Following in Amazon's footsteps, two student projects independently use 'collaborative filtering' to bring recommendations and social networking to online music; soon they will join forces.
What we now know as the “social web” — or Web 2.0 — didn’t arrive until around 2004. But the first inklings of it were emerging a couple of years before. As usual, music was the harbinger.
Last.fm was founded in 2002 by a group of four Austrian and German students from Ravensbourne College of Design and Communication in London. It was fashioned as an internet radio station that allowed a user to build a listening profile and share it with others. The year of its launch, Last.fm won a young talent award at the Europrix, a multimedia awards show based in Vienna. This was how the product was described in a showcase video (embedded below) leading up to the awards ceremony:
“After repeated use, the system builds a listening profile that increasingly reflects the user's preferences. The sum of all profiles is visualized in the ‘Map of Music,’ a presentation of musical connections and genres determined only by the collaborative effort of Last.fm users.”
When the students went up to receive their award, one of them, Thomas Willomitzer, noted the importance of “collaborative filtering” to the Last.fm system. The idea was that the Last.fm algorithm would recommend music you might like, based on your listening history combined with the listening history of other, similar, users. Willomitzer added that this type of algorithm would be familiar to people who used Amazon.com.
Here's a video of the Last.fm founders presenting at Europrix 2002, via Thomas Willomitzer:
Collaborative filtering was a common technique in recommender systems, and its history dated back to before the web — for instance, it was the basis for a 1992 Xerox PARC email system called ‘Tapestry.’ But collaborative filtering really came into its own during the web era, and in particular it was popularised by Amazon. By 2002, Amazon users were familiar with the following message: “Customers who bought items in your Shopping Cart also bought…” There was also a “Your Recommendations” list on the Amazon.com homepage. Both of these features were created using an algorithm that Amazon called “item-to-item collaborative filtering.” As explained in a research paper:
“Rather than matching the user to similar customers, item-to-item collaborative filtering matches each of the user’s purchased and rated items to similar items, then combines those similar items into a recommendation list.”
The key here is that Amazon’s collaborative filtering was done based on the items people bought or rated, not the profiles of its users. This approach was also crucial to how new social web services like Last.fm would develop. The “map of music” that Last.fm created was all about mapping which songs (or genres) were interconnected — so a certain Bob Dylan song might have a strong connection to a certain Joni Mitchell song, based on listener data, and thus the Mitchell song might come up as a recommendation for people who listened to the Dylan song (and vice versa).
Audioscrobbler
By coincidence, another student in the UK was also working on a recommendation system for music in 2002. Audioscrobbler was started as a computer science project by Richard Jones at the University of Southampton. Jones coined the term “audioscrobbling" (later shortened to “scrobbling”) to describe the process of tracking songs that you listen to in order to make a listening profile, which is then used for recommendations.
In an April 2003 interview with his University’s paper, twenty-year old Jones explained how Audioscrobbler worked:
“Users of the system need to download software on to their computer that monitors what artists they listen to. The data is then collated and a pattern emerges by way of a technique known as ‘collaborative filtering.’ The results are then recorded against a username and can be compared with the listening tastes of other members.”
Later, Jones would team up with the Ravensbourne College students and fold his project into Last.fm, but even in 2002 — when they were independent products — it is striking how similar the two systems were. Both used collaborative filtering to create song recommendations, and both aimed to create a kind of social network based around what users listened to.
Escaping the Broadcast Model
The key to the emerging social web would be that you discover new content and communities by following other people. For music, the idea was to help you break away from the established broadcast model. At the Europrix event, Last.fm’s Martin Stiksel brought out a 1980s-style transistor radio to illustrate the point. If you want to listen to music on such a device, Stiksel explained, you have to tune the frequency band to find your station. If you don’t like the music playing on that station, you tune the dial to another radio station and try your luck again.
“The inherent problem with broadcast media is that basically, at the end of the day, it's always somebody else selecting the music for you,” said Stiksel. “So there's always a bunch of editors or programmers that picked the music and put them into into a program for you.”
With Last.fm, the stream of music you heard was a mix of manual choice and algorithmic selection. You might start with a song already in your online “record collection” (the term Stiksel kept using), or start from another user’s profile. From then on, songs would be chosen for you based on collaborative filtering. If you played a song through, the Last.fm software automatically added it to your own collection. You could also press a “love” button to add it. But if you didn’t like a certain track, you could press a “hate” button (so it wouldn’t get played again), or click the “skip” button to move to the next song. There was also a “change” button to go to a different user profile.
The early Last.fm user interface was, in truth, a bit cluttered with all these different buttons and various search boxes — but over time it would get more streamlined.
Stiksel explained that the idea for Last.fm came about when the students asked themselves, “how do you look for something that you don't know?” So in the case of music, how to discover new music when you don’t necessarily know what type of music you’re looking for? The answer, he said, was the social component.
“Then we figured out that it's the social aspect of music — the best music you always find when you go to your friend's house and he plays you records. And we’re taking this concept into an online environment here.”
Value of User Data
What both Last.fm and Audioscrobbler stumbled onto in 2002 was the collective value of user data in discovering new content — something that Amazon was also taking advantage of at this time. The problem with music, though, was that licensing from record companies was still highly restrictive. The Last.fm founders somewhat glossed over it during their Europrix presentation, but they did admit that “due to legal issues, we're only allowed to play 30 second samples.” Unless you already owned a piece of music, 30 seconds was all you got.
By the following year, however, Last.fm had begun turning itself into an "online radio" service, by paying licensing fees to the UK collecting societies PRS (Performing Right Society) and MCPS (Mechanical-Copyright Protection Society).
So pre-Web 2.0, the streaming revolution was only just getting started. But with Last.fm and Audioscrobbler, we at least glimpsed the future of the social web.
Buy the Book
My Web 2.0 memoir, Bubble Blog: From Outsider to Insider in Silicon Valley's Web 2.0 Revolution, is now available to purchase:
Or search for "Bubble Blog MacManus" on your local online bookstore.