Unexpected Discoveries in Randomized Music Playlists

Creating playlists can be a challenging task. The quality of a playlist depends on various factors, such as music genres or artists.

A number of studies have investigated the impact of different factors on users’ perception of playlist quality. For example, Pauws (2002) found that contextualized playlists were preferred over random ones.

1. Surprised by a song you didn’t know you liked

Music lovers can often identify with the experience of discovering a song on shuffle that they’d never heard before and immediately liked. Such songs might be a random song or a track that’s a friend’s recommendation. Moreover, there are also moments when music fans stumble upon a song by themselves, for example, when browsing through a playlist on Spotify or YouTube. These new experiences can lead to discovering a whole new musical world or expanding a playlist with more interesting and relevant tracks.

In our online questionnaire, participants were asked to use their usual music-listening device and start a new playlist with the shuffle function. When a track they had never heard of appeared on the list, they were asked to open and play this track and answer a few questions about it. In particular, they were asked to describe the first impression that the song evoked in them and why they chose to save this track to their device.

Participants were also asked to rate the quality of the resulting playlist. In particular, they were asked to evaluate the homogeneity of musical features such as tempo or energy, the diversity of artists, and the transitions between the ranked tracks. In contrast to previous studies, the lyrics aspect was rated by our participants as one of the most important criteria for a good music playlist.

Furthermore, we found that our participants prefer a good balance between tracks they know and like as well as tracks that are not familiar to them yet. This indicates that playlists containing recommendations should not only be characterized by the presence of already-known songs but should also try to include some less popular or undiscovered music.

Moreover, we found that the quality of playlists depends on the users’ expertise and enthusiasm. For instance, a playlist that promotes music that is new to the user tends to have high usage peaks but also has a shorter life span than a historical compilation or a context-oriented playlist.

Interestingly, we observed that a clear differentiation between the contexts and purposes for which a playlist was created significantly reduced the perceived difficulty of the creation task. This indicates that it is helpful for recommenders to provide a clear and specific definition of the topic in which they are trying to help users.

2. Unexpected transitions

Music fans often complain of “choice overload” when it comes to streaming music services. With so many choices, it is easy to get stuck in a musical rut of listening to the same songs over and over again. The endless options can also make creating a playlist that fits the current mood or occasion difficult. While creating a playlist is challenging for music lovers, technology can help make the task easier by suggesting tracks that fit the situation and mood.

One way to identify songs that are likely to be preferred is to measure harmonic surprise, a measure of deviation from expectations. Harmonic surprise can be measured using the chord type data provided by the McGill Billboard project, which features transcriptions of 732 Western popular music songs taken at random from the Billboard charts over a 34-year period.

We analyzed the chord type data to determine which chords were most surprising and then compared the time course of harmonic surprise for different song genres. The results showed that the time course of surprise tended to be convex and was more gradual at the transition from verses to choruses than at other points within the songs. Furthermore, the results suggest that harmonic surprise tends to be preferred when it is gradual and consistent rather than abrupt and sudden.

To test whether the genres had an effect on average per-song surprise and the average standard deviation of surprise across sections within songs, we conducted two ANOVAs. The first ANOVA yielded no significant interaction effect. However, the second did reveal a significant main effect of the quartile on both measures.

This finding is consistent with the Contrastive-Surprise Hypothesis, which suggests that listeners prefer to be surprised by music that deviates from their expectations and may be less tolerant of sudden surprises that are inconsistent with their previous expectations. Additionally, the trend toward a smoother time course of surprise suggests that listeners prefer gradual changes in mood rather than sudden shifts in musical style. Whether this preference for gradual change is due to the emotional impact of music or other reasons is unclear.

3. Surprised by a song you thought you didn’t like

Despite their rich musical collections, many people find it difficult to identify tracks that match a particular context or purpose. Therefore, they often wish to discover new music that might fit their needs. However, the search for appropriate songs is often time-consuming and requires that users remember a large number of tracks that they may not have heard in a long time.

To address this challenge, researchers have proposed different methods for assisting with playlist creation by automatically suggesting additional music from the same genre. For example, a recent approach (Bonnin and Jannach 2014) takes the artists appearing in the playlist beginning as input and recommends tracks from these artists and other similar artists. The artist’s similarity is computed based on various factors, such as popularity or co-occurrence in publicly shared playlists.

The evaluation of these approaches shows that they perform well in terms of typical information retrieval (IR) metrics, such as recall and accuracy. However, their effectiveness is limited by the amount of music that is available in users’ libraries, which can lead to an insufficient number of recommendations being presented to the user. Hence, other ways of identifying potential music matches are needed.

In order to understand what criteria are the most relevant in this context, we asked participants to rank the quality of playlists with respect to several aspects. Interestingly, the participants ranked homogeneity of track features and artist diversity as the top criteria. In contrast, the ordering of the tracks, their transitions, and the freshness of the tracks were rated lower.

Another aspect that was ranked high was the topic matching of the playlist. This finding is in line with the intuitive belief that the participants want to hear music that fits the intended purpose of their playlist, e.g., popular tracks for road trips, calm songs for chillout, or work-related melodies for the gym.

Furthermore, a correlation analysis showed that the participants who most frequently adopted recommendations were those who ranked the topics of their playlists highly. This indicates that it is crucial to consider the underlying motivation of playlist creation when providing recommendations. For instance, music services might benefit from offering users the means to easily share their playlists with friends or co-workers in the form of links. Then, these playlists might have the chance to reach larger audiences and thus increase their chances of being included in Discover Weekly.

4. Surprised by a song you thought you liked

For the dyed-in-the-wool music lover, the joys of discovering new music have always been an important part of enjoying listening to music. However, as the world of music discovery has evolved from crate-digging in physical collections to the more curated experience of playlist creation and streaming services, many have found that it is becoming harder and harder to find new songs they enjoy. This could be due to a combination of factors, such as the growing ubiquity of digital technology, which makes accessing their current collection easier and potentially leads them to repeatedly listen to the same songs.

Fortunately, there are still ways to find new music, including exploring the playlists of friends and acquaintances. Delving into a friend’s extensive physical collection can reveal unknown pleasures such as Martha Wainwright’s debut or Afrofunk, and expanding your musical horizons beyond English and American staples can open your ears to the likes of rap, K-pop, and alternative/indie.

In addition, there are online sources of music that can provide you with an infinitely scalable number of songs and artists to discover and love. These sites are able to generate recommendations that match your tastes and genre preferences, which is an excellent way to discover new music. However, it is worth considering that these recommendations can be very similar to the music you already know and enjoy and may not provide the new songs you’re looking for.

One way to address this issue is by incorporating the element of surprise in playlist construction. This is a great idea for anyone looking to avoid getting stuck in a rut of listening to the same old songs and is especially relevant to randomized music playlists.

Our research has aimed to assess the extent to which users adopt system-generated recommendations and how this affects their playlist creation behavior. For this purpose, we have conducted an online questionnaire that consists of three sections. In the first section, participants were asked to turn on the shuffle function within their regular music-listening app and play any track that came up. They were then asked to provide the track title, artist, and device/app that they used. Next, they were asked an open question about the track: “What’s the first thing that comes into your mind when you think of this song?”

Lastly, participants were asked to rate the quality of their created playlist based on various criteria. They were then asked if they liked their chosen tracks and why.

Read more:

Lucas Noah

Administrator

With a solid foundation in technology, backed by a BIT degree, Lucas Noah has carved a niche for himself in the world of content creation and digital storytelling. Currently lending his expertise to Creative Outrank LLC and Oceana Express LLC, Lucas has become a... Read more