I based it off a Ruby module called “Twitter Ebooks”, which I believe was based on a Python module called “horsey books”, which could create a bot that would make posts using Markov chains, given a corpus of text. [2]
(I’m not dropping the date to be a jerk, more to point out that Markov chains / n-grams are an interesting concept for experimenting with language models and have been around for a while.)
In my case, I fed a corpus of data (my Twitter history at the point, which was something like 20K tweets) and tried to create a Twitter bot to make intelligent sounding posts.
It worked out, sometimes. But, it was a lot of fun to figure out!
[1] “Building intelligent robots using Node.js that can conquer Twitter” - https://www.youtube.com/watch?v=rMmXdiUGsr4