The secret formula behind Spotify and Tinder’s scary accurate recs

April 24, 20174 Minute Read

Select article text below to share directly to Twitter!

Dismiss

You may not believe in soul mates, but Spotify’s Discover Weekly playlist is pretty close to true love.

When digging into the UX and algorithms of today’s best apps, Spotify is a prime example of brands that really know their customers on an individual level. The Discover Weekly playlists, which refresh on a weekly basis, allow users to discover new music without skipping through irrelevant tracks. Using algorithms to generate recommendations isn’t a new concept, but this is different from your average genre-and-tone-based playlist built by a machine.

The depth of accuracy induced comedian Dave Horwitz to tweet: “It’s scary how well @Spotify Discover Weekly playlists know me. Like former-lover-who-lived-through-a-near-death experience-with-me well.” That level of custom experience is what we’re all striving for in design thinking. Seriously, what’s the Spotify secret, and how can you steal it?

Spotify goes a lot deeper with classification than many other apps. They’re also not shy about pushing the bar further than their competitors. Spotify’s Matthew Ogle told Quartz, “If you’re the smallest, strangest musician in the world, doing something that only 20 people in the world will dig, we can now find those 20 people and connect the dots between the artist and the listeners.”

Innovation where UX and algorithms meet

There are fundamental flaws in algorithms that aim to serve as a replacement for human selection and judgment. Algorithms are only as smart as their rules and input, and they don’t know how to filter out the garbage unless they’re told how. That’s why data science-powered diagnostic tools haven’t replaced doctors. Algorithms don’t operate with the nuance of people yet, but companies like Spotify and Tinder are trying as hard as they can to make this a reality.

Innovation in design occurs when organizations get to know their users better than anyone else and use this knowledge to build products. It’s the bleeding edge of design thinking, which is both informed and improved by data. The best customer-facing apps have excluded your kids’ favorite artists from your custom playlists, while connecting you to your new favorite singer-songwriter.

1. Spotify

Finding odd patterns in the relationships between product and customers isn’t new. Way back in 2004, Walmart discovered that strawberry Pop-Tart sales spiked when a hurricane was about to hit. And Spotify uses many of the same tools Amazon and Netflix use to generate recommendations, like clustering techniques and probability.

However, Spotify uses natural language processing on music blogs to understand cultural context around artists and live events—this is what differentiates their technique. They’re applying deep learning techniques to analyze the way their tracks sound. These insights form a rapidly growing number of specific subgenres, like new americana, alt country, or stomp and holler. These algorithms operate in real-time, creating playlists that speak to your soul. They’re also sensitive enough to pick up on the fact you’re really into “chamber pop” this week.

While Spotify’s engineers were a bit coy in a recent presentation about what’s next for their users, they plan on adding a feedback loop to their selection process. By integrating data on the songs that users skip and save from their Discover Weekly playlists, Spotify can only make your recommendations—and their super-sharp algorithm—better.

2. Tinder

As one of the world’s most popular dating apps, Tinder’s incredible ease of use makes high-volume online dating fun. There’re some hints of gamification; users’ swipes are interrupted by a micro-interaction when they receive a new match. Its instant feedback lets you know you’re likeable or, at the very least, swipe-able. Online dating can feel sketchy, but Tinder even makes room in their super-clean UI to show you mutual Facebook friends with potential matches.

Tinder isn’t just a really smart, addicting design. It’s also a data-based approach to romance that wants to know our behaviors perhaps better than we know ourselves. Every Tinder user has an “Elo score,” a complex number revealing where you fit into their community. Elo is based on real-time feedback from their user base. Tinder wants you to feel good, so you’re fed users with a similar Elo score who may have a similar personal style, sense of adventure, and other characteristics.

One of Tinder’s most recently launched algorithmic features is “smart photos,” or complicated testing of your profile pictures through a constant rotation to determine which are most attractive. Using a method called the Epsilon Greedy, it determines the swipe-attracting value of each photo on a scale of 0.1 to 1.0. Tinder’s engineers are convinced smart photos will land you more dates.

Tinder may not have cracked the whole code to algorithm-assisted compatibility and match potential, but they’ve come close. Is anyone else a bit frightened there’s a massive database containing almost all the answers to the science of human attraction out there?

Human preference is complicated

“Taste” is never simple, and one of the key characteristics of any algorithm that really gets its users is understanding how to filter out the bad data. It’s apps with enough nuance to understand that the yacht rock you played at your dinner party last week sounds nothing like what you listen to while jogging. While Tinder’s Elo score is a closely guarded secret, they’ve managed to assign numeric values to the zeitgeist of human attraction. Tomorrow’s user experience will be powered by algorithms that can discern between the data that’s “really you” and the data that needs to be discarded.

  • Recommended for you
  • Recommended for You