YouTube’s Algorithm-Based Music Playlists Explained – Billboard
From June 2020 to June 2021, YouTube paid out more than $4 billion to the music industry, the company announced this month – a much higher sum from the world’s largest video platform where product reviews, how-to videos and vlogs share the spotlight with music. Rightsholders have always complained that YouTube isn’t paying enough for its offering, but 2020’s payout was a notable leap from 2019, when YouTube paid music rightsholders more than $3 billion. , thanks in part to its powerful algorithms that make music appear. every user wishes.
Unlike other streaming services, YouTube Music relies entirely on algorithms to program its 10,000+ playlists; So, instead of staff members manually selecting songs for listeners, the company emphasizes the editorial design of these algorithms with parameters that constantly evolve to meet listeners’ needs. (YouTube’s Music and Premium services have over 2 billion monthly active users and 30 million subscribers, respectively.) This “algotorial” approach – a combination of “algorithm” and “editorial” – allowed it to to maintain its position as world leader. -music platform used, despite increasing competition from Spotify, Apple and Amazon.
These algorithms work on the “signals” that music fans send to the service when they listen to music. These can range from obvious actions like liking or disliking a track and skipping it to passive cues like adding a song to their library or playlist and listening time. The algorithm ingests this information and adds it together to formulate the individualized options for each user – favorite playlists, related music and albums they may like. This basic process is standard across all music platforms, but the specifics of their formulas can lead to big differences in how users interact with different platforms – and the revenue they generate for the music industry. Now, for the first time, YouTube is opening the hood to its secret recipe. Here, YouTube executives unveil their “algalatory” curation method exclusively for Billboard.
Here’s how it works:
Raw signals are what we know – thumbs up or thumbs down on YouTube – but not all signals are created equal on the platform. For example, skipping a song isn’t as cumbersome as you might think, while other passive cues like saving a song to your library, repeating a song, and listening time of each song are key factors that help the algorithm build what YouTube Music calls your “taste profile.”
“A taste profile is made up of the listener’s favorite artists, the artists they’ve most recently engaged with, and it works down to the long tail of artists they’ve engaged with,” explains T. Jay Fowler, head of music at YouTube.
Like all music streaming services, YouTube Music begins building its taste profile when users initially sign up for the service and take a quiz (called a “taste builder”) that allows users to choose the artists they love. But it’s a little different for YouTube, where users watch more than a billion hours of video every day, searching for everything from music videos to home improvement instructions. “Most users are not new to YouTube,” says Gregor Dodson, YouTube Music Product Manager. “We often see Rick Astley there, or the Gangnam style,” Dodson says of new Music app users, noting that the company doesn’t care too much about your music video consumption, which may not translate to what you want to listen to on YouTube Music.
Fowler says YouTube is also careful about how it weighs the artists you choose in its tastemaker, noting that people can get stressed out by the onboarding process. “They say, ‘If I click on it and say I like Billy Joel, because I like that song or Fleetwood Mac, because I like that song ‘Dream’, will the service now to think I’m a big fan of Fleetwood Macs and who tops all the recommendations they give us?’ “Says Fowler. “So given that, we take it as a light touch and really consider their behavior more once we start packaging and recommending content for them, whether it’s ‘a mix, playlist or individual song recommendations.
In fact, YouTube says it never wants to assume that listeners are completely understood, recognizing, for example, that users who listen to songs on the radio and search for them on YouTube may no longer want to explore those artists’ catalogs.
“What you don’t want is this experience where you select Fleetwood Mac, and the very next screen is, here’s a bunch of Fleetwood Mac record shelves, mixes the playlist, where they’re featured and you say, ‘It doesn’t really have any value, it’s just that it spins what we told it to,’ Fowler says. “The idea is to couple this explicit input, this explicit selection of artists, as a simple rough guide. And then continue to encourage people to come out of this concentric circle.
Minimize skipping songs
Skipping a song also doesn’t carry much weight for YouTube Music, which executives say can be hard to decipher, as the motivation behind the skip may have little to do with whether a user doesn’t like it. not the song.
“One behavior that we talk about a lot is rage jumping, which is you see someone just jumping, jumping, jumping, jumping, jumping,” Dodson says. “So you think, ‘Oh, they’re annoyed. They’re clearly upset if they’re doing this. Well, it turns out that when talking to users, sometimes people do it because they want to hear what’s going on. there’s on the playlist. If you were to take that as a negative signal, you might actually end up delivering something or believing something that was seriously wrong.
Dodson says jumps are expected in things like YouTube’s Discovery Mix, which is designed to get people to listen to songs they don’t know. “A jump might not be now, or never.” Fowler says. “There’s a general rule you can follow with a jump, which is if you assume it’s not now, you can just delay how often you resurface it. We want to be relatively light with the not-never, because as a music fan, these new songs take a while to get to know each other and become your new favorite. So we never take a jump as an explicit, never play that again.
Handle unexpected events and gender lines with a human touch
“It’s really a mixture of human and machine because even the algorithms are written by music fans,” explains Doug Ford, director of music and product programming at YouTube, noting that human interaction is always crucial to creating a successful algorithm for music. “An algorithm can’t cold start a track or a new artist or a trend, so we’re here with other music fans and music experts in the industry, getting insights from partners, artists, direction, from the signals we see you, from search signals from our parent company and from YouTube consumption data.”
Fowler points to users seeking viral hits in TV shows and the boundaries between different genres – the lines of which are becoming more blurred with each passing year – are areas where the algorithms need additional guidance from from the YouTube Music team. Unconventional supergroups will also throw a spanner in the success rate of algorithms, Fowler says, as Labyrinth, Sia and Diplo’s LSD group did when they released their debut album in April 2019. “These three artists have incredibly large followers on YouTube, but the algorithm, when it saw this content coming in, it didn’t know what to do with it,” Fowler says, so the contextual awareness around the following groups had to be added by hand. “A lot of times you need a human being who can actually help provide information,” he says. “Let’s actually provide context to the system that’s important.”
Genres can also be a challenge for YouTube’s algorithm, and the company’s solution to the problem has been the cavalcade of subgenres you’ll see in year-end summaries with names like “pop rap.” or “deep pop emo”. “We use humans to train things like classifiers by creating groups of similar music,” Fowler explains. “Genre is incredibly subjective, and things that are labeled as Brazilian music aren’t necessarily okay if you can lump it all together. There’s Brazilian funk and there’s classical Brazilian music. We think it’s a human and machine marriage, but we allow the algorithms to talk to the user once we’ve given them the full context.
To determine the success of a playlist, Fowler says, “The first metric we look at is the number of impressions and the number of click-throughs. Did we put the right playlist in front of you that caught your eye? This analysis comes down to the taste profile (is it the right playlist for the listener), whether the user is able to understand what the playlist offers through the artwork and title, and how the user consumes the playlist.
“Once you click, we’re looking at what we call super consumption,” Fowler continues. “Have you spent any time with it, and had a long listening session? The jumps are fine. We don’t penalize you for skipping, but have you listened to a bunch of songs and are you getting value from them? »
YouTube also takes into account “tertiary metrics”, including nudging songs and listening to or returning to the playlist, which YouTube calls re-consumption. “That’s how we really know we have a good property,” Fowler says.
There are still some improvements the YouTube Music team would like to make, including developing their algorithm to better understand and predict how a user’s musical tastes change over time.
“We’ll probably struggle with this until the end of time, especially since this next generation is really genre, really period,” says Fowler. “They have their TikTok songs that they know all the words to, and then they have the real songs that they love, and there will be an intersection of those two. This will be an ongoing and evolving issue, but it’s fun.