Next Word Prediction helps complete sentences on mobile phones and tablets. This project explores how to predict the next likely words in real-time using a shiny app in R. we use a Data Set provided by Swiftkey.

• For our model, we used N-grams tokenization, more specifically for N=2, and N=3. After sampling 10% from the given data setand some cleaning We calculated the frequency and probability for each n-gram, removing The resulting indexed data.tables are stored for real-time prediction.

• We use a backoff algorithm for predictions, starting with 3-grams, then 2-grams.

The Milestone Report.

The report which ca be found here compiles basic statistics about the full three data sets, displays strategies for cleaning and processing a corpus which will be used as a training set for our model. The report also show the most frequent words (unigrams) and pairs of words (bigrams) across the three data sets.

The Shiny App.

it can be found here

The First Page.

The Second Page.

We used various sample phrases from three data sets, and our app was able to predict the next word with a 73% accuracy rate. The main challenge we faced in this project was memory usage. We believe that the accuracy of the N-gram prediction model, which employs backoff probability selection, could be improved by including N-grams with larger values of N in the selection process. Unfortunately, as N increases, more memory is required to store the resulting N-grams. Our computer was unable to handle values larger than N=3.