Swiftkey Project

Capstone SwiftKey Predictive Text Model
ShinyApp For Predictive Text Modelling




When someone types:

I went to the

The keyboard presents options for the next word.
The predicted words could be

gym, store, restaurant

Swiftkey - Features / Corpus

Features

  • Free Form Text
  • Prediction Based On
    Last Character Typed
    • [punctuation]
      No Prediction
    • [alphabet]
      Word As Incomplete;
      Predicts Current Word
    • [space]
      Word Complete;
      Predicts Next Word
  • Probability Chart
  • Word Cloud

Corpus Statistics

  • Blogs
    Lines:899K / Words:37M
  • News
    Lines:77K / Words:2M
  • Tweets
    Lines:2M / Words:30M

  • N-Grams Prediction Model
    N-Grams-1 : 113K Terms
    N-Grams-2 : 532K Terms
    N-Grams-3 : 522K Terms
    N-Grams-4 : 108K Terms

Swiftkey - NGrams / Alogorithm

NGrams

  • NGrams-1 Model
      Word   Freq
    1 just 253008
    2 from 243195
    
  • NGrams-2 Model
      Term     Freq
    1 and the   755
    2 good day 1352
    
  • NGrams-3 Model
      Term            Freq
    1 thanks for the   256
    2 for the follow    82
    
  • NGrams-4 Model
      Term            Freq
    1 is a good day    125
    2 best of the lot   99
    
    Note:
    Model Only Contains
    Terms With Freq >= 2

Algorithm

  • For Input String
    Today Is A Good
  • NGrams-4 Search - Last 3 Words; The Search String
    is a good
  • The NGram Found
    is a good day
  • The Predicted Word
    day
  • Weightage When Search Results Have Many Words
    N-Gram-1 :  1 * Freq
    N-Gram-2 :  1 * Freq
    N-Gram-3 :  5 * Freq
    N-Gram-4 : 10 * Freq

Swiftkey Project - ScreenShot

alt text

Swiftkey Project - Signoff

Coming Soon

  • Spell check along with Auto Complete
  • Add Data To Corpus To Personalize Your Model

For Online Demo

For Online Documentation

  • Select “Documentation” tab in above URL

For No Obligation Full Featured Trial Account Contact