Next Word Predictor

Our purpose today is to introduce to you our predictive text model within the confines of the Capstone Project of Data Science Specialisation at JHU.

SwiftKey, the corporate partner in this capstone, builds a smart keyboard that makes it easier for people to type on their mobile devices. The cornerstone of their smart keyboard is predictive text models.

To showcase our Next Word Predictor:

• We will briefly explain how our predictive model Algorithm is built
• We will show the user how this data product works
• We will review the architecture of the Next Word Predictor

The working Model is available for review URL: 'http://frankhasbani.shinyapps.io/final/'
Additional information about the preparation of the model URL: 'http://rpubs.com/FRH123/69391'

We're applying the Markov Chain, using 2, 3 and 4grams, we smooth with the following Interpolation formula:
And we apply the Katz Backoff with K = 3

Read the Instructions, Type in a word, partial or whole sentence, Click the Button, and Experience the predictions with more words

The above caption is the Solution Architecture

Current Status Next Word Predictor and Future plans include but not restricted to the following:

• Using up to QuadGram, Interpolation, Smoothing, Backoff at k=3, and returns 3 results.
• Grams have been pruned to freq of 3 in Bigrams to 1 for Quadgram to manage memory size limitations
• Lambda, Discount ratio and Alpha have not been fine tuned for now, but will be soon.
• Future plans include reactive Autocorrect and Prediction, Addition of new terms, Comparative Testing
• Long Term Plans would include Domain specific applications, Law, Medicine, etc, and Other Languages such as Portugese, French, German, Italian, Spanich etc.