The curation of beauty ideals through Instagram's recommender system: An interdisciplinary approach

This paper investigates how Instagram's recommender system participates in reproducing beauty ideals through an interdisciplinary approach that combines ethnographic methods with supervised machine learning. We conceptualise beauty ideals as socioculturally constructed and intersubjective perceptions and adopt the theoretical framework of ‘curatorial assemblages’ in examining how recommender systems shape feeds. Our novel methodological approach involves training a statistical classifier using an ethnographically informed and manually annotated dataset. The classifier determines whether a post reproduces beauty ideals (RBI) or does not reproduce beauty ideals (NRBI). On this basis, we analyse how the curatorial practices (liking, saving and following) affect the Explore feed composition regarding beauty ideals. Our findings show that Instagram's recommender system strongly amplifies beauty ideals when users routinely engage with such content through likes and saves. Conversely, for content not reproducing beauty ideals, we find that the Explore feed is less adaptive, resulting in feeds that feature fewer posts depicting persons. Experiments to counter-curate showed minimal impact on feed composition. We discuss our findings against the theoretical backdrop of curatorial assemblages and elaborate how both collaborative and content-based recommendation may contribute to the reproduction of beauty ideals. We conclude that Instagram's recommender system structurally entrenches beauty ideals within the platform's global curatorial assemblage. These findings highlight the need for greater algorithmic transparency to enable informed user decisions and critical media literacy regarding automated feed curation.