Modeling Opinion Polarization in Social Media through Abstract Argumentation Frameworks

GABRIELA DIAZ; CARLOS CHESÑEVAR; ESTEVEZ, ELSA; ANA MAGUITMAN

Abuja

Conferencia; ICEGOV 2025; 2025

Opinion polarization in social media poses critical challenges fordemocratic discourse and policy-making. Recent research has ledto the development of specialized frameworks (e.g. the Polvizframework) for analyzing and studying opinion polarization bycombining visual analytics, stance detection, and argumentation-based topic modeling to reveal polarized viewpoints. However, suchframeworks fall short in capturing inter-argument conflicts andthe possible emerging semantics provided by computational ar-gumentation. This paper presents Polviz-AAF, a novel approachthat refines Polviz’s clustered opinions into a Dung-style AbstractArgumentation Framework, where each argument corresponds toa cluster of opinions sharing topic and polarity, and attacks areinduced by opposing stances on related topics. We show how thisalternative characterization enables the construction of argumentsfrom stance trees, the definition of the attack relation via topic-similarity metrics, and the computation of standard semantics. Wedemonstrate the feasibility of our proposal by applying it to climatechange opinions, illustrating how argumentation semantics canidentify dominant and contested opinion clusters, and discussingthe implications for explainable policy analytics