Normalise rankings when displaying

[kaspiana]

Elo rankings only have meaning relative to other rankings in the same group, which makes the rankings on a given document confusing. Intuitively, a ranking of 40 will have the same meaning in any category, but it does not. We can ameliorate this confusion somewhat by normalising the rankings, perhaps to a percentage or ten-point system.

e.g.

Current

Rankings:
 * gothic: -32
 * cool:   -48
 * cute:   0
 * sexy:   -16
 * silly:  0
 * weird:  48
 * shitty: 16
 * stupid: 0
 * chic:   -32

Proposed

Rankings:
 * gothic: [****      ]
 * cool:   [***       ]
 * cute:   [*****     ]
 * sexy:   [*****     ]
 * silly:  [*****     ]
 * weird:  [**********]
 * shitty: [*******   ]
 * stupid: [*****     ]
 * chic:   [****      ]

[kaspiana]

Further normalisation may be necessary: an average Elo rating in cuteness, for example, would intuitively signify that an image is somewhat cute, when this is in fact wrong; the average document is one that is not cute. One can see this in the example above, the normalised sexiness rating is 5/10 because it has an average Elo rating, but the image this example is taken from was never judged sexy at all.

We could weight the ratings so that those at the average and lower are brought to zero. Perhaps anything at the average or lower should be zero?

[kaspiana]

Even with normalisation, Elo ratings will still reflect the composition of the sample space. If images with a certain quality are overrepresented, then all images will have a lower reported rating in that quality.

e.g. If most images are silly, then silly images will tend to have lower Elo ratings for silliness.