87

Streaming music listening services are halfway between two previously existing devices – radio and record collections –, and two related types of agency – the consumer supported by music programming in the case of radio, and the amateur in the case of collection. They therefore offer a hybrid system of personalized services, adapted to users’ tastes, based on their past listening, through the use of recommendation systems. To what extent is the figure drawn by the interfaces reflected in the listeners’ real behaviour? To answer this question, we draw on the analysis of logs of musical activity from a random sample of 4,000 users of a musical platform, monitored over a period of five months. We first describe the distribution of listening, and show that while the diversity of consumption on the platform is greater than on other devices, listening remains highly concentrated on the most popular musicians. We then examine the use of recommendations, showing that it remains moderate and unevenly distributed, and that for musical discovery, it is still marginal, whether the orientation of the recommendations is towards mid-tail and long-tail music.