Behind the scenes, Spotify is working hard to find you new songs that it knows you'll love. But how do its smart song algorithms work? And what's coming next from the streaming service?
With over 20 million songs in its library, Spotify – one of our entertainment partners for Vodafone 4G – presents a pretty unique problem: too much choice. You can never have too much of a good thing, but even so, the company is on a mission make sifting through its massive collection even easier.
This month on 4G
Having difficulty choosing your Vodafone 4G content partner? Take a look at what’s coming up this month on Spotify, Sky Sports and Netflix to help you make your decision…
– Click here for the full rundown
How? With a combination of some super smart computing, expert human input and a smattering of behavioural data. We’ve been speaking with Gary Liu, the head of Spotify Labs, to find out where the future of the service’s streaming suggestions lie…
Left and right brain working together
“The hardest part of having all the world’s music at your fingertips is that you don’t immediately know what to listen to,” says Gary. “So, fundamentally, we want to solve that problem – to maximise the amount of time people have to listen to new music.”
To do that, Spotify’s made a smart move, and bought very a smart computer…
“In March this year we acquired a company called The Echo Nest,” Gary tells us. “The Echo Nest is a data company based out of Boston, and it’s considered the top data company in music. Most people won’t have heard of them, but in the industry we’ve spent a lot of time discussing them because they do a really outstanding job.”
As Gary explains, The Echo Nest is a smart system for cataloguing music, which is built on two separate programmes working together:
“The company was founded by two MIT PhD holders, each of which built half of a computerised ‘brain’. One half is a program that can analyse any piece of music in under three seconds, and break it down into five dozen attributes and components – including some things that even a human ear can’t hear. It goes way beyond just BPM, key, and genre; we’re talking about things like the timbre of the voice, and the structure of the song. So you can literally start grouping music together based on these attributes.”
“The co-founder built the second half of this ‘brain’. It crawls the web every single day collecting billions of points of interest, and is able to understand the sentiment around certain artists and songs.
“The reason why both of those sides of the brain are really important is that if you take a death metal song, and you take a Christian metal song, and you throw them into the side of the brain that analyses attributes, they will look very, very similar and we would probably group them together. But they have completely different context,” Gary says, “so people who listen to them are probably very different in the way that they live their lives. You need that right side of the brain to tell the full story.
“We bought The Echo Nest because our goal is to be able to surface the right song, at the right moment, for the right person, without the consumer having to do anything.”
Getting to know you
That goal of feeding you the right song for every moment is one that Spotify is putting a lot of hours into achieving. And as Gary reveals, with the combination of human input, The Echo Nest’s smart brain and a strong focus on monitoring listeners’ behaviour and data, it’s a goal that’s within reach:
“With a machine like The Echo Nest we’re starting to test a lot of really interesting features. For instance, if we know you’re about to go out on a run, and we know a little bit of your listening history, we should be able to give you a very unique running playlist that will help you get the best results. A lot of these features will try to help the user discover new music without doing much themselves, and there’s iterations of that already in Spotify. The Radio algorithm has already improved because of The Echo Nest’s data.
“If we know you’re about to go out on a run, we should be able to give you a very unique running playlist.”
“Secondarily, we have our Browse section, where we have curated playlists by music experts within Spotify. The Echo Nest has really supercharged that, too. Our human curators are now not only using their own expertise to recommend music, but are also using this machine to surface songs from our deep catalogue that even they themselves might not have known about.
“They can listen to these and decide whether to include them in our themed playlists, making the Browse section an incredible combination of human and machine-based recommendation.
“You need both human and machine input to do this, but the machine piece of the algorithm is incredibly important,” Gary adds. “And to get this right we need to look at all of the data – at the 40 million users who listen to music every single day. The average Spotify user listens to music about 110 minutes a day. That’s a lot of music listening; to have humans analyse that would take a lot of time. With a machine, we can listen to a million songs in about two days.”
And data, once analysed, also plays a bit part in shaping Spotify’s innate music cataloguing skills:
“We recently launched a new category in Browse, called ‘Focus’. That was based on our analysis of user behaviour,” says Gary, “and of user-generated playlists. There was a massive group of playlists that people had made for studying, revising or working, and we saw that there were dedicated parts of the day in which people would turn them on and dive into their spreadsheets or books. So we created that new ‘Focus’ section, and it’s one of our top used sections in Browse – it’s more popular than Rock.
“We would probably never have created that section if we hadn’t looked at user data like that.”
“You’ll start seeing a lot more intelligence in Spotify…”
Gary explains that while The Echo Nest’s influence is already starting to make a difference to Spotify’s song suggestions, that magic combination of experts, intelligent software and user data will soon help the service’s recommendations hit a whole new level in the months to come:
“You’ll start seeing a lot more intelligence in Spotify,” he says. “You’ll see this creep up in ways that might not be immediately obvious, but all of our programmed features like Radio will start to become better at understanding who you are.
“And at the same time, we do expect to launch brand new features. We want to get to the point where you can just press play, and it’ll give you the right music for the moment you’re in.”