On the corner of 500 S. and 200 E., rifling through the stacks of records while his bagel lay waiting next door, Jack Osmond searches for any familiar album kept stocked at Fountain Records. Living not too far away, Osmond comes into Fountain Records to peruse, wanting to expand on his and his roommates’ record collection.
“When we moved into our new house, the previous tenants left their record player,” Osmond said. “Since then, I’ve been trying to find some of my favorite albums to add to our collection.”
While in the store, Osmond pointed to some of his favorite albums that were on his wish list, “‘Yankee Hotel Foxtrot’ by Wilco for sure. I just got ‘Nevermind’ by Nirvana, and then ‘Californication’ by the Red Hot Chili Peppers.”
When we came up on the pile of new releases, however, they were met with just a passing glance. As a longtime Spotify user, Osmond said, “I usually will listen to my Daily Mix or Discover Weekly to find new songs.” As a Spotify user myself, I often find myself wondering how these old and obscure songs find their way into my Daylist or other Spotify-generated mixes.
Rediscovering lost sounds
Two thousand miles away in Brooklyn, representatives for Mexican Summer Records dig through old vinyl and singles, looking for those that time passed by. They search for, in their words, “pop culture heritage.” Songs and artists that encapsulate their moment in time perfectly, regardless of their commercial success.
Robert Lester Folsom, Art Lown and F.J. McMahon — all of whom released their work with limited to no success, were rediscovered and reissued by Mexican Summer. I was able to see Folsom perform live at Kilby Court in April, a man who put out his only studio record, “Music and Dreams,” in 1978 to no avail. With financial responsibilities mounting, he quit pursuing music full time and for 40 years painted houses in my native Jacksonville area.
However, since Mexican Summer found “Music and Dreams” in a small record shop in New York in 2010, he has been able to amass over five hundred thousand monthly listeners on Spotify. But how did someone coming from such obscurity find such success on these modern streaming platforms? The answer lies within their multi-faceted algorithms.
Inside Spotify’s algorithm
Spotify’s recommendation algorithm uses three main components to suggest songs to its users: collaborative filtering, natural language processing and audio analysis. Collaborative filtering uses massive matrices to look at users who have similar listening habits and tries to fill in the blanks. For example, if Person A and Person B both enjoy the same three albums, but Person A also listens to another, very similar album, the algorithm is likely to recommend songs of that similar album to Person B.
How does the Spotify algorithm know how these albums are similar? That’s where natural language processing comes into play. Natural language processing scans metadata throughout the internet to identify how songs are described and connected. Meanwhile, audio analysis breaks down the tracks themselves into a few main features — tempo, key, danceability and mood. All of these components work together to tailor personalized playlists and enhance user experience.
Whether it’s a programmer tweaking code in Spotify HQ or a University of Utah student flipping through albums at Fountain Records, discovery still starts with curiosity. Regardless of what algorithms may serve up, it’s up to the listener to decide to venture out and listen to something new.