UPDATE: Scottbot has been removed from GitHub and will not be coming back. Find out why.
After talking about it with Fred for a couple of weeks, I sat down this morning and started scottbot, an IRC bot that will learn how to make good “that’s what she said” jokes.
It’s built in Node.js on two libraries by the inimitable Harth Vader, more commonly known as Heather Arthur: nomnom, an excellent argparse/optparse implementation for node, and brain, a machine learning library with neural net and Bayesian classifier support. So thanks, Heather!
I’m also using a slightly hacked node-irc from Martyn Smith.
Right now it’s a pretty simple Bayesian method. It’s been learning all day on Moznet and has made one or two pretty-good jokes, and a couple of stinkers. I think it’s already better than our other TWSS bot.
There’s plenty of things to do to improve it. I don’t know a whole lot about machine learning or anything Bayesian beyond what I can recollect from my old stats-for-math-majors class, so if you’ve got any suggestions, I’d love to hear them (especially in the form of pull requests)! I want to learn, too.
If you’ve got node and Redis, you, too, can have your very own scottbot.
Scottbot, and it’s default IRC nick, “mscott,” is named affectionately after Michael Scott.