Section: New Results

Argumentation Systems for Decision Making

Participants : Rallou Thomopoulos, Madalina Croitoru, Jérôme Fortin, Marie-Laure Mugnier.

In collaboration with: Joël Abecassis (IATE/INRA), Jean-Rémi Bourguet (UM3), Patrice Buche (IATE/INRA), Sébastien Destercke (IATE/CIRAD) Nir Oren (Univ. of Aberdeen, Scotland)

Scientific investigations in this axis are guided by applications of our partners in agronomy (IATE laboratory). Substantial part of the work has consisted of analyzing the proposed applications and the techniques they require in order to select appropriate applications with respect to our team project.

Argumentation is a reasoning model based on the construction and the evaluation of arguments. In his seminal paper, Dung has proposed an abstract argumentation framework [56] . In that framework, arguments are assumed to have the same strength. This assumption is unfortunately strong and often unsatisfied. Consequently, several generalizations of the framework have been proposed in the literature. In [49] and [50] , we have led a comparative study of these generalizations. It clearly shows under which conditions two proposals are equivalent. We have also integrated those generalizations into a common more expressive framework.

An instantiation of Dung's abstract framework with the conceptual graph framework has been proposed. This representation uses default conceptual graph rules, an extension of classical conceptual graph rules (equivalent to existential rules, see Axis 1) with Reiter's defaults [67] allowing for non-monotonic reasoning, that we developed independently of the argumentation framework [42] , [43] . In the conceptual graph representation, arguments are represented as nested graphs, attacks between arguments can be computed from the structure of arguments and default rules allow to compute several kinds of extensions (i.e., maximal sets of arguments jointly acceptable according to a given semantics).

This approach has been applied to agrifood chain analysis, which is a highly complex procedure since it relies on numerous criteria of various types: environmental, economical, functional, sanitary, etc. Quality objectives imply different stakeholders, technicians, managers, professional organizations, end-users, public organizations, etc. Since the goals of the implied stakeholders may be divergent, decision-making raises arbitration issues. Arbitration can be done through a compromise—a solution that satisfies, at least partially, all the actors— or favor some of the actors, depending on the decision-maker's priorities. We have analyzed a case study concerning risks/benefits within the wheat-to-bread chain. It concerns the controversy about the possible change in the ash content of the flour used for commonly used French bread. Several stakeholders of the chain are concerned, in particular the Ministry of Health through its recommendations in a national nutrition and health program, millers, bakers and consumers.

As already pointed out, the proposed approach is novel both for theoretical and application aspects.

Let us mention additional results related to the applications in agronomy on decision making combining machine learning based on decision trees and ontologies [58] ,[30] , as well as results obtained by our collaborators on semi-automatic data extraction from web data (tables), data reliability, and the representation and flexible querying of imprecise data with fuzzy sets [16] , [14] , [17] , [26] , [31] , [25] , [27] , [33] , [34] . These investigations are complementary to the above mentioned results on argumentation and generally relate to other aspects in the same applicative projects.