Making decisions through preference-based argumentation

Leila Amgoud, Yannis Dimopoulos, Pavlos Moraitis

Decision making is usually based on the comparative evaluation of different alternatives by means of a decision criterion. The whole decision process is compacted into a criterion formula on the basis of which alternatives are compared. It is thus, impossible for an end user to understand why an alternative is good, or better than another. Recently, some decision criteria were articulated in terms of a two-steps argumentation process: i) an inference step in which arguments in favor/against each option are built and evaluated, and ii) a comparison step in which pairs of alternatives are compared on the basis of ``accepted'' arguments. Thus, not only the best alternative is provided to the user but also the reasons justifying this recommendation. % However, a two steps approach is not in accordance with the principle of an argumentation system, whose accepted arguments are intended to support the ``good'' options. Moreover, with such an approach it is difficult to define proof procedures for testing directly whether a given option may be the best one without computing the whole ordering. Finally, it is difficult to analyze how an ordering is revised in light of a new argument. This paper proposes a novel approach for argumentation-based decision making. We propose a Dung style system that takes as input different arguments and a defeat relation among them, and returns as outputs a status for each option, and a total preordering on a set of options. The status is defined on the basis of different inference mechanisms. The total preordering privileges the option that is supported by the strongest argument, provided that this argument survives to the attacks. The properties of the system are investigated.

Subjects: 15.5 Decision Theory; 3.5 Qualitative Reasoning

Submitted: Jun 16, 2008


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.