Possibilistic preference networks

This paper studies the use of product-based possibilistic networks for representing preferences in multidimensional decision problems. This approach uses symbolic possibility weights and defines a partial preference order among solutions to a set of conditional preference statements on the domains of discrete decision variables. In the case of Boolean decision variables, this partial ordering is shown to be consistent with the preference ordering induced by the ceteris paribus assumption adopted in CP-nets. Namely, by completing the possibilistic net ordering with suitable constraints between products of symbolic weights, all CP-net preferences can be recovered. Computing procedures for comparing solutions are provided. The flexibility and representational power of the approach is stressed.