🔬 Original article by Nils Aoun, Chloé Currie, Ava Harrington, and Cella Wardrop from Encode Justice Canada
This is a part of our Recess series in which university students from across Canada briefly explain key concepts in AI that young people should know about: specifically, what AI does, how it works, and what it means for you. The writers are members of Encode Justice Canada, a student-led advocacy organization dedicated to including Canadian youth in the essential conversations about the future of AI.
Introduction
Chloé Currie
In 2022, we live in an interconnected world where everyone can share the highlights of their life across social media. One of the applications most commonly shared is Spotify, the Swedish-based audio streaming service that houses a range of recorded music and podcasts for public consumption. Spotify is embedded within social media applications, such as Instagram and Twitter, to allow for easier sharing, making it accessible to a wide audience. Spotify curates your personal entertainment tastes to create playlists based on aspects ranging from your location to your current mood. Although this technology can be seen as interesting and harmless, it is important to understand how this algorithm works and affects your personal privacy.
Discover: How the Spotify Algorithm Works
Nils Aoun
Based on installs and active users, Spotify is currently the most popular music streaming application. [1] With various initiatives to attract and preserve users, it has made a name for itself in the music streaming market. Among these initiatives, Spotify Wrapped – which will be developed upon later in this brief – has intrigued many about the methods of data collection, and how such data is used. The answer to these questions lie in algorithms and machine learning.
What Users See – User Interface
Spotify’s home screen welcomes users with playlist recommendations such as their Discover Weekly, Recent Releases, as well as another section specifically custom-made for users. The latter includes playlists with names such as “Jump Back In, Recently Played, or Recommended for Today.” [2] These recommendations are the result of extensive data gathered and used by Spotify; the most prominent example of this being Spotify’s Year In Review where it provides users with data regarding their music consumption over the year.
Detailed Look at Data Collection
Spotify collects all data that is entered by the artists: songs names, description, lyrics, genres, images, and song files. In addition to these, the algorithm collects and tracks “provider side” data, like listening history, skipped songs, downloads, social interactions – in addition to external data coming from text data about songs or artists themselves. The statistics in the Spotify algorithm includes auditory history, emission rate, listening time and playlist features. Auditory history consists of identifying a song’s mood style and musical genre. The emission rate consists of transforming the least number of omissions into a greater number of recommendations. Listening time also helps determine how much a user enjoys a specific item; when exceeding 30 seconds in musical audio, it is considered positive data. Before then, the transmission is not monetized. This is referred to as the 30-second rule. [3]
Spotify’s Recommendation System
Similar to other online platforms, once data has been collected, recommender systems are used. In the case of LinkedIn, the recommender system delivers suggestions of people a user may know based on their current network, working history, and interests. Its role is to provide suggestions based on behavior or characteristics that have been tracked by the system. So, what about Spotify? With such an overwhelming quantity of content: “recommender systems are necessary to help navigate and facilitate the decision process.” [4] These systems achieve both collaborative-filtering and content-based recommendations. The former assumes that past behavior predicts future behavior – that people will like similar things that they have liked in the past. The latter relies on available features of the user-item relationship to build a mode, such as demographic data comparison. It also takes advantage of Natural Language Processing (NLP) and raw audio analysis. [5] NLP is software’s ability to understand spoken and written human language. It allows for the audio analysis that follows the collection of audio data from songs.
BaRT
The sections making up the home screen are created and curated by an AI called BaRT (“Bandits for Recommendations as Treatments”). Its goal is to keep listeners entertained and incentivized to keep listening to the music delivered to them on the platform; it is the real reason why it does not take long for someone to find a playlist they enjoy. [6] This algorithm provides users with new music that it thinks they will enjoy based on previous listening activities, to avoid them getting annoyed with redundancy. It is based on the two concepts: exploitation and exploration. [7] When exploiting, Spotify accounts for every activity its user engages in: listening history, songs skipped, playlists created, social media activity, and even location. When exploring, Spotify studies the rest of the world and searches for playlists and artists similar to one’s tastes, while also looking at popularity within one’s area. By analyzing the mood of the audio of a song (the language, lyrics, and contents) and then comparing them to other songs within Spotify’s repertoire, the streaming service can provide very accurate recommendations.
Result
The result of this entire process of data collection, analysis, and recommendation system is what users end up seeing in their Spotify home screen feeds. Spotify gives a customized experience to the user while considering their musical tastes, and overall activity on the platform and beyond, to provide them with an experience that will allow them to not only listen to but discover music that fits their taste. Many would say that this exchange of privacy for a customized experience is worth it; however most simply do not consider this trade-off. This is why it is important to know about the ways how such experience is provided – to make a clear decision while knowing the implications behind each of them.
