• Here is the App link: Shiny App

Objectives

• This app predicts the next word(s) with one or more words as input
  • For this, a large corpus of twitter, news and blog content was analyzed
  • We extracted N-grams from the corpus and used them to build the predictive model
  • We also explored various models for improving the prediction accuracy and speed

• N-gram model with back-off strategy has been used to train the algorithm
• Dataset was been cleaned, lower-cased, links, twitter handles, emojis, punctuations, extra whitespaces, numbers etc. removed
• Matrices from uni-gram to hexa-grams were extracted and sorted by frequency of occurrence
• Size of model was reduced by dropping least frequent N-grams
• Speed and memory usage was further optimized by dropping the least frequent bigrams and monograms since they do not appear to improve accuracy

• Input Word(s): text input box for user to type a phrase / word
• The words typed are detected and the next word(s) predicted reactively
• Output iterated from longest N-gram (hexagram) to shortest (bigram)
• The last word in matching N-gram is used as predicted word
• Predictions are made using the longest, most frequent, matching N-gram
• If no matches are found using the existing {6:2}-grams, it selects the most frequent word from monogram
• User can configure the number of words the app should suggest