Spotify Knows What I like Before I Know I do
https://qz.com/571007/the-magic-that-makes-spotifys-discover-weekly-playlists-so-damn-good/
In class, we learned about how informational and direct-benefits reasons can drive decisions. Since we are all part of a larger group, society, and whole we interact and learn from others. We can infer from others’ actions some unknown information then we apply that in our own decision-making. Or perhaps, after knowing what our friends and peers have enacted in their lives, we too follow them out of peer pressure and desire to conform. Regardless, other people’s decisions inevitably end up affecting our own. We may like to believe that individuality and uniqueness, and passion define us. And they certainly do. But, at least in the music industry, our favorite album and songs may be more rational than we think.
Such is the case with Spotify which is a online streaming service that has in the last decade become extremely popular. One of the reasons that Spotify has distinguished itself from the countless other streaming and pay per song services that exist nowadays is that Spotify makes awesome playlists.
As a Spotify user myself, I love to listen to a playlist called “Discover Weekly”. It is a playlist Spotify assembles with its secret sauce algorithm. Every week I am strangely pleased and have never hated a song on that list. How does Spotify know me so well?
As shown in the article, the answer lies in networks, specifically we can think of how a song ended up in my playlist as a cascade threshold model. The process of curating the playlist is complex and not revealed, but the article hints at what is likely happening.
Spotify instead of looking only at me, looks at the others around me. Spotify analyzes what other users have personally added to their stations, playlists, queues, favorited songs, etc. By clustering individuals with similar music tastes defined by genre, artists, etc, Spotify recommends me songs that others in my cluster have liked. We can imagine there exists a q, such that if more than q fraction of users in my cluster have added that song, then the threshold is surpassed and the song is recommended to me. This explains why I never have certain artists or genres in my discover weekly playlist. I personally don’t listen to those kinds of songs, and there is a group of others like me that Spotify uses to gauge what songs I do like.
I like the basic stuff and admittingly do like the top 200 US songs. But I like to believe that I have a few preferences that make me unique because I like what I like. However, do I like the song because I like it, or do I like it because it is acceptable and familiar to me because Spotify said it is? Its just interesting to consider that a machine knows what I like before I known I even do.