Capstone Project - Next Word Prediction

The purpose of the Project has been the generation of a Web App based on Shiny R, able to receive a writting input from a user and predict the next coming word.
Main features of the developed App:

• Reactive update to the writting of the user.
  
• Showing the result of the most probable next word.
• A didactic bar plot is displayed with the top words showing the probability of each one in the group.
• It is based on N-Gram analysis and probability of them.
• Main packages used: tm and RWeka.
  
  

• The Web App is developed in Shiny framework.
  
• There is a side panel and the left in which the user input the text.
• The main panel shows mainly three results:
  
  1. A reproduction of the current text input by the user,
     
  2. The most probable next word predicted,
  3. A plot with the ranking based on probability of the next words results of the analysis.
     

The follwoing is a screenshot of the App described in the previous page. Try it by yourself at the link below:

• The data has been cleaned and subsetting to an reasonable amount considering the tradeoff between processing time and quality of results.
  
• Bi-gram, Tri-gram and Qua-gram tdm matrices constructed (tm & RWeka). All analysis done in lowercase, no punctuation, additional spaces removed. .
  
• Many failure points handled, like unrecognizable words with no prediction, input with excess spaces, Upper/Lower case.
  
• The algorithm detects the input number of words and adapt them for searching based on maximum probability recursively in 4-gram –> 3-gram –> 2-gram database. The top probable word is suggested to the user and the top ranking is shown.
  

• Database is from Swiftkey alliance with Coursera: get it here!
• Main packages used: TM: Text Mining and RWeka: Machine Learning

“This is the End of a Journey for all of us. Long nights working for assessments and Quizzes. For me this has been a real approach to Data Science beyond construction of algorithms for Machine Learning in which I involved before. I learned how to use and interpret many tools and I am equiped with them now. Thanks to all the Instructors and Peer Reviewers for the efforts, dedicating time to check my uploads! Greetings from Chile and see you around!”