Let's say I have 1000 objects and 10000 comparisons between individual
objects of the form
(object1, object2, which is better, magnitude)
So, for example I might have:
("Bob","Jim","Jim","smarter")
("Bob","Jay","Jay","smarter")
("George","Jay","George","way smarter")
Given a large set of those (10,000 in this example), how do I take
that data and determine an overall ordering? In the above example it
would be something like
Bob, Jim, Jay, George
or it could be
Bob, Jim, Jay, George
Also, there are going to be plenty of contradictory comparisons where
some think Jay is smarter than Jim while others think Jim is smarter
than Jay.
I don't expect to be able to get any sort of perfect ordering because
I doubt there is one in most cases. But if most people think Jim is
smarter than Jay and Jay is smarter than Bob then I'd like to get a
final ordering that order Bob then Jay then Jim.
If it makes the problem simpler I could get rid of the magnitude in
each data point so that it's just a binary comparison where A > B or A
< B. But, if possible I'd like an way to include the magnitude. |
Clarification of Question by
mxnmatch-ga
on
17 Apr 2006 00:04 PDT
And just to be clear, the real data will be of the form
(objectId1, objectId2, comparison)
with data like:
(92311,991101,1)
(123572,917213,-2)
which would mean that 991101 is better than 92311 while 123572 is way
better than 917213.
|