Algorithms for Solving Satisfiability Problems with Qualitative Preferences

In this work we present a complete picture of our work on computing optimal solutions in satisfiability problems with qualitative preferences. With this task in mind, we first review our work on computing optimal solutions by imposing an ordering on the way the search space is explored, e.g., on the splitting heuristic in case the dpll algorithm is used. The main feature of this approach is that it guarantees to compute all and only the optimal solutions, i.e., models which are not optimal are not even computed: For this result, it is essential that the splitting heuristic of the solver follows the partial order on the expressed preferences. However, for each optimal solution, a formula that prunes non-optimal solutions needs to be retained, thus this procedure does not work in polynomial space when computing all optimal solutions. We then extend our previous work and show how it is possible to compute optimal solutions using a generate-and-test approach: Such a procedure is based on the idea to first compute a model and then check for its optimality. As a consequence, no ordering on the splitting heuristic is needed, but it may compute also non-optimal models. This approach does not need to retain formulas indefinitely, thus it does work in polynomial space. We start from a simple setting in which a preference is a partial order on a set of literals. We then show how other forms of preferences, i.e., quantitative, qualitative on formulas and mixed qualitative/quantitative can be captured by our framework, and present alternatives for computing "complete" sets of optimal solutions. We finally comment on the implementation of the two procedures on top of state-of-the-art satisfiability solvers, and discuss related work.