From Tumblr to Twitter: Evidence of NSFW Community Migration via Bio Keyword Prediction

Tumblr was one of the most popular microblogging platforms throughout the 2010s, boasting loose regulations around adult content that fostered the formation and growth of various fandom communities. After Tumblr’s December 2018 ban on adult content, these communities were driven to migrate to another platform. This study provides evidence for a migration of users from Tumblr to Twitter following the ban using bio keyword prediction. We collect 400,000 Twitter bios for every year between 2017 and 2022 and train a logistic regression model to predict the presence of ‘porn’ and ‘NSFW’ given the other tokens in the bio and evaluate the models on 100,000 held-out bios. Through the learned model weights, we find evidence for ‘porn’ being primarily predicted by keywords advertising nudity for sexual consumption, while ‘NSFW’ correlates with accounts of individuals that present nudity as just one part of their online presence, mirroring how NSFW existed on Tumblr. Using the models’ testing accuracy as a proxy for predictability, we find that both ‘porn’ and ‘NSFW’ originally existed in small niches on Twitter which grew in scope over the years. Taken together, we conclude that there was a mass influx of users from Tumblr to Twitter following the porn ban that (re-)formed community around the keyword ‘NSFW’. Finally, we make available BioKPD (Biographical Keyword Prediction Dataset), a dataset of model weights and testing accuracies for n=1630 frequently used identity signifiers.