Predicting next word Shiny app

The app is available at:
https://pavelzimin.shinyapps.io/nextWord/

• This Shiny app was created to fulfill requirements for the Capstone Project for the Data Science Specialization at Coursera.
• The objective of this Shiny app is to predict the next word from the preceding text.
• This task is widely used in mobile phone applications to assist the user with completing a typing task.

• The corpora were provided by the Data Science Specialization at Coursera.
• They were collected with the web crawler from online sources.
• The provided corpora are collected in 4 languages: Finnish, Russian, English and German. This report analyzed the English data set.
• The data were collected from blogs, news and twitter.

• The data were first read and cleaned by changing letters to lowercase, removing punctuation and removing profanity words.
• Data from blogs, news and twitter were combined and sampled for further analysis. Due to the large size of the provided data set, only 1/100 of the data were used.
• Unigrams, bigrams and trigrams were identified and their frequencies.
• Katz Back-Off Algorithm was used for creating the model. In this algorithm trigrams were used with some weight redistributed to the bigrams and unigrams.
• 5-fold Cross Validation was performed to optimize the model parameters.