Many applications call for techniques for representing and reasoning about preferences, i.e., relative desirability over a set of alternatives. Preferences over the alternatives are typically derived from preferences with respect to the various attributes of the alternatives (e.g., a student's preference for one course over another may be influenced by his preference for the topic, the time of the day when the course is offered, etc.). Such preferences are often qualitative and conditional. When the alternatives are expressed as tuples of valuations of the relevant attributes, preferences between alternatives can often be expressed in the form of (a) preferences over the values of each attribute, and (b) relative importance of certain attributes over others. An important problem in reasoning with multi-attribute qualitative preferences is dominance testing, i.e., to find if one alternative (assignment to all attributes) is preferred over another. This problem is hard (PSPACE-complete) in general for well known qualitative conditional preference languages such as TCP-nets.

We provide two practical approaches to dominance testing. First, we study a restricted unconditional preference language, and provide a dominance relation that can be computed in polynomial time by evaluating the satisfiability of an appropriately constructed logic formula. Second, we show how to reduce dominance testing for TCP-nets to reachability analysis in an induced preference graph. We provide an encoding of TCP-nets in the form of a Kripke structure for CTL. We show how to compute dominance using NuSMV, a model checker for CTL.

We address the problem of identifying a preferred outcome in a setting where the outcomes or alternatives to be compared are composite in nature (i.e., collections of components that satisfy certain functional requirements). We define a dominance relation that allows us to compare collections of objects in terms of preferences over attributes of the objects that make up the collection, and show that the dominance relation is a strict partial order under certain conditions. We provide algorithms that use this dominance relation to identify only (sound), all (complete), or at least one (weakly complete) of the most preferred collections. We establish some key properties of the dominance relation and analyze the quality of solutions produced by the algorithms. We present results of simulation experiments aimed at comparing the algorithms, and report interesting conjectures and results that were derived from our analysis.

Finally, we show how the above formalism and algorithms can be used in preference-based service composition, substitution, and adaptation.