This app takes in a phrase from the user and outputs a list of next 3 predicted words.

Training process: It took data from a vast amount of blogs, tweets and news for training and forms a dataset for unigram, bigram and trigram.

Prediction process: For phrase input it receives, it takes into account the last 2 words and predicts the next word by using the 3-gram dataset, which if it fails, will move on to looking at the last word and predict using the 2-gram dataset, which if it fails, then will use the unigram dataset to predict the most common word it learned from the training process

The app takes a dataset of around 90 MB to load up and each input responds its output with the following performance:

• Cold start load time: ~3–5 seconds (loading 90 MB of data)
• Per prediction: under 300ms
• Predictions returned: always exactly 3 words
• Fallback works: unknown words still return a prediction via unigram

The nextWordPredictor app: https://icedmcstuffin.shinyapps.io/nextWordPredictor/

RPubs article explaining the model:(https://rpubs.com/IcedMcstuffin/1428801)

The dataset used to train the model: (https://d396qusza40orc.cloudfront.net/dsscapstone/dataset/Coursera-SwiftKey.zip)

The github repository for this model: (https://github.com/Icedmcstuffin/Text-Prediction-Model)