Genres, subgenres, and storytelling tropes: a data science approach

Abstract

**Author(s):** Sebire, I; Camargo, C; Introduction: Popular media plays a significant role in shaping and reflecting culture and society. In this paper, we investigate patterns in the use of recurring storytelling devices, i.e. common tropes, in popular media. Examples of tropes include the "mad scientist", "villain wearing a dark palette", or even "melancholic song playing during sad moment". Using a database of media tropes, we compare clusters of tropes with actual genre classifications, and study the differences in media through geography and time. Methods: This study uses a dataset collected from TVTropes, a wiki-format crowdsourced website describing tropes in popular media. We apply approaches from cultural evolution along with data science methods such as topic modelling, sentence embedding, and stochastic block models, to simultaneously cluster tropes and the works they feature in. # Results & Discussion: We show the applicability of the nonparametric stochastic block model-based topic model topSBM to identify a hierarchy of genres and subgenres, as well a hierarchy of storytelling tropes present in the data over time. We end by discussing the strengths and limitations of machine learning in cultural evolution.