Facing the Music: How Systems Like Spotify Know What We’ll Like

Have you ever wondered who is behind music streaming services like Spotify and Pandora, and how they are able to predict what kind of songs you will like? That’s exactly what Professor Nick Seaver, an assistant professor in the Department of Anthropology, is interested in. As an anthropologist, Professor Seaver chooses to study the individual people who create and maintain the systems rather than the algorithms used to produce the recommendations. Particularly, he researches how the people behind music recommendation systems think about their work.

After studying literature and completing his masters in media study, in which he researched the history of the player piano (the self-playing piano), Professor Seaver realized his love for music and anthropology, leading him to apply for a PhD in music recommendation. Once accepted into the program, he found that the most difficult barrier of studying music recommendation was access. As an anthropologist, typically research is done by going out into the field and observing people. However, because of his focus on commercial algorithms, Professor Seaver had a difficult time conducting research due to non-disclosure agreements. As a result, much of his writing discusses the challenges of studying the music recommending systems. In order to learn about their systems, Professor Seaver had to interview people multiple times to slowly gain their trust. Another way he conducted research was through attending conferences, where conversations surrounding the algorithms are a little less guarded and he could more easily learn about them.

Part of why Professor Seaver is interested in the people behind the software rather than the actual algorithms themselves is because change occurs so rapidly in the code of these systems. By orienting his research on the more stable features of the system, such as who the people that work on the software are, Professor Seaver can be sure that his research will be more applicable in the long-term as well.

So how do music companies recommend music to listeners? The people behind these recommendations create categories based on certain predictions about their listeners to create the algorithms. These categories tend not to be demographic-based, but instead constructed on other factors such as the level of avidity – how much people are into music. Most of the software engineers identify themselves as musicophiles (lovers of music) and the imagined listener as someone who does not share the same level of passion. Depending on their assumptions, the algorithm will be built differently. For example, some algorithms bring in demographics like age, race, and gender for users who demonstrate low avidity, or they try and gain information about the user through their connected Facebook account to better recommend music. The less information on a listener, the less accurate the recommendations become. However, designers are aware that many listeners still have low expectations when experiencing the system’s recommended songs; this is helpful for software designers because there is more room to be experimental in the songs they recommend.

Professor Seaver is currently writing a book on his research into algorithmic music recommendation, and his next topic of interest is the anthropology of attention. He would like to research attention as a cultural and social concern that people value and is beginning to conduct interviews on the subject while also teaching at Tufts.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s