Preference queries over sets

We propose a “logic + SQL” framework for set preferences. Candidate best sets are represented using profiles consisting of scalar features. This reduces set preferences to tuple preferences over set profiles. We propose two optimization techniques: superpreference and M-relation. Superpreference targets dominated profiles. It reduces the input size by filtering out tuples not belonging to any best k-subset. M-relation targets repeated profiles. It consolidates tuples that are exchangeable with regard to the given set preference, and therefore avoids redundant computation of the same profile. We show the results of an experimental study that demonstrates the efficacy of the optimizations.