This presentation is for Data Science Capstone Course Project. We will cover the features of the application, a description of the algorithm and instructions of the user interface.

• Github repo please see: https://github.com/YihaoYang/DataScienceCapstone.
• Shiny Application please see: https://yihaoy.shinyapps.io/next_word_prediction/.

The next word prediction app is located at https://yihaoy.shinyapps.io/next_word_prediction/. The webpage consists of two tabs. The main tab is named as App, where you can enter some texts and the app will try to predict your next word. The second tab About leads to the source code and this presentation.

Please give it a try!

Text data were sampled (10%) from a large corpus of blogs, news and twitter posts provided by Coursera-Swiftkey. After cleaning the sampled data, the N-grams were extracted using the quanteda package. Moreover, sorted token-frequency data frames were built and saved for the use in prediciton part.

• Please see MilestoneReport in the github repo for details of sampling.
• Please see predNext.R in the github repo for details of N-grams generation. Unigram, bigram, trigram, quadgram and pengram were built for the application.

The server first reads in the N-grams built from predNext.R. As text is entered by the user, the app splits the text input into ordered words, feed into the longest n-gram possible to find a match. If a match is not detected, the app switches to a shorter n-gram token-frequency data frame for matching until unigram.

• Please see server.R in shinyApp folder in the github repo for more details.

Thank you!