Originally an LSE MSc dissertation (Applied Social Data Science, 2024).
A Spotify field experiment on how recommendation algorithms form and reinforce music-genre tastes, measured along variety and atypicality with a co-genre network — connecting cultural sociology with computational methods. The study uses the Spotify API and the Million Playlist Dataset, and builds a co-genre network (3,824 nodes, 46,711 edges) to quantify genre "atypicality." Controlled accounts are classified as mono-purist, mono-mixer, poly-purist, or poly-mixer, plus controls, and analyzed with one-way ANOVA, repeated-measures ANOVA, and Tukey's HSD post-hoc tests.
Algorithms narrow diversity over time for users who start with broad, typical tastes, producing more homogeneous listening. Users with narrow initial tastes see some diversification, but within constrained genre boundaries.