Everybody likes contributing an opinion, especially in subjects on which we consider ourselves experts. Rank'em (located at gorankem.com), now in beta testing, uses this human urge as the basis for a crowdsourced song recommendation engine.
This obviously wouldn't work without some structure--asking users to pick their 10 favorite songs at random, for instance, would be too scattershot and yield useless results. So Rank'em has users select particular artists, then asks them to choose between 5 and 20 songs and rank them in order. Users must also rate their level of enthusiasm--or "fanstanding" as the site calls it--for each artist. These self-ratings are limited by the number of songs picked. If you only know a handful of Modest Mouse songs, Rank'em will assume you can't be a very big fan, and will only let you rate yourself on a scale of one to five. If you pick 20 songs, they'll let you give yourself up to 10 points.
When other users search for that artist's name, they'll see aggregate rankings of the songs, weighted by each voter's fanstanding, along with links to buy them on iTunes, Amazon, and eMusic (assuming they're available through these sources). Although the site is still in beta, so far it looks pretty promising: for example, the results for Jimi Hendrix place well-deserved fan favorites like "Spanish Castle Magic" and "Voodoo Child (Slight Return)" above overplayed radio hits like "Purple Haze" and "Foxy Lady."
You can try it yourself: enter the code "CNET" when you register (you'll also need a valid e-mail address) and you can start ranking songs and reading aggregate rankings in a few minutes.
Unfortunately, Rank'em is only as good as the song data it offers, and that data's flawed right now. The site draws song data from the user-maintained MusicBrainz database, but MusicBrainz is obsessively complete, listing multiple releases for the same album. This completeness is a drawback when you're just trying to find a song and rank it. For instance, the Who--notorious for repackaging songs and albums--has some songs show up five times. This can really mess up the rankings, as votes are split among multiple versions of the same song.
There are other data-related problems as well: albums often have the wrong years associated with them and some albums are listed twice with two different years. Rank'em has neglected to import data from some singles and soundtrack albums, which means several of my favorite songs--"Hey Jude," "Hey Bulldog," and "It's All Too Much" by The Beatles, "Positively Fourth Street" by Dylan--are completely missing from the site. And many artists operate with different bands--Neil Young solo, Neil Young & Crazy Horse, and Neil Young & the Shocking Pinks, for instance. Rank'em lists these as separate artists, but casual fans just want to know which Neil Young songs are the best.
For the time being, rating songs on Rank'em is a fun diversion, and the site could be a useful starting point if you're just discovering a new artist. But the company will have to do some pretty heavy manual data scrubbing--or find a more appropriate data source--before it can become a truly great recommendation engine.