White Paper: How Your Favorite Music Discovery Services Work


Listen to the mind-numbingly repetitive radio programming on the FM dial long enough, no matter which genre you prefer, and you might conclude that only a handful of recording artists are worth listening to.

Fire up your PC and tune in to Internet radio, on the other hand, and you’ll discover an embarrassment of riches, nearly all of which you can enjoy for free and without—or at least with very little—commercial interruption. In fact, there’s so much music that you might find yourself overwhelmed. That’s where the music discovery services Last.fm, Pandora, and Slacker come in. All three services help you discover new music based on the songs and artists you express a preference for. As interesting as that concept is, what’s even more remarkable is that each service takes a completely different approach to the mission. Let’s take a look at all three.


Last.fm mixes Internet radio with aspects of social networking. The service started out as an Internet radio station that allowed listeners to express their preference or disdain for particular songs by using a Love or Ban button. Last.fm used this information to develop a unique profile for each user and to create dynamic playlists.

The service was later merged with the Audioscrobbler music-discovery system, a stand-alone application originally designed to record the music played on registered computers. This enabled Last.fm to collect statistics that could chart a track’s worldwide popularity. Each time you listen to a song—whether it be online or from your personal library—the tune gets “scrobbled,” meaning its title is sent to Last.fm and added to your music profile.

Last.fm analyzes your initial list of favorite artists (which you provide when you first sign up), your personal music collection, and your expressed preferences (based on your use of the Love and Ban buttons) and begins streaming songs it thinks you’ll enjoy. In that respect, it’s not terribly different from Pandora or Slacker; Last.fm becomes unique when you take its community aspect into consideration. Subscribers can join groups based on common interests, create friends lists, and view each other’s profiles. Profiles list tracks the person has recently listened to, songs in their library, as well as charts listing their top artists and tracks. The service also uses a collaborative filtering algorithm to compare your preferences with those of like-minded subscribers and build a personal recommendations page.


Pandora takes a more scientific approach to music discovery. The company’s founders started an initiative called the Music Genome Project in 2000, with the goal of analyzing the fundamental elements of a song, and Pandora uses this data to analyze the music you listen to on the service and then recommends other songs and artists you might enjoy.

Pandora’s musicologists go much further than simply breaking a song down to its hook, chorus, and bridge; following the basic tenets of music theory, they scrutinize each song for as many as 400 distinct musical characteristics. They identify attributes ranging from major/minor key tonality (whether a song’s harmony is based on a major or minor musical scale), level of syncopation (a rhythmic quality in which emphasis is placed on upbeats, versus the more conventional downbeat), and instrumentation (which types of musical instruments are featured in the song, including a distinction between electrified and acoustic instruments). You’ll find a more complete listing at Pandora's website .

Pandora has 50 musicologists on staff adding some 15,000 tracks to the Music Genome and Pandora databases each month. When you sign up for the service, you provide it with one of your favorite songs or artists and it will use an algorithm to pick other songs and artists from its database that it predicts you’ll also enjoy.

When we told Pandora we enjoyed folk artist Guy Clark, for instance, it began playing Kris Kristofferson’s “Pilgrims Progress.” Since we hadn’t given Kristofferson much thought since the 1998 Wesley Snipes vampire flick Blade, we clicked the “Why was this song selected?” button. Pandora replied, “Based on what you’ve told us so far, we’re playing this track because it features folk roots, country influences, gospel influences, a subtle use of vocal harmony, and acoustic sonority.”


Slacker’s name is somewhat ironic since it could be argued that this service is powered more by human effort than either Last.fm or Pandora. Like those two Internet radio stations, Slacker uses algorithms to analyze your expressed preferences and then recommend music it thinks you’ll enjoy, but Slacker is unique in that it hires professional deejays to program its stations (which is to say the deejays are choosing which songs are played on the radio stations, not that they’re writing the software that runs the show).

Slacker has 120 prefab stations, and each deejay is responsible for programming just one or two of them to ensure that the person populating the playlists is an expert in that genre. The deejays also monitor what users are listening to (or skipping, as the case may be) in order to track which songs are trending popular; they’ll then increase their rotation so that they’re played more often. But they also take care to avoid playing a particular song so much that listeners grow tired of hearing it.

Subscribers can also create custom stations based on their own musical tastes, using the familiar Love It/Ban It buttons. As you populate the station with your favorite artists, Slacker will recommend other artists in the same genre. You can fine-tune your Slacker stations with slider controls that boost or limit the degree to which the service’s recommendation algorithms expose you to new artists, play popular or more obscure tracks, and select primarily older classics or new releases.

Around the web