What's Next (Text Prediction Application)

Omayma Said
30 Dec 2016

What's Next is a Shiny application that uses NLP models for text prediction. It provides two options as shown in the screenshots:

-What's Next on Shinyapp.io: HERE
-Created by OmaymaS

Next Word Prediction

Enter text to get the top-3 predicted words based on your input. (similar to mobile phones and other gadgets' keyboards).

Input	W1	W2	W3
here we go	to	on	again
I'd like to see	you	the	what
I'd like to say thanks for the	follow	first	RT

Trigram Model Probaibiliy

Enter three words (Trigram) to get its probability

Input	Output
here we go	P(Here we go)
and we can do the	P(can do the)
the first	Not found in the trigram model

If Input>3 word, only the last 3 are considered.
If Input<3 word, or if the trigram is not found in our model, the user gets a message "Not found in the trigram model

Using 70% of the given dataset:

data was cleaned removing symbols, profanity, extra spaces, non-latin characters and certain words/letters.
unigram, bigram and trigram datatables were generated with the count of each entry.

Good-Turing discounts were calculated for Nr<=5 (Nr: frequency of frequency).
Probability is calculated using Katz’s Backoff model as shown in the example in Fig-2.

Advantage: Reactive

Tradeoffs: Accuracy Vs. Speed

Areas for Improvement: Application Startup Speed