Queer Encodings: An NLP Study of Challenged Books in the United States

Attempts to ban books through complaints of "Sexually Explicit" content, "Body Description or Function," or just "Homosexuality" have escalated dramatically in the United States in recent years. In response, many researchers and library organizations tag texts as LGBTQIA+ to illustrate the relationship between anti-trans panic and backlash to queer visibility with these book bans. But what makes a text queer? And how do both book banners and defenders constitute new definitions of queerness through their respective projects? Deploying a comparative approach of close reading and two "distant" reading methods represented by challenger complaints and computational topic modeling, this paper explores constructions of queer(ness in) literature in the hopes of critically assessing how classification around book bans protects or endangers queer people and cultures. This thesis draws from computational frameworks – and critiques – from queer and feminist literary scholars, as well as critical catalogers. Computational methods include analysis of metadata from the American Library Association and a BERTopic model trained on a subset of titles challenged from 2010-2023. Ultimately, I demonstrate some of the strengths and weaknesses in deploying topic models to analyze book bans and show that book challenges disproportionately target sex, gender, and queer topics as a source of harm in the texts even compared to more prevalent topics like violence.