Conditional Preference Networks - Refining Solution Orderings Beyond Pareto Dominance.

When handling preferences, the inclusion-based partial order (called here Pareto dominance) is a natural basis for comparing solutions: one is then preferred to another as soon as the set of preferences violated by the former is a subset of the preferences violated by the latter. In order to obtain refinements of this partial order, various general principles can be used, such as Ceteris Paribus, or optimistic (resp. pessimistic) principles stating that what is not explicitly said to be satisfactory according to the preferences should be regarded as satisfactory (resp. disregarded). Algorithms taking advantage of the two last principles are presented and discussed. They enable us to refine the Pareto ordering of solutions.