Word Prediction with R and Shiny

Demonstration of a word prediction using R and Shiny

• Corpus generated from text scraping blogs, news and twitter (832 Mb)

• 50% Used as training data, cleaned and tokenised

• Unigrams, Bigrams and Trigrams generated and trimmed to create lookup tables

• The prediction takes a string and integer input

• The last and second last words are used to search through matches in the lookup tables

• Results are added together with adjustable weighting and the score used as a measure of likeliness of a match

• The prediction returns a list of the n most highly scored predictions for the input string

• Go to the Shiny app

• Enter your text and select how many results you want returned with the slider

• Click the Submit button

• Try change the context or use the first two words of common phrases to see how results from the trigram are weighted above bigram and unigram results

• Read the readme file at https://github.com/SeanRossouw/WordPredictionR/blob/master/README.md

• Visit my Github page at https://github.com/SeanRossouw

Thanks to Jeff Leek, Roger Peng and Brian Caffo