How Contextualized Music Recommendations Are Shaping The Future of Music Consumption
Level up your digital marketing at: http://joshdviner.com/
I’m on a run. Inching towards my last kilometre. My heart rate is climbing. I’m dripping in sweat, trying my best to keep my pace. And then I hear a voice through my headphones say “1KM left, push it” and one of my favourite high-tempo songs starts playing. I do as I’m told, push it, and finish the run.
The above never actually happened. Yet, there’s no reason why it can’t sometime soon. This article will explore how Digital Service Providers (DSPs) ( i.e., Spotify, Apple Music, Amazon Music, etc.) can make more effective recommendations by leveraging personal taste as well as contextual information.
First, this article will define what contextual music recommendations are and review how some DSPs have added contextualization and personalization into their music services. Second, it will explore how these services can enhance their personalization and recommendation efforts. Finally, it’ll conclude with a glimpse into the potential future of music recommendations.
What Are Contextual-Based Music Recommendations?
As described by an article in Computer Science Review, the basic idea behind contextual (or situational) music recommendation and retrieval is “to retrieve and suggest music depending on the user’s actual situation, for instance emotional state, or any other contextual conditions that might influence the user’s perception of music.” In real life this equates to DSPs serving up music specific to a listener’s environment or emotional state. For example, if a DSP can discern that a user is about to go for a run, it can suggest a high-tempo running playlist; or if one is at their office (wherever that may be these days), the service can suggest a productivity playlist. Recommendations become even more powerful when DSPs can combine a user’s contextual information with their past listening behaviours, thereby serving up a playlist full of music one will enjoy, while also being suitable for one’s context. In the past number of years, DSPs have started along this path, but mainly only leveraging past listening behaviours.
The Beginnings of Music Contextualization and Personalization
Two years ago, music data analytics company, Chartmetric, wrote an article about the rise of contextual playlists on Spotify. In the article, Chartmetric found that context playlists (e.g., playlists based off an activity, time, or day of the week) have a higher median follower count compared to content playlists (e.g., based off typical genres). They also found that hybrid playlists (a mixture of context and content — e.g., “Bleeps & Bloops” is based off of Gaming (context) and Electronic/Dance (genre)) outperform both content and context playlists in terms of median followers as well as year-to-year growth percentage in followers. Hybrid playlists have a median follower count of 237K, compared to context playlists with 160K, and content with 103K. This difference in follower count helps demonstrate the power of the combination of contextual and content based recommendations. Chartmetric concludes that “music startups are moving more and more towards meeting the audience on their own terms.”
In 2019 Spotify announced that it will algorithmically personalize some of its curated playlists. This means that a typical playlist (such as the one titled “Happy Hits”) will not look the same for every user — it will be slightly personalized to one’s taste, based off past listening behaviour. Spotify claimed that seeking out songs after discovering them through a playlist is up 80% on personalized playlists and track saves increased by 66%, compared to non-personalized playlists. Once again, this insight shows the power of personalization combined with contextual data. For example, in the playlist “Happy Hits,” one is presumably listening to it for feel-good music; Spotify will personalize it to include feel-good music the service knows you love.
As will be a running theme with other DSPs as well, Spotify is continuously upgrading their playlist strategy in order to better contextualize and personalize their music recommendations.
A year ago Apple Music updated the “For You” section of Apple Music to provide subscribers with more customized suggestions and music recommendations based around certain themes. Playlists include titles such as “Start Your Week Right,” and “To Make You Smile.” Recommendations can be altered using the “Love” and “Dislike” features in Apple Music so that the music Apple serves up is continuously adjusted to cater to one’s taste. During the COVID-19 pandemic, Apple recently released a new personalized playlist called “Get Up! Mix,” which updates automatically every Monday with new music. The playlist features songs personalized to the user based off past listening, as well as humanly curated high energy favourites.
In contrast to Spotify, Apple Music has leaned more heavily on human curation; however, it has recently delved deeper into algorithmic curation.
In December 2018, Amazon added a playlist recommendation assistance feature to their popular Alexa voice-controlled device. When a user asks Alexa to help pick a playlist, the voice assistant engages in a conversation about what type of music the user wants to hear and subsequently starts offering samples of different playlists for the user to accept or reject. For example, it may ask if the user wants something more laid back or upbeat, what type of genre, what tempo, and much more. Additionally, one can “like” or “dislike” music via Alexa so that her recommendations when asked to “play something” become more aligned with one’s music preference.
Amazon’s differentiating factor is its popular smart speaker Echo devices. In fact, nearly 70% of smart speaker owners in the U.S own an Amazon device. Amazon has leveraged this competitive advantage in order to create a more conversational recommendation system.
In December 2019, YouTube Music rolled out its first personalized playlists, “Discover Mix,” “New Release Mix,” and “Your Mix.” “Discover Mix” introduces a user to new artists YouTube think they’ll like based off past listening; “New Release Mix” is filled with recent releases from ones favourite artists; and “Your Mix” is a more general playlist consisting of music a user has shown to enjoy. As with the other platforms, the more one interacts with the platform the better the playlists become.
