Celebrity Twitter Generator

The Celebrity Twitter Generator (CTG) collects tweets from 30 different celebrities, and creates a language model out of each one, thus allowing a front-end GUI to dynamically create a tweet in the style of a given celebrity. The program works by creating Markov Chains with a variable N-Gram size, which allows the user to adjust the coherency or accuracy of the generated tweet. As part of evaluating how well the CTG works, I put 301 generated tweets (10 from each celebrity, expect 1 which had 11) onto mechanical turk and had 10 turkers for each tweet attempt to guess the author of the tweet (for a total of 3010 guesses). The system performed admirably, with 1933 out of 3010 guesses being correct (64.22%). Currently this lives as a Java applet, but will soon be on this website for your enjoyment. Update: Now online, converted to a processing app, check it out here: http://benatwork.cc/CTE/applet/index.html