Musical tastes fluctuate throughout the day: "By integrating an artificial neural network with Spotify’s API, we show a general awareness of diurnal preference in playlists"

Diurnal fluctuations in musical preference. Ole Adrian Heggli, Jan Stupacher and Peter Vuust. Royal Society Open Science, November 10 2021. https://doi.org/10.1098/rsos.210885

Abstract: The rhythm of human life is governed by diurnal cycles, as a result of endogenous circadian processes evolved to maximize biological fitness. Even complex aspects of daily life, such as affective states, exhibit systematic diurnal patterns which in turn influence behaviour. As a result, previous research has identified population-level diurnal patterns in affective preference for music. By analysing audio features from over two billion music streaming events on Spotify, we find that the music people listen to divides into five distinct time blocks corresponding to morning, afternoon, evening, night and late night/early morning. By integrating an artificial neural network with Spotify's API, we show a general awareness of diurnal preference in playlists, which is not present to the same extent for individual tracks. Our results demonstrate how music intertwines with our daily lives and highlight how even something as individual as musical preference is influenced by underlying diurnal patterns.

Statement of relevance: Today, most music listening happens on online streaming services allowing us to listen to what we want when we want it. By analysing audio features from over two billion music streaming events, we find that the music people listen to can be divided into five different time blocks corresponding to morning, afternoon, evening, night and late night/early morning. These blocks follow the same order throughout the week, but differ in length and starting time when comparing workdays and weekends. This study provides an extremely robust and detailed understanding of our daily listening habits. It illustrates how circadian rhythms and 7-day cycles of Western life influence fluctuations in musical preference on an individual as well as population level.

3. Discussion

In this work, we have shown that the rhythms of daily life are accompanied by fluctuations in musical preference. We show that the diurnal patterns of audio features in music can be treated as five distinct subdivisions of the day, with the musically meaningful distinction between them found in the range and distribution of the musical audio features. Our follow-up studies indicate that individuals hold a general awareness and agreement of diurnal musical preference in playlists consisting of multiple tracks, but that single tracks do not necessarily elicit the same diurnal associations. Taken together, this points to the circadian rhythms governing life being reflected in the highly individualized and often subjective preference for music.

The next step in this line of research would be to examine the degree to which the diurnal patterns documented herein reflect universal psychological phenomena in music perception. As previously discussed, some types of music often occur at a specific time of the day and often with a clear link to activities, with perhaps lullabies being a prime example. As lullabies are intended to ease falling asleep, they tend to occur at night and have been found to have partly universal features such as reduced tempo [40–42]. If similar time-dependent songs could be collected into a database, it would then be highly interesting to investigate if the audio features of such songs match up with the features that drive the time-of-day preferences uncovered herein. Here, the Spotify API's ability to search user-made playlists for name and description is a highly productive approach, as shown in a recent study uncovering a large amount of variation in sleep music [43].

While the diurnal patterns in musical audio features uncovered in this work are robust and consistent with previous research, there are nonetheless limitations to highlight. In particular, our analysis has not addressed demographical and geographical influence on the results. In part, this is due to the lack of both demographical and individual-level information in the MSSD, and due to our data being based on Spotify, biasing the findings towards the population with access to the service. This means that our results are inherently biased towards Western culture, and we are unable to investigate factors such as age and occupation which have previously been found to impact listening behaviour [44,45]. We would encourage future research to work on combining datasets from multiple providers, such as QQ Music, Gaana and Boomplay, to ensure a wider geographical and cultural representation. Collating such datasets would require collaboration with the music streaming industry and work on harmonizing the many approaches to calculating musically meaningful audio features [46,47]. In addition, the audio features may miss out on nuances in high-level understanding of musical behaviour such as the behavioural functions of the music, and aspects of emotional content [48,49].

As a final note, we would highlight that this project has been carried out using open-source software and publicly available data, with all analysis and programming performed on laptop computers, and that the data collection processes in this work were undertaken without incurring any direct costs. This shows how the availability of digital traces from online activity can be used to investigate human behaviour by scientists both affiliated and independent alike [50].