Word Prediction Capstone Project

• Application will predict the next word the user is typing
• Similar applications used by Android, Apple, and Swiftkey to increase the speed of typing text
• Code written for efficiency and was compiled and run on macbook pro
• 90% prediction accuracy on Coursera's benchmark tests

• 19,000,000 words used to train model
• 3 different sources of text data (blogs, news, and twitter)
• Important R packages: dplyr and quanteda
• Similar algorithm used by Google for text prediction in search engine

“Backoff Algorithm used with up to four-gram text tokenization”

• A four-gram is a four word text string “Sally is a dog”
• Tri-gram, Bi-gram, and Uni-gram are lower count phrases
• Backoff Algorithm looks for four-gram match to predict the user's next word. If no four-gram match is found, the model looks for a tri-gram match. If no tri-gram match is found, looks for Bi-gram. etc.
• Application shows the 15 most frequently found matches in training dataset

Click Here >> Word Prediction Ap <<