Spotify Unwrapped: A Deep Dive into the Yearly Review
Cella Wardrop
At the end of each year, millions of Spotify users patiently await the release of their Spotify Wrapped, a collection of their yearly Spotify activity. Spotify Wrapped includes a user’s most listened to artists, songs, and genres, but has expanded over the years. Since the first version of Spotify Wrapped, Year in Music, was released in 2015, new features such as the user’s most listened to astrology signs have been introduced. [8] Spotify Wrapped has won multiple marketing and design awards, such as the 2020 Bronze Clio Award for Design in Direct Marketing, due to its marketing success with design and graphics. [9] However, from the algorithmic and data standpoint, Spotify Wrapped has received some criticism.
The Algorithm
Spotify’s algorithm works by collecting users’ data to provide features such as personalized playlists and song suggestions. Spotify’s algorithm can go as far to provide mood-based features, which many believe is the cause of Spotify’s success as a music streaming app, while others deem it “emotional surveillance.” [10] As AI is shown to be discriminatory, there are also concerns as to whether Spotify’s algorithms are biased against the promotion of music created by visible minorities. [11] As Spotify algorithms can feed users new content related to the content they enjoy, this process could reinforce the social biases of the user. [12] For example, if the user mostly listens to music by white artists, the algorithm may only suggest content by white artists, excluding similar music created by visible minorities. [13] The relationship between genres and race also plays a role in this bias, as many black artists have criticized their categorization into a ‘black’ genre, even if their music does not align with said genre (for example, Tyler the Creator in response to his 2020 Grammys Best Rap Album win). [14]
Self-Image
Critics have also pointed out the relationship between Spotify Wrapped and an individual’s self-image. [15] One’s Spotify Wrapped provides an insight into the user’s identity, and provides competition-inducing status rankings; for example, telling a user they’re in the ‘Top 1%’ of listeners for their favorite artist. [16] Sharing Spotify Wrapped on social media and with friends can induce a sense of solidarity with those with similar music taste and impact others’ perception of the user. [17] The ‘bandwagon effect’ is also at play, as users may feel the need to share their Spotify Wrapped and promote their favorite artists to seek feelings of validation and belonging. [18] When considering that Spotify Wrapped is rooted in Spotify’s algorithm and its potential social biases, the relationship between Spotify wrapped and individuals could be a concern.
Spotify’s Response
Spotify has been attentive to concerns raised over its algorithm’s capabilities. When country music star Martina McBride shared that her ‘country music’ playlist suggested mostly male artists in 2019, Spotify reached out to her for feedback on any potential negative biases that need to be corrected. [19] Spotify also describes its curated playlists as being user-based: if a user wishes to change their suggested content, they can “like/dislike” songs or diversify the content they listen to. [20] If a user wishes to be exposed to different artists, they can go to one of Spotify’s many public playlists. For example, “Black History is Now” is a Spotify collection of playlists and podcasts promoting Black voices across the world. [21]
On Repeat: The Effects of Spotify on Data Privacy
Ava Harrington
Spotify Wrapped has morphed into a kind of cultural phenomenon. The marketing strategy gives users the opportunity to assess their yearly listening history through colorful graphics and an interactive story format. [22] With features such as listening times, “audio auras”, and competitive artist statistics, Spotify Wrapped has made data collection fun with the social media presence of Wrapped strengthening this grip. [23] In 2020, Spotify saw more than 60 million social media Wrapped shares, but lurking under the bright displays are darker concerns surrounding data privacy and the future of tech and individualism. [24]
Capitalization of User Data
One concern with Spotify data collection is a general privacy issue. Specifically, that collected information can be used to understand more specific data, which can be sold and used for predatory and harmful marketing. The data collected by Spotify extends beyond one’s listening habits, including gender, IP information, and location. [25] Listening data can also be used to determine other aspects about our lives – our family status, our emotional status, or even our health information. [26] This is especially concerning given that AI like the kind used for Wrapped has been shown to perpetuate racism and other biases, information which could potentially be extracted. [28] Collected user information can also be used by Spotify for marketing and consumer trends. [29] By collecting and extrapolating personal information, Spotify is able to better determine market forces and identify better means of advertisement. Users effectively serve as a data source to fuel Spotify’s own commercial success. Spotify’s use of data may serve for fun, shareable graphics, but this collection can be harmful.
Personal Branding through Spotify
Another issue specific to Spotify Wrapped is its use of data and algorithms to dictate a kind of individual persona or “digital construction.” Through Wrapped, people feel a sense of identity based around personal data, encouraged by quips from Wrapped like: “If 2021 was a movie, you were the main character.” [30] This personal-data identity results in people basing their personality in data and (imperfect) algorithms. [31] The personal branding of Wrapped encourages users to share their data on social media to advance their social status or to connect with others about music. [32] This also serves as free advertising for Spotify, driving more listening and engagement on the platform.
