Finding tags with "discriminating property"

[Benjamin Mesing]

Hello,


>  1) take a list of packages
>  2) find 2 the best selections of tags A and B that approximate that
>     ~50% of the packages have tag A and ~50% of the packages have tag B
>  3) present the user the choice "Do you want A or B?"
>  4) weed out the half of the list that the user doesn't want
>  5) continue until the user is satisfied
> 
> Now there is to be seen how to implement point 2).  But the idea is
> interesting and I'm noting it here so that at least it doesn't get lost.
The question is, if it is possible at all. Especially so, since every
package must be contained in at least A or B, being in both won't hurt
much, but being only in one would be bad.
You could of course do: 
        A = { tag1, tag2, tag3, .. }
        B = { not tag1, not tag2, not tag3, ...}
But this looks not very smart.

Even if it is possible to find such two tagsets, I doubt that it is a
good approach to select packages. Giving the user two sets of totally
unrelated and tags could lead to confusion. It might help narrowing
searches that are already reduced to a few (perhaps up to 500?) packages
though.

Best regards 

0Ben