My fourth Metis project. Using the New Yorker cartoon caption contest submission data collected by NextML and archived here, I used topic modeling to group about 100 cartoons into twenty broad categories, like Restaurants, Dogs, and Work. I then used Max Woolf’s textgenrnn, running on an AWS EC2 instance, to train an RNN on the caption submissions for each cartoon category.
Finally, I created a Flask web app that allows users to generate an unlimited number of new and bizarre captions for each cartoon category.