July 26, 2022 Revealed by Lillio Mok, Samuel F. Means, Lucas Maystre, Ashton Anderson
“Selection is the spice of life”, because the saying attributed to poet William Cowper goes. Folks crave heterogeneity and keep away from boredom throughout all features of their on-line experiences. As providers that present huge quantities of content material for person consumption, streaming platforms like Spotify are eager to higher perceive how folks hunt down selection to maintain themselves and engaged. If we will determine when and how listeners need to develop their musical tastes, then mechanisms like recommender programs might assist them discover novel items of music.
Our present information of how folks discover assorted items of content material is proscribed to self-reported, usually inconsistent, proof supplied by small teams of individuals in slim time slices. As an example, the scientific literature is considerably undecided on how folks search for heterogeneity over time, with two apparently incompatible concepts:
- Choice 1: Folks “ossify” as they get older and grow to be much less in selection as a result of they already know what they like and are much less simply bored.
- Choice 2: Folks broaden as they get older and grow to be extra in selection as a result of they’ve skilled extra issues and have refined palates.
So, how do folks truly eat heterogeneous content material as they get older? On this examine, we examine how a big group of Spotify listeners actively discover musical selection at totally different factors of their life cycles.
Exploration on Spotify
To begin with, what does “selection” even imply? A method of measuring variety-seeking behaviours is to determine when individuals are exploring particular person items of novel content material. If an individual is listening to a monitor for the very first time on Spotify then that might be a really novel tune discovery. Alternatively, if the listener hasn’t listened to a monitor shortly and comes again to it later, then the monitor is barely regionally novel. To maneuver between items of regionally novel content material throughout totally different time frames (weekly, for the aim of this examine), listeners must flip over the tracks they’re listening to.
Conceptual schematic of exploration at totally different time scales.
We measured how usually folks found and turned over musical content material throughout 100 thousand US-based Spotify Premium customers between 2016 and 2019, an entire listening hint of over 8 billion distinctive listening occasions.
The Offline Lifecycle: Ageing and Exploration
The primary and most simple mind-set about lifecycles is the trajectory alongside which customers develop offline – i.e. their age. We in contrast how a lot and when folks explored throughout totally different age teams and located three key outcomes.
Firstly, youthful listeners constantly discover much less throughout our dataset. For a similar variety of tracks listened to and at totally different factors in calendar time, they’re prone to have fewer discoveries and to have decrease content material turnover. At a primary look, this implies that listeners don’t ossify or grow to be narrower as they age, and in reality broaden their consumption habits as they develop.
Anticipated variety of discoveries per stream listened to separate by imply person in age group (left) and total distribution (proper). Baselines in gray (stable: every hear is a discovery; dashed: discoveries drawn from a hard and fast set).
Secondly, exploration by no means stops. Folks don’t behave like they’re drawing from a finite set of content material that may fulfill them, as an alternative persevering with to build up novel content material over time. Nevertheless, that is diluted by repeat listening of previously-discovered tracks, which might be seen within the rising must turnover content material over time. Folks more and more must cycle by way of tracks between time frames to take heed to the rising variety of tracks they found. Once more, it doesn’t seem like folks ossify and cease exploring over time.
Anticipated content material turnover fee every week, cut up by age group.
Thirdly, exploration is topic to different components like seasonality and the best way through which listeners work together with the platform. Exploration spikes round Christmas, for instance, throughout which individuals could also be influenced to hunt out season-specific music. Moreover, customers are prone to uncover novel music at a constant fee when listening to programmed music like by way of algorithmically-curated radio.
The On-line Lifecycle: Exploratory Phases
We’ve regarded on the offline lifecycle, however what occurs throughout a person person’s on-line lifecycle – particularly their tenure on the platform? This time, we studied how folks distribute exploratory behaviours inside their particular person traces. Is exploration unfold out and dispersed, or is it clumped collectively and clustered?
Illustration of how exploration might be grouped collectively or unfold out over time.
To determine this out, we checked out how a lot listeners explored between time home windows (once more, we used every week) of their particular person traces and measured the invention fee interquartile vary (IQR) over home windows with a number of streams. In brief, the upper the IQR, the extra exploration is clustered into exploratory home windows with many discoveries. We discovered that younger customers have a lot decrease IQRs than older customers, indicating that they discover extra evenly than older customers. In distinction, older listeners group their discoveries into clusters.
Distribution over discovery fee IQRs between totally different time home windows, inside every person. If 25% of a person’s exercise in home windows is made from pure discoveries whereas the remaining 75% home windows had 50% discoveries, their IQR can be 50%.
However what in regards to the distribution of those home windows? We discovered that customers truly chain exploration-heavy home windows collectively. In the event you already began exploring, you usually tend to preserve going; when you’ve got not began exploring but, you’re additionally extra prone to not begin within the subsequent week. So, there appear to be phases of each content material exploration and content material revisitation.
Chance of every week having many discoveries after a earlier week of many discoveries (deciles). Listeners are prone to chain many discoveries collectively (backside proper sq.) or go on prolonged runs with few discoveries (high left sq.).
Content material Exploration and Style Variety
We’ve mentioned exploration loads thus far. Whereas it has its benefits as a measure of selection – it’s granular, considers particular person items of content material, and is time-sensitive – it additionally doesn’t take note of how comparable two items of content material are. An individual who listens to Home and Electronica music might be not as variety-loving as somebody who listens to Bluegrass and Ok-Pop. So, we additionally in contrast our exploration metrics to style range (see right here). This range metric measures how broad (and never essentially novel) somebody’s listening habits are, and is scaled in order that 0 could be very specialised and 1 could be very numerous consumption.
Left: a listener’s weekly style range predicts their range within the subsequent week; Proper: weekly style range is poorly correlated with content material exploration.
It seems that range and exploration are virtually unrelated, each on the weekly timescale and throughout your entire hint. By the use of comparability, a person’s range in a single week is pretty predictive of their range within the subsequent week. In actual fact, it seems as if youthful listeners are the most numerous, whereas older listeners are extra specialised. By the lens of range, then, there is some assist for the ossification speculation. As listeners get older, they seem to take heed to a narrower set of content material.
Listeners’ weekly style range over time, cut up by age teams. Age ordering is reversed in comparison with discoveries over time.
What do these outcomes inform us? For one, that selection is a posh assemble. The totally different, seemingly incompatible theories about how folks search selection as they develop could also be defined by other ways of satisfying heterogeneity wants. Our outcomes are in step with youthful folks being generalist repeat customers who don’t usually search for very novel music in comparison with what they know, however the music that they do know is comparatively broad. Older listeners are specialist explorers, who search for a selected and doubtlessly ossified set of music however are additionally consistently discovering and turning content material over.
Moreover, we uncovered various factors that will affect variety-seeking habits and thus listeners’ receptiveness to new music. On the one hand, exploration might depend upon how folks work together with programmed music, new releases, and totally different genres. On the opposite, exploration additionally fluctuates throughout seasons within the calendar, in addition to in phases inside people’ listening trajectories. These components might doubtlessly assist on-line platforms determine when and the way customers ought to be guided in the direction of assorted content material to fulfill their wants and preserve them . Extra info might be present in our paper:
The Dynamics of Exploration on Spotify
Lillio Mok, Samuel F. Means, Lucas Maystre, Ashton Anderson
ICWSM 2022