Slashdot Mirror

← Back to Stories (view on slashdot.org)

More iTunes Math

Posted by Zonk on from the fun-and-learning-like-pb-and-j dept.

markmcb writes "OmniNerd once again digs into the math behind iTunes. The 5-star ratings in iTunes are actually stored in a variable allowing values from 0 to 100, with 20 = 1-star, 40 = 2-star, etc. Known to few, if you set a rating to 30 (manually), it will show up as 1 1/2 stars in the program's GUI. Matt Schinckel provides interesting statistical evidence showing that not only do non-whole, non-half values (e.g., 47) not increase the amount a song will play, but neither do the half-star ratings (with the exception of .5 star)."

8 of 167 comments (clear)

  1. How does the Windows version work? by CyricZ · · Score: 2, Interesting

    Has anyone been able to replicate this behavior with the Windows version of iTunes? Is it built upon the same core as the Mac OS version, or does Apple have a separate implementation that may behave differently?

    --
    Cyric Zndovzny at your service.
  2. Re:How does it handle values outside the range? by frostilicus2 · · Score: 5, Interesting

    I've just tried this - If you insert a value outside the range, iTunes assumes the nearest value to 0, 20, 40, 60, 80, and 100 and doesn't crash.

    --
    Nothing sucks like a Vax, nothing blows like a PowerMac G4
  3. closed source.. by GrAfFiT · · Score: 2, Interesting

    Interesting to see that what would just involve reading a few lines of code on open source software takes a whole complicated statistical analysis on closed source software...

  4. why linear? by kavehkh · · Score: 5, Interesting

    Why is the rating system linear instead of exponential. I would like a song that has one more star than another to play TWICE more often.

    1. Re:why linear? by Anonymous Coward · · Score: 1, Interesting

      as another poster has pointed out, zero means no rating, so we're talking 16 times. Now: imagine the worst song you don't delete: it's fair that a dozen turds together should get less of your time than a gem. Or, in other words, if you have two 5-star gem cd's, and 24 CD's of more or less crap, then short of deleting them, you should certainly not have to hear much more (as you suggest) than one song from one of the crap CD's for each good song on one of the good CD's! Imagine if for every +5 mod'd post on slashdot, you had to read through many more posts at 0 until you could see the next one? If anything, the algorithm should give LESS than 1/2 as much weight for each fewer star (to the point that some people hard-hide posts at 0 and -1 -- they don't even know they're there!)

      I believe that each star should be one standard deviation, and that the mean should be at 2, and that 0 should not be "no rating" but "total and worthless crap (like -1 on slashdot). Like: if I recorded myself shitting and looped it for the length of a song (remember The Turd Report troll ["News for Turds. Stuff that splatters"] on slashdot?), that's what 0 should be. It doesn't really need to be more fine-grained than that.

      Then 1 should be one standard deviation below the average, two stars should be about average, three stars should be a standard deviation above average, and four stars should be two standard deviations above the average.

      Five stars is if you'd gladly memorize the entire 72 minutes of the CD. (Although, in practice, I think it would have to be like Slashdot's +5 posts.)

      I think it is fair to listen to more than 40 hours of mostly fantastic music and some good music before being subjected to three minutes of a song so awful you have to skip it: therefore, I think a 2-base logarithm decays too slowly. Under "Rules for normally distributed data" in the Standard Deviation Wikipedia article you can note that I propose lumping the low-end green and brown to 0-stars, the low blue to one star (*), the high-blue to two stars (**) the high brown to three stars (***) and the high green to ****.

      This accounts for 99.73% of all music, half of the remainder of which is "better" than four-stars, meaning 0.135% of music in the whole of the iTunes music store should be at 5-stars (my definition is, I'd gladly memorize every second of the CD-length).

      the iTunes music store says: "Featuring more than 2 million songs", so there is 2,700 songs or about 270 CD's worth memorizng. That is more selective than The Rolling Stones 500 Greatest Albums of All Time, but then I don't think anyone would like to hear those 500 albums enough to memorize them. I would not consider it inappopriate to spend 75% of my time listening to the 2700 songs off of iTunes that are rock-solid five-star songs, since my definition is that they're pretty much good enough to memorize. Any one of those songs might very well be good enough to listen to more or less attentively for a few hours straight, under the right circumstances or with the right intention.

    2. Re:why linear? by NeMon'ess · · Score: 2, Interesting

      I agree that the fixed settings suck. As a user actually comfortable operating a computer, unlike much of the market, I want advanced options that let me adjust how much more and less often I hear songs. So Party Shuffle isn't worth using to me.

      Instead I rate my tracks like I would a movie.
      1-star is probably going to get deleted eventually.
      2-stars play half as much as No-stars.
      No-stars play twice as much as 2-stars.
      3-stars play 3 times as much as 2-stars.
      4-stars play 4 times as much as 2-stars.
      5-stars play 6 times as much as 2-stars.

      Then I make an enormous playlist containing 1 play of 2-star songs, 2 plays of No-star songs... and 6 plays of 5-star songs, shuffle them, and save the result in a new playlist. Takes between 3 and 5 months to go through the list.

  5. 5 Stars Rating Systems are Poor Design by ChristopherA · · Score: 3, Interesting

    If you read my article on the Rating Systems you'll see that 5 point rating systems are a poor way to do ratings. In our analysis, only 10% of raters use the bottom two star in a 5-point scale, and only 2% use the lowest rating of 1 star. The median of the 5-point scale is actually the fourth star, with a neat bell curve arranged around it. In my own personal use of iTunes, I've forced myself to be much more consistent and lower with my ratings. 1 star means that it has been rated, but in general I don't like it and don't play it. 2 stars means average, which means play it some of the time. 3 stars is above average, and play it more often, but I can get bored with these. 4 is good, and can basically listen to these regularly without getting tired. 5 stars are exemplars -- I try to find more songs like these.

    1. Re:5 Stars Rating Systems are Poor Design by ibennetch · · Score: 2, Interesting

      I think this is because people tend to think of a 5-point rating system as being 5 stars for love it and 0 for hate it -- where as if they hate a song, they probably don't even own it. Even "strongly dislike" which is what most people would figure as 1 or 2 stars, doesn't make the cut to be purchased for my music collection. I like your system, where 5 is exceptionally good and 2 is average; it's much more useful for this type of thing because the music that would typically get rated 0, 1, or 2 isn't even in one's music collection, so the whole rating system needs to compensate for that.