Graph Convolutional Networks and Graph Attention Networks for Approximating Arguments Acceptability -- Technical Report

An Argumentation Framework (AF) is a directed graph F = A,R where A is a set of abstract entities called arguments and R A A is the attack relation. Although Dung does not put additional constraints regarding the set of arguments, in this work we assume that A is a non-empty finite set of arguments. When(a,b) R, we say that a attacks b, and similarly if a S s.t. a attacks b, then the set of arguments S attacks b. Classical AF semantics rely on a notion of collective acceptability: the semantics allow to determine sets of extensions, which are sets of jointly acceptable arguments. Most extension-based semantics satisfy two basic properties: Definition 2 (Conflict-freeness and defense). Given an AF F = A,R and the set of arguments S A, we say that S is conflict-free if a,b S, (a,b) R. Then, given an argument a A, we say that S defends a if b A s.t.