Spotify Wrapped also informs users when they are in the top percentage of an artist’s listeners. [33] The “achievement” of being a top listener fulfills a personal individuality and also encourages competition – competition which drives more Spotify listening and more data collection. [34] An individual’s online activity and perception of self becomes an extension of oneself, which is then exploited for marketing and engagement.
The Price of Personal Privacy
Finally, Spotify Wrapped normalizes and encourages data collection and the invasion of personal privacy in the name of sharing an online identity. People interested in Spotify Wrapped have less incentive to push back against the harmful data collection it relies on, especially when considering the strong social presence of the service. [35] This may serve to decrease resistance to future privacy collection not only from Spotify, but also from other companies. Worryingly, young people already report a lack of concern with or even enthusiasm for Spotify’s invasion of privacy in the name of shareable stats. [36] Rather than working to hide data collection, media-savvy enterprises may find success in providing users with their own data habits, even at the heavy price of personal privacy.
Conclusion
Chloé Currie
The complexities of Spotify have grown immensely since its creation in 2006, specifically through introducing personalized mixes and creating a global audience. Through the popularization of multiple social media platforms, as well as the creation of specific advertisement campaigns such as Spotify Wrapped, Spotify has become a dominant music streaming campaign. Spotify collects a user’s data and inputs it into an algorithm to determine these personalized mixes and Spotify Wrapped. This algorithm acts as a type of “emotional surveillance” that concerns many users. It is important to understand how Spotify’s algorithm collects data and uses it to create a type of personal branding, but also to understand its implications on personal privacy.
Notes
[1] Top music & audio apps ranking – most popular apps in the United States. Similarweb. (n.d.). Retrieved March 15, 2022, from https://www.similarweb.com/apps/top/google/app-index/us/music-audio/top-free/AndroidPhone/.
[2] Marius, H. (2021, November 23). Uncovering How the Spotify Algorithm Works. Towards Data Science. Retrieved on March 1, 2022, from https://bit.ly/3CpMldD.
[3] Balaganur, S. (2020, January 26). How Spotify’s Algorithm Manages to Find Your Inner Groove. Analytics India Magazine. Retrieved on March 1, 2022, from https://bit.ly/3C8VY01.
[4] Marius, H. (2021, November 23). Uncovering How the Spotify Algorithm Works. Towards Data Science. Retrieved on March 1, 2022, from https://bit.ly/3CpMldD.
[5] Balaganur, S. (2020, January 26). How Spotify’s Algorithm Manages to Find Your Inner Groove. Analytics India Magazine. Retrieved on March 1, 2022, from https://bit.ly/3C8VY01.
[6] Ibid.
[7] Ibid.
[8] Cadenas, C. (2020, December 7). Spotify Wrapped: How its Stats & Features Have Changed Over the Years. ScreenRant. Retrieved March 7, 2022, from https://screenrant.com/spotify-wrapped-stats-features-evolution/.
[9] Spotify – 2020 wrapped. Clios. (2020). Retrieved March 7, 2022, from https://clios.com/awards/winner/design/spotify/2020-wrapped-94650.
[10] Ion, F. (2021, June 2). Spotify’s ‘Only You’ is a Glaring Reminder of its Emotional Surveillance. Gizmodo. Retrieved March 12, 2022, from https://gizmodo.com/spotify-s-only-you-is-a-glaring-reminder-of-its-emotion-1847020039.
[11] Illing, S. (2018, April 3). How search engines are making us more racist. Vox. Retrieved March 7, 2022, from https://www.vox.com/2018/4/3/17168256/google-racism-algorithms-technology.
[12] Hardt, M. (2014, September 26). How big data is unfair: Understanding unintended sources of unfairness in data driven decision making. Medium. Retrieved March 12, 2022, from https://medium.com/@mrtz/how-big-data-is-unfair-9aa544d739de.
[13] Carver, D. (2020, June 18). CT no. 48: Intentionally diversifying your algorithms. CT No. 48: Intentionally diversifying your algorithms. Retrieved March 12, 2022, from https://technologist.substack.com/p/ct-no-48-intentionally-diversifying?s=r.
[14] Harrison, E. (2020, January 27). Tyler, the Creator calls out the Grammys for racism in their awards categories: ‘The rap nomination was a backhanded compliment’. The Independent. Retrieved March 12, 2022, from https://www.independent.co.uk/arts-entertainment/music/news/grammys-2020-tyler-creator-best-rap-album-urban-racism-winners-a9303351.html.; Schaap, J., van der Waal, J., & de Koster, W. (2021, October 3). Black rap, White Rock: Non-declarative culture and the … Wiley Online Library. Retrieved March 12, 2022, from https://onlinelibrary.wiley.com/doi/10.1111/soin.12461.
[15] Pau, K. (2021, December 2). Spotify Wrapped, unwrapped. Vox. Retrieved March 7, 2022, from https://www.vox.com/culture/22814121/spotify-wrapped-2021-algorithm-data-privacy.
