February 4, 2018

What is it?

  • The Next Word Predict-ROAR is an app that pulls from news and Twitter data to tell you what the next likely word will be given your chosen phrase
  • The user's input will be stripped of punctuation, numbers, and turned into lower case
  • Currently, the app can search up to 4 consecutive words to predict the 5th (any inputs greater than 4 will be shortened to the last 4 words)

How does it work?

  • The Model is based off the Stupid Back-off Model. The Next Word Predictor will try to find the user's search in the appropriate n-gram table (up to 4-grams).
  • If a 5 word phrase is entered, the model will shorten it to 4 words then search for a match and, if there is a match, calculate the likelihood of the next word given the number of occurances of that phrase divided by total occurances of the phrase.
  • If there is no match, the model will back off one less word to only the last 3 words and repeat the process.
  • If the model can not find a match, it will display the top 3 most common words.

Performance

  • Out of 20 random phrases, the app was able to predict the correct word in the top three 40% of the time.
  • The app could be improved with the creation of 6-grams (or more) to allow the model to search for 5 words in a row. With your generous funding this could be possible!

Try it out yourself!

  • Just enter a phrase in the text box to see the top 3 likely words [there may be a small delay]
  • Enter another phrase to continue the fun!
  • In the app, click on the Explanation tab for more information and source code
  • Go have some fun! Next Word Predict-ROAR

Thanks!