Using social networks to find and evaluate online information
Online subscription music services are developing approaches that find useful and relevant information through social networks.
By Gillian Kerr, RealWorld Systems
The information in this article is current as of September 7, 2005.
One of the biggest problems on the Web is locating high quality information that is relevant to your specific needs. Search engines have been grappling with this since soon after the Web was first invented, and the most successful engines use strategies based on some sort of human filtering. For example, when you enter a search term into Google, you will get sites that other sites link to. When you search Yahoo, you will be using directories built by expert surfers. (For anyone who's interested in the world of search engines, check out SearchEngineWatch.)
But to really get a handle on how human filtering can locate relevant and high quality material, you should look at the emerging music subscription sites. Music fanatics, many of whom are youth, can be thought of as 'lead users' in technology. They spend huge numbers of hours listening to music. They endlessly discuss the advantages and disadvantages of competing services with their peers, and they are highly demanding regarding price, quality and functionality. They are also a very large market with money to spend. As a result, watching the online music industry is like looking into the future for the rest of us.
I'm going to focus on music subscription services because this is where search functions get really interesting. These services play songs over the Internet like a radio, but with a twist - they allow listeners to create their own radio stations.
Right now, most of these new-generation services are only available in the U.S. because of music license restrictions, so I was only able to test Pandora and Yahoo's Canadian version of LAUNCHcast. Hopefully the other services will become available soon. For the benefit of U.S. readers, Napster, MusicMatch (recently bought by Yahoo) and Rhapsody also provide subscription music with a collection of about one million songs, almost all rock, pop, urban and the like.
***********
Two approaches to human filtering: Pandora and LAUNCHcast
The two services I tested this month are beautiful examples of the two major approaches to human filtering. The first service uses experts to classify music based on defined attributes, and the second service uses automatic filtering based on usage patterns.
Pandora, a new subscription music service, offers members up to 100 personal radio stations. Pandora has coded about 300,000 songs on almost 400 musicological attributes. Trained analysts take about 20-30 minutes per song to "capture the unique and magical musical identity of a song - everything from melody, harmony and rhythm, to instrumentation, orchestration, arrangement, lyrics, and of course the rich world of singing and vocal harmony." When you select one or more songs or artists, Pandora creates a playlist of similar songs from its "Music Genome" database. Every time you give feedback on a song by rating it as "I really like this" or "I don't like this", Pandora immediately changes the playlist of the station to incorporate your preferences. You can play stations that have been created by other members, and share your station with others. The service is free for the first 10 hours and after that costs $36/year.
If you're interested in finding more music like the kind of music you already listen to, it's a terrific service. If you want to learn more about a genre, it looks like a really helpful service because it defines similarity at the song level, not on the reputation of the artist or the genre the artist usually belongs to. And it's easy to use.
However, I found Pandora unsatisfying because it didn't seem to filter by quality. A crummy copy of a great song has many musicological similarities, but it's still crummy. I'm not looking for music that is similar to the music I already like (I could just keep playing my Elvis Costello records over and over again). I'm looking for recommendations from serious music-lovers who like the same kind of music I like, but who will suggest new songs that I would never have thought of, or wouldn't have expected to like.
So I'm looking for another definition of 'similar' that meets my needs. Which brings me to Yahoo.
Yahoo Music's LAUNCHcast is a free music service (its paid version is not yet available in Canada) that creates a personalized radio station matching your special tastes. Here's how it works:
- During initial setup, you select your favourite genres and artists from a list of popular artists. Genres include pop, rock, adult alternative, R&B, rap, country, electronic/dance, jazz, blues, latin and reggae. You can select the entire genre, or specific artists within each. For example, under 'Adult Alternative' you can choose any combination of Matchbox 20, Barenaked Ladies, R.E.M., Enya, Train, Goo Goo Dolls, Jewel, Tori Amos, Sarah McLachlan, Alanis Morissette, Coldplay, and Snow. If that's not your idea of a good time, you can add up to four additional artists and click on 'Create My Station'.
