Word Prediction app
Abhay
9-OCT-2016
Executive Pitch
We present you the Word Prediction app, which is intended to compliment several text input application.
By being able to predict the next word to be typed the algorithm can substantially reduce the typing effort on the part of user.
Care has been taken to optimize the loading times and processing speed of the algorithm for smooth user experience.
Working Methodology
The model makes use of Ngrams model (till quadgram). The input training corpus was used to form lookup tables (upto quadgrams).
User Input text is first cleaned and last three words are then used to form input to the unigram, birgram & trigrams lookup tables. Frequency count of ngrams are moderated by log transformation.
Also inorder to reduce the size of lookup tables word coverage of 90% was used. For prediction, model makes use of a modification of stupid back-off algorithm. Scoreing is done parallely for three ngrams where 0.4 penality is imposed for subsequent lower level ngram lookup hits.
Incase of new input not covered in the app, we are simply displaying the most frequent unigrams.
Application Usuage - Settings tab
Application is divided into two tabs
Settings Tab:
Click on the Load Ngram button to upload the multiple Ngram tables.
A Message on right will display the action results. Ideal state all 4 ngram tables should load up successfully. Word Cloud on Most frequent word is shown on successfull loading.
Application Usuage - Prediction tab
Text Prediction Tab: Enter the phrase into the into the input box and press the “Predict Next word” button. Output should display a table of upto top 20 next predicted words and its word cloud with word size weighted with word possibility scores.