A few years ago I reconnected with an old classmate via Spotify. We had barely spoken to each other in real life until we found that we had eerily similar music taste while looking at each other’s playlists . And then we started talking on Facebook and became friends. While we both moved to Europe from the United States we started planning to go see gigs together and eventually met up in Zurich and spontaneously went to see a concert by a band called A Place to Bury the Strangers, which also happened to show up in my Spotify recommendation lists.
The merit of this story is not about how the internet connects people. The merit of this story is about how the algorithms used by Spotify played a crucial role in establishing the profile of “music tastes” during that time. I started listening to a few artists from a genre a I like, then I keep finding out about other artists I like who are associated with them from the Spotify weekly recommended list, then I started making playlists with all of them together, and the next thing I knew there was a clear pattern in the type of music I listen to just by looking my Spotify page.
But what exactly does ‘algorithm’ mean? In Tarleton Gillespie’s essay1, she dissects the term “algorithm” and outlines how the practical meaning of this word has grown far beyond its literal meaning. Due to the broad cultural impact of the word “algorithm”, the general public started to adopt a range of references that they associate with ‘algorithm’, whereas the social scientist is more critical of the impact of algorithm and adopted a variation of such references. Gillespie elaborates on the meaning of the word to different groups of audiences and emphasizes the importance of making such a distinction in the discourse without being tyrannical by insisting on its literal meaning as defined by the technical community – a set of logical steps. Popular websites have been using “algorithms” for many years now to recommend content based on a user’s taste profile. It’s perhaps one of the most prevalent “algorithms” in our everyday life. From the social scientists’ point of view, lots of issues and concerns have thus been raised regarding the popularity of these “algorithms” such as privacy, data security and internet culture becoming more and more like an echo chamber. The algorithm becomes simultaneously the mysterious thing that magically returns to you so much knowledge about you and the creepy thing that steps in the potentially uninvited zone of privacy, while in reality it merely comprises of a cluster of of logical steps.
In fact, unlike how the singular word “algorithm” is casually used in these contexts, any website that’s a platform of content display probably uses more than one algorithm. Spotify’s Discover Weekly, for example, uses three main types of algorithm models: Collaborative Filtering, Natural Language Processing and Audio.
Let’s focus on Collaborative Filtering. While the other two models analyze the content itself, Collaborative Filtering analyzes patterns of the profiles of users. The benefit of this model is that it’s so universal that it can essentially be used in all types of platforms that recommends content, rather than just music/video streaming websites. YouTube and last.fm use it, as well as Amazon and Facebook. There are several types of Collaborative Filtering Models, but what they all have in common is that they look at users who view the same content, and they cross reference their history and find out what these users have in common and group them together. In the case of Spotify, the recommendation model takes data from a huge set of user playlists and analyzes the patterns of what artists or types of music are generally put together, so when a user displays a pattern similar to others who also listen to these artists, items from the playlists of these users would be recommended to that user.
Returning to the story that I began with: I was connected with someone because 1)Spotify recognized that we were listening to the similar types of music; 2)Spotify categorized us in the same group of profile and thus recommends similar music to us; 3) the music that we listened to became more similar because they were also influenced by the algorithms. This becomes a feedback loop that propels the formation of not only groups in the virtual space (ie. as part of process of the algorithms) but also groups in real life because they are creating what people have in common.
What particularly interests me about Collaborative Filtering algorithms is how when we consume the content from the internet we are presented with what the algorithms believe we want to see, depending on what group the algorithms put us in. Without criticality our taste, our references and even values can be strengthened based on what we are already consuming.
1. Gillespie, Tarleton. “Algorithm [draft] [#digitalkeywords]”
2. Ciocca, Sophia. “How Does Sporify Know You So Well?”.