- Start playing your station. The songs will be played on your computer using Internet Explorer or the free Yahoo Music Engine music player (still buggy and slow - wait for it to improve), or on the new Yahoo Messenger with Voice instant messaging program (I recommend this option, but be sure to choose the 'custom installation' option and refuse to change your Internet settings. The program rudely changes your home page unless you tell it not to.) The music players will all offer you the option to rate each song, as well as the song's artist and album, on a five point scale from 'never play again' to 'can't get enough'.
- You can now refine your personal radio station to match your preferences. Each time you give a high rating, your station will be reprogrammed to play more songs by the same artist, more songs from that album, and/or more songs that are similar. When you give a low rating, your station will play fewer songs like that.
- LAUNCHcast defines similarity completely differently from Pandora. When you give an artist a top rating, your station will play more songs by other artists that are highly rated by other people who love that artist.
For example, yesterday my personal radio station played a song by Joan Armatrading. I really like her, so I rated her as a '4' ("can't get enough"), which automatically increases the number of her songs on my station. But it also adds songs that are liked by her other fans. By clicking on Armatrading's name (as her song is playing), I'm taken to her dedicated page on Yahoo Music. The 'Fans' link pulls up a list of 93 personal stations that have given Armatrading a top rating. When I gave a top rating to Armatrading, my personal radio station started selecting songs that tend to be highly rated by these other fans of Armatrading, and in turn their playlists are now influenced by mine. If I want a more personal set of recommendations, or if I want to get more deeply into any particular artist, I can listen to any of those 93 fan stations. If I really like one of the stations in particular, I can add it as an 'influencer' to my own radio station. That person's preferences will influence my own playlist. I can have as many influencers as I like, assuming I have signed up to the paid service. Or I can just listen to that person's station from now on instead of my own. Some of these fan sites are extraordinarily diverse and interesting; it's like having a personal mentor who shows me through her record collection.
Speaking of diversity, personal recommendations by a knowledgeable guide is the best way to expand musical horizons. If you only listen to songs that fans of Joni Mitchell tend to like, you're unlikely to bump into Outkast; it's a form of groupthink. Well-designed reputational systems like Yahoo's can allow you to use a variety of strategies to find relevant results. For future improvements, Yahoo should make it easier to locate influencers by their names or other characteristics. Wouldn't it be cool to subscribe to the personal station of your favourite musician?
- Most subscription services seem to allow users to share their playlists with friends. If you use Yahoo Messenger, you can set the options to show what song you are listening to when you are online. In fact, that's a growing problem with instant messaging applications; you have to be careful about your privacy preferences or you may reveal a whole lot more about yourself than you intend. But for youth (and music fanatics), music is a way to create and define community boundaries, and it's interesting how music services are contributing to this use.
To some extent, the creativity of the new music subscription services are driven by the weird rules that are incorporated into current music licenses. Services have to design clever workarounds that satisfy their demanding and bad-tempered customers, while following license restrictions so they can keep access to commercial music catalogues. Subscription services can't play a specific song on request; they have to play a "similar song". Users can't rewind to listen to a particular song, though it is allowable to skip to the next one. All of these rules are intended to discourage illegal copying (sigh), but they lead to ingenious solutions.
How is music like information, and what do these services mean for the future?
Music is a form of information, but a very complex form. "Good music" is hard to define, since it relates as much to a listener's personal taste, mood and context, as to its quality as rated by experts. If we can figure out successful search strategies for music, those strategies should be useful for plain old textual information as well.
Reputational systems (see this pdf document for definitions and examples) like Amazon.com's recommendation engine and Slashdot's 'karma points' are getting more popular for all kinds of uses, not just music. Expect to see increasingly sophisticated ways of searching for information, based on usage patterns and social networking approaches.
**********
Gillian Kerr, Ph.D., C.Psych.
President, RealWorld Systems
gkerr at realworldsystems.net
Read my weblog at http://blog.realworldsystems.net
