This app predicts the next word in a given sentence.

The algorithm uses the last words of the sentence as the context to predict the new word.

Usage:

• type a few words in the text area and then press ‘Predict word’ button

• the application will suggest the next word of your sentence

For example you can type: ‘as a matter of’, ‘to buy or not to’, etc

Since the algorithm has not been optimised, word prediction can take up to 45 seconds

• Data files contain tweets, news and blog entries. Only a small portion of them is used to create the prediction model

• Texts containing profanity are excluded from model creation

• The algorithm constructs a large set of ngrams. ngrams which appear less than 10 times are discarded from the model

• The algorithm processes a sentence and looks for the largest ngram that contains the last words of the sentence

• If no ngram is found, a shorter ngram is searched and so on. If no ngram can be found, the algorithm returns a null

Word Prediction is a shiny application. It is based on two files:

ui.R

• implements the presentation logic and draws the user interface (UI)
• provides the interactivity with the user

server.R

• loads data, implements the word prediction algorithm
• sends data to the UI

When the user inputs a sentence, the server side is executed and it returns the predicted word to the user.

Application code: https://github.com/claudiolande/10-Capstone/tree/master/WordPrediction

Application access: https://claudiolande.shinyapps.io/WordPrediction/

Data files: https://d396qusza40orc.cloudfront.net/dsscapstone/dataset/Coursera-SwiftKey.zip

List of profanity words: https://www.freewebheaders.com/full-list-of-bad-words-banned-by-google/