The ratings strategy described above applies to push attacks; this strategy can
easily be adjusted for nuke attacks. For example, positive correlations but negative
prediction shifts can be achieved by assigning the target item a rating of rmin, and
ratings of rmax and rmax − 1 to the more- and less-liked selected items.
The knowledge requirement here is intermediate between the bandwagon attack
and the average attack. Like the bandwagon attack, the popular items can usually be
easily estimated from outside the system; but because there are no filler items, the
Popular Attack will need more popular items. The attacker then needs to guess at
the relative average preferences between these items in order to provide the correct
rating. It might be possible to extract such distinctions from the system itself, or if
not, to mine them from external sources; for example, by counting the number of
positive and negative reviews for particular items to find general trends.