Characterizing End-User Auditing of Recommender System Behaviors Affecting Music Discoverability: a Diary Study

Abstract

Recommender systems shape what people encounter online, yet their role in facilitating or constraining music discoverability is rarely examined from the listener’s perspective. To help address this gap, we present a two-week diary study with nine Italian young adults who reported problematic recommender system behaviors as encountered in everyday music listening. Combining quantitative and qualitative analysis to characterize distinct auditing practices, we show that detection, sense-making, and response to such behaviors are structured by engagement mode and prior platform knowledge. Notably, for some participants, narratives shifted over the study toward more selective platform use, while most described their platform relationship as unchanged, and even those who recognized potential harm to artists continued using the audited platforms, nonetheless adapting their navigation strategies and developing workarounds. Building on these insights, we propose a cumulative spiral as a frame for end-user algorithmic auditing, in which repeated encounters may support algorithmic literacy through interaction, sensemaking, and folk theorization, which in turn shapes future detection and response.