: Use of undefined constant user_level - assumed 'user_level' (this will throw an Error in a future version of PHP) in /home/commons/public_html/wp-content/plugins/ultimate-google-analytics/ultimate_ga.php
on line 524
By Huazhi Qin and Kevin Ackermann
Sociotechnical Perspective of Spotify
Spotify as a social network
Generally speaking, Spotify has already built up the most important part of a social network: people. They owned a hundred and forty million users. More than 50 million of them pay for its premium service. In addition, it makes a connection with Facebook and Twitter. It allows its users to integrate their account with their existing Facebook and Twitter accounts. Then they are able to access their friends’ or followees’ favorite music and playlists. They can like or listen to the song that their friend was just listening to — making it as if they are with their friends. In other words, their existing ties in other social media platforms can be transferred to and pushed deeper in Spotify.
The function including “following” and “sharing” make the connections among its users, and make it go beyond just a music player.
Spotify as a new marketing channel
Spotify owns a large user base and database, which is the solid foundation for its ad experience. According to Spotify, it has 100% logged-in audience and more than 2 hours a day for multiplatform users. Also, Spotify collects billions of data points every day. The data they collect reflects the real people behind the devices, revealing users’ preferences, behaviors, and mindsets. As what Spotify emphasize, “the more the user stream, the more they learn”. Its streaming intelligence provides a new marketing channel, helping its customers to locate and reach the right audiences in the right context.
Spotify as a representation of the new music production and consumption activities
Spotify is one of the most popular streaming music platforms. It can be seen as a great example to elaborate the change the widespread of streaming technology bring to music production and consumption activity. The introduction of streaming technology addresses the conflicts between the users’ “objective” or needs and other elements in human activities, including tools, rules, community and division of labor. (Adamides) Streaming technology makes music more “playful, short-term, social, visual and mobile”.
How Does Spotify’s Recommendation System Create the Discover Weekly Playlist
Spotify’s Discover Weekly playlist, a weekly, personalized mix that is meant to help you “enjoy new discoveries and deep cuts chosen just for you,” works so well, it’s hard to believe that each Spotify account doesn’t actually come with its personal DJ. The song recommendations in the Discover Weekly playlist feel like they come from a close friend who knows you deeply, but in reality, this close, personal friend was created from a combination of several pre-existing filtering and analysis methods (Cowan).
The Discover Weekly playlist is fresh and new every Monday.
While recommendations on Spotify feel entirely unique, there was a long history of computationally-mediated recommendation systems from which Spotify borrowed and combined to make its Discover Weekly playlist. Borrowing a system of recommendation from Pinterest, Spotify originally tailored the design of its recommendation to mimic Pinterests with cards and panels that the user could interact with (Cowan).
This was the original design for Spotify’s recommendation system. The design drew inspiration from similar content recommendation systems, like Pinterest.
So, how was this technology designed to replicate a feeling of personal knowledge about a user to provide such intimate recommendations? Spotify combined three filtering/analysis technologies to create the Discover Weekly playlist: Collaborative filtering, natural language processing, and audio models (Ciocca).
Sharing music isn’t new or unique to technological mediums. Have you ever thought of how you determine if you should take someone’s suggestions to heart or not? You probably have an idea of if someone has “good” or “bad” taste, and you decide whether or not to listen to their suggestions based on the worthiness of what you determine their taste to be. You probably use shared interests and preferences to determine this worthiness.
In this scenario, Person 2 would leave with the suggestion of “Song A,” and Person 1 would leave with the suggestion of “Song E.”
Spotify realized they did not have to redesign this core action of recommendation in which people with similar tastes will like other things that the other person likes. Spotify just had to design a technologically-mediated version of this process that fit within a music streaming service. Netflix already popularized this process of collaborative filtering with its rating system, and Spotify took this system of modeling user behavior and comparing it with other users’ data to suggest new content and made the ratings into implicit data within the streaming service, such as play count. Spotify then creates a massive data matrix, in which every user of the platform is a column and every song is a row. An algorithm compares this absurdly massive matrix of data to find similar listening patterns among users. Collaborative filtering is now often viewed as the “starting point” for making a suggestion system (Ciocca).
Imagine a chart like this, but with literally millions more columns and rows.
Natural Language Processing
Another layer of Spotify’s recommendation system uses natural language processing to determine commonalities between songs and artists. Spotify has crawlers that search out what people and organizations are writing about certain songs on the internet. Once the sentiment of the song is analyzed and turned into a mathematical representation, the data is compared to other songs to find similarities among a user’s listening patterns (“Ever Wonder How Spotify Discover Weekly Works? Data Science.”)
Audio Models and Convolutional Neural Networks
The above methods are great for artists and songs that already have a large base of listeners, but how does the Discover Weekly playlist serve undiscovered hits to the user so regularly? Using convolutional neural networks, Spotify analyzes audio data from songs to determine certain characteristics such as tone, tempo or mood. Then using this extracted data, Spotify can compare and discover new, fresh songs and artists that sounds similar to artists and songs that a user already loves (Ciocca).
Adamides, Emmanuel. (2018). Activity-based analysis of socio-technical systems innovations.
“Audiences: You Are What You Stream.” Spotify For Brands, spotifyforbrands.com/is/audiences/.
Ciocca, Sophia. “How Does Spotify Know You So Well? – Member Feature Stories – Medium.” Medium.com, Medium, 21 June 2018, medium.com/s/story/spotifys-discover-weekly-how-machine-learning-finds-your-new-music-19a41ab76efe.
Cowan, Matt. “How Spotify Chooses What Makes It onto Your Discover Weekly Playlist.” WIRED, WIRED UK, 27 Jan. 2017, www.wired.co.uk/article/tastemakers-spotify-edward-newett.
“Ever Wonder How Spotify Discover Weekly Works? Data Science.” Galvanize Blog, 22 Aug. 2016, blog.galvanize.com/spotify-discover-weekly-data-science/.
“Spotify: The New Social Network.” Campaign Creators, www.campaigncreators.com/blog/spotify-the-social-network/.