[16] Fomina, M., Kerins, T., MacIntosh, K., & Somerville, K. (n.d.). The behavioral science behind Spotify Wrapped’s viral success – the decision lab. The Decision Lab. Retrieved March 12, 2022, from https://thedecisionlab.com/insights/consumer-insights/the-behavioral-science-behind-spotify-wrappeds-viral-success.
[17] Ibid.
[18] Ibid.
[19] Reuter, A. (2019, September 16). Martina McBride ‘Felt Like We’d Been Erased’ When Spotify Didn’t Recommend a Single Female Country Artist. Billboard. Retrieved March 12, 2022, from https://www.billboard.com/pro/martina-mcbride-spotify-female-country-artists-interview/.
[20] Spotify. (2018, May 18). How Your Daily Mix “Just Gets You”. Spotify. Retrieved March 12, 2022, from https://newsroom.spotify.com/2018-05-18/how-your-daily-mix-just-gets-you/.
[21] Spotify. (2020, June 1). Spotify Stands With the Black Community in the Fight Against Racism and Injustice. Spotify. Retrieved March 12, 2022, from https://newsroom.spotify.com/2020-06-01/spotify-stands-with-the-black-community-in-the-fight-against-racism-and-injustice/.
[22] Pau, K. (2021, December 2). Spotify Wrapped, unwrapped. Vox. Retrieved March 7, 2022, from https://www.vox.com/culture/22814121/spotify-wrapped-2021-algorithm-data-privacy.
[23] Richardson, J. M. (2021, December 31). Richardson: Spotify wrapped makes invasive data collection look cool and hip. Retrieved March 18, 2022, from https://ottawacitizen.com/opinion/richardson-spotify-wrapped-makes-invasive-data-collection-look-cool-and-hip.
[24] Shalvoy, J. (2021, December 21). Spotify Unwrapped: Inside the Company’s Biggest Marketing Campaign. Variety. Retrieved March 18, 2022, from https://variety.com/2021/music/news/spotify-wrapped-marketing-shares-1235139981/.
[25] Pau, K. (2021, December 2). Spotify Wrapped, unwrapped. Vox. Retrieved March 7, 2022, from https://www.vox.com/culture/22814121/spotify-wrapped-2021-algorithm-data-privacy.
[26] Metz, R., & CNN Business. (2021, December 2). Spotify Wrapped shows how our personal data gets sliced and diced. CNN. Retrieved March 18, 2022, from https://www.cnn.com/2021/12/02/tech/spotify-wrapped-data/index.html.
[27] Burgess, M. (2021, August 7). All the Ways Spotify Tracks You—and How to Stop It. Wired. Retrieved March 19, 2022, from https://www.wired.com/story/spotify-tracking-how-to-stop-it/.
[28] Pau, K. (2021, December 2). Spotify Wrapped, unwrapped. Vox. Retrieved March 7, 2022, from https://www.vox.com/culture/22814121/spotify-wrapped-2021-algorithm-data-privacy.
[29] Burgess, M. (2021, August 7). All the Ways Spotify Tracks You—and How to Stop It. Wired. Retrieved March 19, 2022, from https://www.wired.com/story/spotify-tracking-how-to-stop-it/.
[30] Richardson, J. M. (2021, December 31). Richardson: Spotify wrapped makes invasive data collection look cool and hip. Retrieved March 18, 2022, from https://ottawacitizen.com/opinion/richardson-spotify-wrapped-makes-invasive-data-collection-look-cool-and-hip.
[31] Pau, K. (2021, December 2). Spotify Wrapped, unwrapped. Vox. Retrieved March 7, 2022, from https://www.vox.com/culture/22814121/spotify-wrapped-2021-algorithm-data-privacy.
[32] Metz, R., & CNN Business. (2021, December 2). Spotify Wrapped shows how our personal data gets sliced and diced. CNN. Retrieved March 18, 2022, from https://www.cnn.com/2021/12/02/tech/spotify-wrapped-data/index.html.
[33] Pau, K. (2021, December 2). Spotify Wrapped, unwrapped. Vox. Retrieved March 7, 2022, from https://www.vox.com/culture/22814121/spotify-wrapped-2021-algorithm-data-privacy.
[34] Ibid.
[35] Metz, R., & CNN Business. (2021, December 2). Spotify Wrapped shows how our personal data gets sliced and diced. CNN. Retrieved March 18, 2022, from https://www.cnn.com/2021/12/02/tech/spotify-wrapped-data/index.html.
[36] Richardson, J. M. (2021, December 31). Richardson: Spotify wrapped makes invasive data collection look cool and hip. Retrieved March 18, 2022, from https://ottawacitizen.com/opinion/richardson-spotify-wrapped-makes-invasive-data-collection-look-cool-and-hip.