SwiftPredict. Ngram text prediction model

is a text prediction model for mobile keyboards. Given some text input, it produces the next word in the sequence, saving keystrokes and time.

• Technology: Implemented in R using quanteda package for text mining and data.table for data storage and search optimization

• Data set: plain text files harvested from Twitter, blogs and news sites

  1. 556 MB
  2. 2.4 million lines of text
  3. 105 million words
  

• Model: stupid backoff using 1- to 6-grams

  1. 22.8% accuracy for top-3 predictions
  2. 20,000 words dictionary
  3. 226 milllion ngrams analyzed
  

https://dmitrytoda.shinyapps.io/SwiftPredict/

• accepts arbitrary text as input
• produces top-3 predictions
• explains how the predictions were obtained