I have spent a bit of thought trying to work out the best way of rating my iTunes music. Most of it I haven't rated, or have rated only by album. It's mostly an academic question, because there isn't much you can actually do with the ratings once you've stuck them in.
At first I found myself rating everything as three stars or more. I thought, if it's rated less than three then why not just delete it. Those were times of precious disk space for me, but it does make sense in general. Scrolling through a whole lot of junk music is a pain, even if you've got endless disk space. Not mention that obsessives like me obtain a great deal of satisfaction from having a neat music collection, with all the ID3 tags in good working order.
So I thought I need to reinterpret the ratings. Normally one star means something is shite. In my case I wanted one star to mean "just good enough not to get deleted". So I started doing that. And it worked pretty well. Most of my playlists drop stuff with one star, but the music is still there if I want to listen to it - which occasionally I do.
I also thought about the top rating - five stars. Should I keep it reserved for only the best, life-changing music? Should I assume that music goodness is uniform and give five stars to 20% of the collection? Or is it some sort of normal distribution (or something non-symmetric but similar)? Tough questions I know. Should I worry about trying to make my ratings comparable to other people's ratings. What about rating inflation? Will I one day need a 6th star? And when I'm in a good mood I'm likely to give everything good ratings. I find that every time I really enjoy a song I end up giving it four stars because giving it five makes me feel like a rating floozy, and giving it three just feels mean. I suspect there's a decided spike at four, and this troubles me.
Another problem is that I never take the effort to rate anything that I don't like. So everything has four stars or no rating at all. Which is totally useless. I've started bulk rating whole albums, hoping that over time I'd use some Bayesian logic to gradually make each song rating more appropriate. I did this just to get some data in there but it isn't very satisfying.
I'm inclined to think that I should maintain some sort of histogram and try to ensure that the ratings distribution sticks to something reasonable. Perhaps just a uniformish distribution. Hopefully, I'll subconsciously adjust my ratings based on short-term distortions in distribution over the long-term it stays reasonable.
Clearly, plenty more thought needs to be done on this. Although I can't help but wonder if my life would be better if I completely removed the whole rating column from iTunes altogether.
You’re such a good brother. I read attentively and well, I agreed with all the problems. And then, I realised that well, my way of dealing with it is really just not using the ratings system. Except this doesn’t work either because iTunes is really biased and is absolutely in love with Ben Harper mainly. Oh the traumas of the digital age!
jem / 1:36pm / 12 October 2007
I know. How much does iTunes love Ben Harper. I had to delete him entirely for iTunes to get over its obsession.
Ryan / 1:40pm / 12 October 2007
I too, believe that a 5-star system for a playlist is the wrong solution.
It scales poorly, it pans poorly, and the resolution is terrible.
I propose a system based on better-than. If you specify (y > x) relationships between songs, you will automatically have a naturaly sorting order for music. This may apply to higher levels as well: albums, artists — but maybe not genres because “good music” does not really corellate with genre.
Wil / 8:56pm / 13 October 2007
That’s interesting because a lot of the economic research I’m looking at is on exactly this problem. The only empirical thing you can really say about how good something is, is that it’s better or worse than something else. They’ve built a whole system of surveys on it (discrete choice experiments). It’s quite nifty.
The problem with their approach (and also what you’re suggesting) is that the comparisons are either between arbitrary items, or you need to randomly compare a very large number of pairs in order to get a complete ordering.
One solution might actually relate to something I was suggested to Denzil the other day. As each song is played, iTunes could propose another song that it thinks you would like equally (based on artist, genre or past ratings etc). You would then tell iTunes whether that song was better or worse than the song playing.
However, I suspect there would be a strong bias towards the current song. Particularly since we like and can evaluate a lot of songs that we don’t know the name of. If you were listening to a randomised playlist it could ask you to compare the current and previous song. It would overcome the recall problem, but would force you to listen to random music to give you statistically valid music ratings. Which probably seems silly to some people.
Ryan / 1:44am / 14 October 2007
And if you simply let people say A > B on the fly, then you’ll quickly get prference loops like A > B > C > A.
I think we should step back, and wonder, what is really our main problem? “sorting the list” is a solution to something else, because “properly” is currently undefined.
Why do we need the sorted information?
I have my singular playlist which has, over the years, grown to be sorted in some sort of arbitrary combined ordering of preference and category. My mostest favouritest artists are at the top, but in the middle is a set of band/pop/rock artists which are like apples & oranges compared to the top ones of Underworld, Orbital, Autechre; and at the very bottom I have loose un-albumed songs which are sometimes absolutely genius.
But point is; it’s not formally sorted, and I don’t have a formal problem.
Wil / 8:18am / 14 October 2007
You’re right. There is no concrete problem here. It’s merely about my dissatisfaction with the current solution to something which isn’t a problem either. If the feature was removed tomorrow, I would have forgotten about it in a week.
Ryan / 10:53am / 14 October 2007
Happiness achieved!
Wil / 8:11pm / 14 October 2007