@bronnerbot was my first ebooks-style Markov Twitterbot. I had the idea after absentmindedly reading a Dr. Bronner’s soap bottle, which is a treat I recommend to anyone who hasn’t already had the pleasure. The label, rambling, whimsical, and circuitous, almost seems as if it were generated from a Markov chain rather than written by Dr. Bronner himself. So, I figured, why not train a model on copy transcribed from each individual label?
A PSALM OF SOAP EXCELLENCE— bronnerbot (@bronnerbot) July 9, 2017
I found several labels ready-transcribed online, but for the remaining few I found an image and typed by hand. (This explains bot typos, which happen every so often). Each variety of soap has a slightly different label, which made for an 8,000+ word text file to train the model. After generating the model, I scheduled the bot to tweet every 6 hours and @bronnerbot was live. It has been tweeting regularly since 2015, and has tweeted over 5,000 times. The bot gained only a handful of followers (39, at time of this writing), but ultimately I made the bot for myself. Most of its tweets are nonsensical riffs on ‘ALL-ONE! ALL-ONE!’ (which is a crutch Dr. Bronner leaned on more often than any other), but occasionally the bot tweets something serendipitously poetic. I’m convinced all that transcription was worth it.
Soap can make Love last 1 trillion years of sweet eternities!— bronnerbot (@bronnerbot) June 16, 2017