I looked at a number of the nominees and winners, then decided to explore Pandora, the Second Place winner in the "Music" category. I've been using First Place winner last.fm since last fall. I love last.fm as a place to discover new music as well as get reacquainted with music I loved years ago but lost track of. Since MTV many years ago stopped showing videos and became a repository of "reality" programming, I've had a hard time finding new music. A few years ago I got an iTunes account, and my husband got Sirius satellite radio installed in his car, and between those two, I was able to begin finding music I loved again. Unlike fm radio, Sirius had a truly diverse selection of music, and four or five stations that focused on genres I really liked. I'd frequently hear new (or new-to-me) songs I loved almost instantly, then go to i-Tunes to purchase them. (I missed out on Napster, etc, and never got the hang of breaking international copyright law by downloading mp3's from file-sharing sites (Press the play button in the upper right.)
In any event, last.fm is a wonderful additional channel through which to find new music. One drawback, however, is that, while it organizes music by user tags and genres, it doesn't seem to "learn" one's general musical tastes. You can press a button to say you "love" a particular song, and, using a paid membership ($3.00 a month) you can play a shuffled selection of the songs you love. You can "ban" a particular song. You can ask it to play you music labelled as belonging to a particular genre. You can ask it to play "Your Recommended Songs," but it seems that it simply bases this upon the genre of songs you listen to and love, and perhaps on existing correlations between people liking one artist and liking another. Many people who like Nine Inch Nails also like Rob Zombie, for instance. I like NIN, but not Zombie. Yet last.fm keeps on playing him when I ask it to play "recommended" music for me. It's already decided what tastes go with what other tastes. That is, last.fm doesn't seem to have a mechanism for dynamically anticipating a user's tastes.
Such capabilities appear to be very important to a number of online businesses, particularly Netflix. It has an ongoing contest to improve its collaborative filtering algorithm, to better anticipate what films a given viewer will like, based upon what films they have liked in the past. For an improvement of 10%, Netflix promises a million dollar reward. It might sound like a lot of money for such a "small" percentage, but Netflix has found that anticipation of tastes in movies gets extremely difficult past a certain point, or when certain movies enter into the mix. I read an article about this a while ago where Netflix gave the example of the movie "Napoleon Dynamite." Apparently, it's one of a select group of films that it is extraordinarily difficult to predict whether a specific person will like, even if one has many other ratings of other movies to base the prediction on. It seems that being able to win the Netflix prize would somehow involve the ability to predict the previously unpredictable.
Soooo...I'll test out Pandora some more. Pandora asks you to select a genre or artist you like, then plays other songs and artists it thinks you might like, giving you the opportunity to give a "yes" or "no" to each song. My primary goal here is simply to find a way to discover more music I enjoy, while minimizing the need to listen to lots of things I dislike while waiting for "the good stuff" to pop up. It's an intriguing thing to me, the mathematical, computerized prediction of what will touch a specific person's soul.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment