Subsampling Methods in Abstract Argumentation Frameworks

In abstract argumentation, arguments are modeled as nodes in a directed graph, with edges designating attack relationships. As argumentation frameworks scale up, practitioners seek smaller yet representative subgraphs that retain core properties of the original. This paper compares three subsampling methods—Random, Breadth-First Search (BFS), and Community-based—under a unified evaluation framework measuring both structural and extensional fidelity. Surprisingly, our results show that simple random subsampling often performs comparably to specialized methods when balancing these dimensions. This non-intuitive finding indicates that random node selection can effectively preserve essential properties. We analyze the trade-offs of each method and identify conditions under which random sampling succeeds, providing insights for practitioners working with large-scale argumentation frameworks.