YouTube Music is the most recent of the DSPs, taking over from Google Play. Google’s differentiator is its vast trove of search and watch data. Utilizing users’ YouTube data, Google has the potential to make extremely powerful recommendations.
In December 2019, Pandora released a redesigned mobile app that builds upon its personalization capabilities. The app features a new “For You” tab, which offers a personalized experience for each user, along with more station customization features and more. In terms of personalization, the new tab places a bigger focus on enabling the user to listen to music by genre, mood, activity, trending, new releases, and more. Pandora also created a new “Pandora Modes” feature which allows users to customize Pandora Stations, somewhat similar to how, as mentioned, Spotify personalizes some of their general playlists to each user (although in Pandora’s case, the user customizes it themselves). In a two-month test of the redesign, Pandora claimed that users engaged with personalized content they found in the “For You” tab three times more than the content they found through traditional browsing. Once again, the more personalized the recommendations, the more engagement these services tend to receive.
The oldest of the group, Pandora is known for its “Music Genome Project,” in which they have, for over a decade, used musicologists to study, analyze, and categorize 450 different musical details on every track. This incredible database enables Pandora to create effective playlists and personalized playlists.
The Future of Contextualized Recommendation Systems
As streaming services become bigger entities, accumulate more data, and advance their machine learning capabilities, music personalization will not only become about the “what” in music listening (i.e., the music) but also start to include contextual markers such as the “where,” “when,” and “how.”
Smartphones, with ubiquitous data and WiFi, have enabled music listening to take place in most areas of one’s life. Where one listens to music can play a role in the type of music they may want to consume. In 2014, Spotify and Uber announced a partnership where, if enabled by the driver of the car, a user can control the music in the car. In 2018, both Google Maps and (Google-owned) Waze added audio integrations that allow users to sign into their preferred DSP and control their music during a commute without leaving the navigation apps. Most recently, Google Maps added YouTube Music playback controls within the Google Maps app. These types of integrations can further stimulate personalized and convenient music listening. If DSPs know how long one’s commute is, perhaps they can suggest a playlist for the exact length of time; or if they know one is going to a bar, the playlist may be different than if they know one is going to work. Recommendations may differ if one is on vacation compared to when one is at work.
Some people enjoy listening to an upbeat playlist in the morning, some may prefer a calmer playlist. When one listens to music is a critical component of music personalization. Some DSPs already play with this contextual marker by serving up specific playlists at a certain time of day (e.g., Apple’s “Start Your Week Right” on Monday morning). Perhaps DSPs can go even further by better knowing user’s schedules. Does a user need a 3PM pick me up playlist? Or a 3PM nap time playlist? Serving up playlists, based off past listening behaviour, as well as personalized parts of the day, will make for better recommendations.
Between smart speakers, smartphones, and smartwatches, users listen to music across various devices. How one is listening to music can play an important role in the music they want to listen too. Especially during the current COVID-19 pandemic, many may be listening to music on shared speakers. Amazon Music is apparently working on a new feature, part of which will ask whether family and friends are around as a way to screen out explicit lyrics. In another example, if one is about to commute or travel without solid data networks available, perhaps the DSP can predict this and download certain playlists before one takes their trip. The Chinese streaming service Xiami performs a variation of this function by informing users when they are not on WiFi that they will incur data fees and offering one of their “traffic-free” plans (zero-rated or data-free). By considering how consumers are listening to music and on what devices is another way in which DSPs can continue to curate the most personalized, effective possible recommendation system.
Where Do We Go From Here?
The initial years of music personalization and contextualization have been about building contextual playlists and serving them at appropriate times via a DSPs home screen. Many of the major DSPs have multiple touchpoints for consumers that would enable them to cater music recommendations appropriately. Apple has the Apple Watch, iPhone, Mac, HomePod, and Apple TV. Google has YouTube, Google Home, Nest, Fitbit, and Fossil. Amazon has Alexa devices, Twitch, and Whole Foods. All of these touchpoints, especially the watches that provide biometric information, provide the companies with more data, which in turn will allow for better recommendations. Apple or Google may be able to use one’s heart rate, past listening behaviours and length of workouts, to serve up the perfect workout playlist. Amazon may be to integrate Amazon Music within the Whole Foods app — imagine walking through Whole Foods listening to music and while you walk around the store, a voice reads out items on your shopping list depending on your location within the store. As DSPs continue to personalize and contextualize their playlist strategy, it will be interesting to see the balance they strike between data privacy and convenience. In the meantime, I’ll be waiting for that moment when I get the perfect song to power me through that last kilometre of my run.
Clap 👏 👏 if you liked this article to help others find it
Comment 💬 if you want to discuss (or just say hello! 👋)