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 advice algorithms has reworked music listening, the place customers have an infinite provide of latest music to find. In this empirical examine, we explored a elementary query: How do end-users stream newly launched music? Using knowledge from 1 million Spotify customers’ historic listening and streaming patterns of 282K newly launched music on the platform, we addressed two analysis questions:
RQ1: How does consumption of newly launched music fluctuate throughout genres?
RQ2: How do customers’ style and new music preferences impression their new launch streaming?
TL;DR: We noticed that new releases in genres that always serve useful functions, corresponding to classical music for leisure, are consumed much less. Surprisingly, new releases in Pop typically don’t exhibit excessive consumption charges regardless of being characterised as common music. On the opposite hand, customers’ new launch preferences are distinct from their general music preferences, and constant curiosity in newer content material can predict new music consumption.
How does new music launch consumption fluctuate throughout genres?
To evaluate newly launched music consumption throughout genres, we think about each the variety of listeners and the overall stream rely. A monitor excessive in each the variety of listeners and complete stream rely would point out widespread repeated consumption, suggesting that the customers actually “love” the track. We draw inspiration from the h-index, a metric that assesses the analysis output of scientists. An particular person’s H-index is the most h such that h of their papers are cited at the very least h occasions. Similarly, we outline a music monitor’s h-index as the utmost h such that h distinctive listeners have listened to the monitor h occasions. As such, a excessive H-index would point out {that a}
substantial variety of listeners repeatedly streamed the monitor. We then evaluate genres by inspecting the proportion of tracks with excessive h-indices (as proven within the determine under).
We noticed that New Age, Classical, Children’s, and Jazz exhibit decrease repetition charges, as most tracks have low h-indices. In distinction, Rap, R&B, Country & Folk, and Indie Rock boast a better prevalence of tracks with larger h-indices. Pop falls within the heart of the plot, signifying that Pop songs have average h-indices. In addition, utilizing a multi-output deep neural community regression mannequin, we discovered that style and artist recognition are most predictive of the variety of distinctive listeners for a given newly launched track.
How do customers’ style and new music preferences impression their new launch streaming?
We evaluate 1 million randomly sampled customers’ prior music preferences with their new launch preferences. We use a yr’s value of person historic streaming knowledge as their prior music preferences. For every person, we collect their normalized streaming proportion of latest releases (and non-new releases) within the present interval and calculate the cosine similarity between them and the normalized streaming distribution over the yr earlier than. This produces a distribution of cosine similarities for every person. We observe that the cosine similarity based mostly on new releases (median = 0.89) is lower than that computed based mostly on different tracks (median = 0.98), indicating {that a} person’s new launch style is distinct from their general music style, as proven within the determine under.
Further, we clustered customers based mostly on their month-over-month longitudinal music preferences utilizing content material age (e.g., new releases vs non-new releases vs very previous songs). We noticed that there’s a particular person phase (cluster 3) that constantly prefers newer content material month-over-month, as proven under.
What does this imply for music suggestions?
From our findings, we imagine that algorithms recommending newly launched can each enhance end-user expertise and develop artist viewers by:
- New releases from genres getting much less repeat listening (based mostly on the H-index) might have extra algorithmic boosts to succeed.
- When a track is newly launched, concentrating on customers who constantly discover newer music would possibly assist develop the viewers within the early days of launch.
- New launch advice must be handled as a definite/devoted activity vs. common music advice, given the distinctiveness of person tastes for brand new releases vs. any music.
More particulars in regards to the work might be discovered within 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.