How do people stream newly released music?

May 02, 2024 Published by Julie Jiang, Aditiya Ponnada, Ang Li, Ben Lacker, and Samuel F Way

The rise of on-demand music streaming supported by novel recommendation algorithms has transformed music listening, where users have an endless supply of new music to discover. In this empirical study, we explored a fundamental question: How do end-users stream newly released music? Using data from 1 million Spotify users’ historical listening and streaming patterns of 282K newly released music on the platform, we addressed two research questions:

RQ1: How does consumption of newly released music vary across genres?
RQ2: How do users’ genre and new music preferences impact their new release streaming?

TL;DR: We observed that new releases in genres that often serve functional purposes, such as classical music for relaxation, are consumed less. Surprisingly, new releases in Pop generally do not exhibit high consumption rates despite being characterized as popular music. On the other hand, users’ new release preferences are distinct from their overall music preferences, and consistent interest in newer content can predict new music consumption.

How does new music release consumption vary across genres?

To compare newly released music consumption across genres, we consider both the number of listeners and the total stream count. A track high in both the number of listeners and total stream count would indicate widespread repeated consumption, suggesting that the users truly “love” the song. We draw inspiration from the h-index, a metric that assesses the research output of scientists. An individual’s H-index is the maximum h such that h of their papers are cited at least h times. Similarly, we define a music track’s h-index as the maximum h such that h unique listeners have listened to the track h times. As such, a high H-index would indicate that a 

substantial number of listeners repeatedly streamed the track. We then compare genres by examining the proportion of tracks with high h-indices (as shown in the figure below).

We observed that New Age, Classical, Children’s, and Jazz exhibit lower repetition rates, as most tracks have low h-indices. In contrast, Rap, R&B, Country & Folk, and Indie Rock boast a higher prevalence of tracks with higher h-indices. Pop falls in the center of the plot, signifying that Pop songs have moderate h-indices. In addition, using a multi-output deep neural network regression model, we found that genre and artist popularity are most predictive of the number of unique listeners for a given newly released song.

How do users’ genre and new music preferences impact their new release streaming?

We compare 1 million randomly sampled users’ prior music preferences with their new release preferences. We use a year’s worth of user historical streaming data as their prior music preferences. For each user, we gather their normalized streaming percentage of new releases (and non-new releases) in the current period and calculate the cosine similarity between them and the normalized streaming distribution over the year before. This produces a distribution of cosine similarities for each user. We observe that the cosine similarity based on new releases (median = 0.89) is less than that computed based on other tracks (median = 0.98), indicating that a user’s new release taste is distinct from their overall music taste, as shown in the figure below.

Further, we clustered users based on their month-over-month longitudinal music preferences using content age (e.g., new releases vs non-new releases vs very old songs). We observed that there is a specific user segment (cluster 3) that consistently prefers newer content month-over-month, as shown below.

What does this mean for music recommendations?

From our findings, we believe that algorithms recommending newly released  can both improve end-user experience and grow artist audience by:

  1. New releases from genres getting less repeat listening (based on the H-index) may need more algorithmic boosts to succeed.
  2. When a song is newly released, targeting users who consistently explore newer music might help grow the audience in the early days of release.
  3. New release recommendation should be treated as a distinct/dedicated task vs. general music recommendation, given the uniqueness of user tastes for new releases vs. any music. 

More details about the work can be found in the following paper:
A Genre-Based Analysis of New Music Streaming at Scale. 
Julie Jiang, Aditya Ponnada, Ang Li, Ben Lacker, and Samuel F Way. 
WEBSCI 2024.