So I'm looking at my charts, which are, as things stand, fairly accurate. However, I've come to the conclusion that a scoring system, rather than purely play-count might be a superior option for generating charts. The current system isn't bad, but could probably use improvement in a couple of areas (as far as ranking artists, not individual tracks). I'm thinking of this because:
Time. Ranking artists by number of times played ignores the time one spends listening to said artist. For example, based on pure play-count, I like 45 Grave or Leonard Cohen (24th in my current overall) far better than Merzbow (number 63), although I've spent more time listening to the latter. The difference is that many of Merzbow's tracks exceed 20 minutes in length, and tracks by the former two artists are far shorter. Using play count as the sole criterion for chart numbers and positioning is, IMHO, inaccurate in this case (as well as some others) in terms of positioning the respective artists in the charts.
It seems to me that a more accurate scoring system would take time into account, if it's done right.
A flat modifier (such as play count x track length) wouldn't work, for the same reason I have against pure play-count, as it would skew the results too heavily for artists who have insanely long tracks, and against those who only have short tracks (most punk bands would be scored at a low level, and most classical scored highly).
What I'd like to see with a scoring system:
Two metrics: Play count, and time modifier. Play count would remain as it always has, but the time modifier would have to be dynamic, and change periodically (such as every x number of tracks or days).
This modifier could be made by making a bell-curve of the length of tracks played (each play adding another count at its length), with the largest part of the curve making a standard modifier of 1. Example: If most of the bell encompasses a range of 3-6 minutes, the play count for a given track would be multiplied by 1 and added to the score. Using a bell curve would provide more accuracy than simple averages.
Along the curve would be different modifiers. In the scenario above, a 2 minute song may rate a multiplier of .8, and an 8 minute song 1.1, before being added to an artist's score (I'm not anywhere near qualified to write the algorithms, so these are just hypothetical).
Finally, the change in the time multiplier should change more slowly as it nears the edges of the curve to avoid skewing the results.
I'm sure that someone who knows maths better than I do will have objections on a technical basis, that I am not qualified to either support nor refute; as far as I can tell, this sort of scoring system would help score artists more accurately than pure play-count, even if play-count remains the first metric and influences the second.