6/18/2017
Overview
- Prediction Algorithm - Efficient Modelling - Ngram backing off to next smallest Ngram
- Quick predictive response for use with mobile texting
- Using the last for 1-5 words of the text to predict their next word
Algorithm
- Simple Backoff A ngram language model that estimates the conditional probability of a word given its history. Each n-gram structure is searched for from 5-grams to 1-grams for a match. There was no smoothing is applied.
Instructions
- Input : Enter your message as mobile text
- Output : As you type, the recommended predicted next word(s) will appear.
App Design
- The applications utilizes shiny framework with 'ui.R' and 'server.R'.
- The predictive model model consists of some R code and data structures provided by the R data.table stored in the cloud. The code uses SQL to search the ngram data structures for matches.
Summary
- Our goal was create a simple real-time predictive next work as you typying for mobile texting to replace existing iphone and android applications.