Using Graph Convolutional Networks for Approximate Reasoning with Abstract Argumentation Frameworks: A Feasibility Study[zur Übersicht]
We employ graph convolutional networks for the purpose of determining the set of acceptable arguments under preferred semantics in abstract argumentation problems. While the latter problem is complexity-wise one of the hardest problems in reasoning with abstract argumentation problems, approximate methods are needed here in order to obtain a practically relevant runtime performance. This first study shows that deep neural network models such as graph convolutional networks significantly improve the runtime while keeping the accuracy of reasoning at about 80% or even more.
28.11.19 - 10:15