Yes, No, Maybe: Can You Repeat That Song?

image from On Pandora and Slacker, choices fall into three categories. Users may give a song thumbs up or love it, making it so that song plays more regularly on that station, or they may give it thumbs down or ban it, ensuring that they never hear that song again. On both services, users have the option of the skip, which, in effect, signals a mixed review—indifference, not loving nor hating—or not now, the response that the user does like the song, but isn't interested in hearing it at this moment. Soon, most fans will interact with music in interfaces that treat choice in this manner. The question worth asking is whether or not this is the ideal representation for choice and if there's a better way to characterize user taste. Is articulating preference for a song a simple yes or no question?

The alternative appears no better. Rating systems work well on Netflix, but aren't ideal for songs. Liking a song isn't as easily characterized in terms of a three or four-star experience, often times, either you like the tune playing or you don't.

In his book Program or Be Programmed, media critic Douglas Rushkoff asserts that, "The digital realm is biased towards choice, because everything must be expressed in a discrete, yes-or-no, symbolic language. This, in turn, often forces choices upon humans operating in the digital sphere." In order for Pandora and Slacker to interpret a user's intent, everything must be broken down into these decisions, because that's how databases work. They require discrete answers.

Many argue that this process fails take "growers" into account. Those venerable songs that are complicated or just aren't easy to like at first. It takes a little bit, maybe even several exposures, before fans have that "aha!" moment where they realize that they really do like the song. The very same one that they may have previously expressed their despisement for. Do Pandora and Slacker take these songs into account? It's hard to say. In a music world, where our choices define our experiences and the trajectory they take, there's no perfect answer to be had.

If Pandora or Slacker replay the song, they risk alienating their userbase. After all, they already indicated that didn't favor the song. On the web, everything is a choice, a preference to be expressed at any moment. Granted, while these interfaces aren't the only way that fans are exposed to music, it makes you challenge if being reexposed to certain songs, at different times of the day, or at a later date, would have the effect of changing user preference. Causing fans to give it thumbs up or love, where they previously indicated disinterest. Perhaps, there isn't a large enough swath of songs that cause a mixed reaction. After all, most pop songs today can't afford to be objectionable; they have to stick right away. There are few songs on the radio that require multiple listens to get, but in our history, there are many that need several listens to discern and love. Yes or no interfaces, it would seem, fail to consider the nature of these eccentric songs.

What do you think? Is the way Pandora and Slacker handle choice the ideal representation? Is there a better way to characterize user taste?

Share on:


  1. We have a new music search and discovery technology (Myna Music) that automates playlist generation based ONLY on musical qualities.
    Although this does not entirely solve a user’s choice problem, it helps significantly by playing music that is much more likely to be satisfying to a listener’s up-to-the-minute taste than playlists generated by metadata/consumer data.
    Kyle, you mentioned that it’s difficult for a Pandora to know when they should try a song again that the listener didn’t like once but might like now. Myna can help solve that problem.
    If a user was listening to, say, Lady Gaga and thumbs-downed a Dirty Projectors song, Myna could easily be told to bring that song back when that person was listening to something more similar (eg Animal Collective).
    It’s a really cool and completely unique technology that is going to change the hands-free listening experience over the next couple of years. Check it out!
    Thanks for the interesting read…cheers!

  2. Yes, it’s true that there are certain kinds of music that are great every time they play. But then, there are also those songs that you get sick of after a relatively low number of plays – and those that you can’t get enough of after only a certain number of plays.
    In my early years of music discovery in the 90s, I used to find the growers by going to the store and re-listening to albums I didn’t buy last time around.
    But if clicking it away means it doesn’t get anymore plays, it’s just gone. Sad.

  3. I think a five star system that Netflix uses is vastly superior to the simple thumbs up/thumbs down/next experience in Pandora. I use star ratings (along with other metadata) in my personal collection to generate playlists all the time. I think five levels is much better than the coarser-grained option from Pandora.

  4. Eh, I gotta agree with the article: the quality of a song isn’t easily represented in a five star rating system. Whether within iTunes, on Amazon, or wherever else…it’s too subjective. Beyond what rating two different people give the same song, it’s that one person would rate the same song differently depending on their mood, musical tastes at the time, etc.
    In other words, it’s not efficient to apply a 5-star rating to your music library, because you know in a year it won’t be accurate.

Comments are closed.