Abstract
Intelligent systems that assist users in fulfilling complex tasks need a concise and processable representation of incomplete and uncertain information. In order to be able to choose among different options, these systems also need a compact and processable representation of the concept of preference. Preferences can provide an effective way to choose the best solutions to a given problem. These solutions can represent the most plausible states of the world when we model incomplete information, the most satisfactory states of the world when we express user preferences, or optimal decisions when we make decisions under uncertainty. Several domains, such as, reasoning under incomplete and uncertain information, user preference modeling, and qualitative decision making under uncertainty, have benefited from advances on preference representation. In the literature, several symbolic approaches of nonclassical reasoning have been proposed. Among them, logic programming under answer set semantics offers a good compromise between symbolic representation and computation of knowledge and several extensions for handling preferences. Nevertheless, there are still some open issues to be considered in logic programming. In nonmonotonic reasoning, first, most approaches assume that exceptions to logic program rules are already specified. However, sometimes, it is possible to consider implicit preferences based on the specificity of the rules to handle incomplete information. Secondly, the joint handling of exceptions and uncertainty has received little attention: when information is uncertain, the selection of default rules can be a matter of explicit preferences and uncertainty. In user preference modeling, although existing logic programming specifications allow to express user preferences which depend both on incomplete and contextual information, in some applications, some preferences in some context may be more important than others. Furthermore, more complex preference expressions need to be supported. In qualitative decision making under uncertainty, existing logic programming-based methodologies for making decisions seem to lack a satisfactory handling of preferences and uncertainty. The aim of this dissertation is twofold: 1) to tackle the role played by preferences in logic programming from different perspectives, and 2) to contribute to this novel field by proposing several frameworks and methods able to address the above issues. To this end, we will first show how preferences can be used to select default rules in logic programs in an implicit and explicit way. In particular, we propose (i) a method for selecting logic program rules based on specificity, and (ii) a framework for selecting uncertain default rules based on explicit preferences and the certainty of the rules. Then, we will see how user preferences can be modeled and processed in terms of a logic program (iii) in order to manage user profiles in a context-aware system and (iv) in order to propose a framework for the specification of nested (non-flat) preference expressions. Finally, in the attempt to bridge the gap between logic programming and qualitative decision under uncertainty, (v) we propose a classical- and a possibilistic-based logic programming methodology to compute an optimal decision when uncertainty and preferences are matters of degrees.