I’ve been using two different internet music streaming websites on and off for a while now – Last.fm and Pandora. They’re different, and complementary in a way – one of them relies on statistical analysis, the other on analysis of the music itself.
Last.fm, in essence, looks at what music you’ve listened to so far, finds other people who listen to the same sort of music, and suggests for you music that they listen to. It does a decent job of finding bands close to the ones you listen to – it’s how I found several of my favorite bands. But it does a rather bad job, I think, of diversifying – it won’t really suggest bands you might like that aren’t in almost the exact same genre as what you’re currently listening to. It doesn’t help that you can’t divide your music into different profiles – all of them are lumped together for suggestion purposes, even if, while you’re interested in genre A and genre C, but have no interest in genre B, which is between them.
The people at Pandora, on the other hand, actually sit down, listen to each song, and record information about it, and give you songs that are musically similar to the ones you have listened to already. This does a decent job of suggesting music you’d like, and I think manages to be more diverse in its suggestions (I’ve found several bands I find interesting and not exactly the same as the other stuff I listen to through it, though most of those I’ve never pursued). It helps that you can have different, unrelated “stations”, and in each of those explore a different sort of music. But, since the people at Pandora are working quickly, some of the songs are rather badly described, and since they’re doing the descriptions by hand they can’t get that many done. Last.fm has more music available, I think.
This dichotomy seems to apply to a few different things… just using statistics versus actually looking at what is being sorted. Movies, for example; a lot of movie services will do “suggestions” based on what other people listen to (a la Last.fm), but I read recently about a service that plans to be more like Pandora – actually watching the movie and describing it.
Or take automatic translation. My understanding is that currently, it’s done mostly with the Pandora method – it translates individual words using a dictionary and applies different preprogrammed grammatical rules. But they’re thinking that a way to get more accurate translations might be to use the Last.fm method – create a database of a bunch of documents that are considered well-translated (by humans, I guess), and then when automatically translating a document, take each phrase and find where it is used in one of the documents in the database and use that translation. The translating device doesn’t have to actually know what the words mean to do this.
After describing these things, I can’t help but think, not about how these two stack up against each other, but how they are both somewhat deficient. Neither of them is as good as just getting a person to do it. Actual music recommendations from an actual person are probably better than either Last.fm or Pandora; movie recommendations from someone you can trust (be it a friend, a professional critic, or a blog you read regularly) are better than the suggestions of a machine; getting a human to translate a document is surely more accurate than either automatic method. Both automated contextual and textual analysis are poor substitutes for an actual person doing it – their only advantages are in speed and
Not that those aren’t great advantages. I’d definitely recommend using Pandora and Last.fm, for as long as they’re both free. But neither can substitute for a person who is knowledgeable about music, if you know one. The same for music; the same for translations.