Capstone Project - Text Prediction

This is an introductory pitch presentation that introduces the text prediction final project for the data science specialization.

The aim of the project is to:

• Use the input data from swift key efficiently to build a data base.
• Build an algorithm using the n-gram and conditional probabilities models.
• deploy the working product to a shiny application for the user.

• The n-gram model was built using a unigram and trigram.

• The unigram was built on the previous two words:

• P(wn|wn−1)=C(wn−1wn)C(wn−1)

• The trigram was built on the previous three words:

• P(wn|wn−2wn−1)=λ1P(wn)+λ2P(wn|wn−1)+λ3P(wn|wn−2wn−1)

• The frequency of the least observed n-gram was assumed for the unobserved n-gram

• The Data was 10% from the Twitter, Blogs, and News data.

• The shiny app is built around searching the optimized data base built on the algorithm on the previous slide.
• It loogs for trigrams, then ngrams, then unigrams - which are list of the most common words used.
• The app works by just entering the input text, and it tries to predict the next word.
• If there is no input, the algorithm will recommend three of the most common english words.

There are a couple of things that might enhance the working of this algorithm

• Using more of the input dataset to expand the database.
• Using n-grams which have n > 3.
• Better optimizing the serach functoins to search the recommendation quickly.