final assignment of the Data Science Capstone Coursera course by Johns Hopkins University.

The main goal of this project is to create a shiny application containing a predictive text model capable of predicting the next word against a sentence entered by a user.

For Shiny app: https://duplo59.shinyapps.io/Data_Science_Capstone_Project/

Before building the prediction model, data have been preprocessed and cleaned. Steps:

• reading of the three given txt files about blogs , news and twitter;
• file cleanup (strip whitespace, removing punctuation and numbers, character conversion to lowercase);
• create an unique file;
• file sampling;
• creation of the corpus;
• tokenization and buildings of 2-grams , 3-grams and 4-grams dataframes;
• saving n-grams files as RData files (compressed files).

As the application starts, the RData files are loaded . Then the sentence typed by the user is cleaned (eg. punctuation is removed) and analyzed with the following criteria:

• quadgram is first used (looking for the first three words of quadgram using the last three words of the sentence provided by the user);
• if there is no match, then trigram is used (looking for the first two words of trigram using the last two words of the sentence provided by the user);
• if there is no match, then bigram is used (looking for the first word of bigram using the last word of the sentence provided by the user);
• if there is no match, a 'no match found' warning is displayed, reporting the word with the highest frequency ('the').