Maximising goals achievement through abstract argumentation frameworks: An optimal approach. (1st March 2020)

• Record Type: Journal Article
• Authors: Cohen, Andrea; Gottifredi, Sebastian; Vallati, Mauro; García, Alejandro J.; Antoniou, Grigoris
• Is Part Of: Expert systems with applications. Volume 141(2020)
• Publication Date: 2020-03-01
• Subjects: Abstract argumentation -- Argumentation dynamics -- Goals achievement -- MaxSAT; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2019.112930
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: Abstract argumentation frameworks (AFs) are updated to maximise goals achievement. Goals specify arguments that should be made accepted and/or rejected. Strategies for updating AFs are obtained via a combined argumentation-MaxSAT approach. Optimal strategies minimise the number of actions used to maximise goals' achievement. Strategies and updates are formally characterised and empirically analysed. Abstract: Argumentation is a prominent AI research area, focused on approaches and techniques for performing common-sense reasoning, that is of paramount importance in a wide range of real-world applications, such as decision support and recommender systems. In this work we introduce an approach for updating an abstract Argumentation Framework (AF) so that achievement with respect to a given set of goals is maximised. The set of goals identifies arguments for which a specific acceptability status (a labelling) will be pursued, distinguishing between " in " and " out " goals. Given an AF, a set of goals and a set of available actions allowing to add or remove arguments and attacks from the AF, our approach will select the strategy (set of actions) that should be applied in order to obtain a new AF where the goals achievement is maximised. Moreover, the selected strategy will be optimal with respect to the number of actions to be applied. In the context of argumentation-based expert and intelligent systems, our approach will provide tools allowing the user to interact with the argumentative reasoning process carried out by the system, learning how the strategy she undertakes will affect the recommendations she receives. For that, we propose an encoding of the AF, the available actions and goals as weighted Boolean formulas, and rely on MaxSAT techniques for selecting the optimal strategy. We provide an experimental analysis of our approach, and formally show that the results we obtain correspond to the optimal strategy.