Quantitative Modeling and Analysis of Argumentation Polarization in Cyber Argumentation

Cyber argumentation platforms offer specially designed environments for users to discuss and debate their stances and viewpoints on important issues. However, argumentation polarization often occurs in discussions and debates on these cyber argumentation platforms. Several researchers investigated argumentation polarization qualitatively in the past, but none of them have developed a quantitative model for measuring the degree of argumentation polarization. We addressed this important and challenging issue by developing an innovative argumentation polarization model to measure argumentation polarization by incorporating four important attributes of argumentation polarization: 1) the total number of argumentation poles; 2) the population size of the argumentation poles; 3) similarity within argumentation poles; and 4) the dissimilarity between argumentation poles. Its baseline model was derived from an economic polarization model proposed by Esteban and Ray, which measures polarization using three features: 1) homogeneity within each group; 2) heterogeneity across groups; and 3) a small number of significant groups. We adapted their model by incorporating population sizes for poles, normalizing for population size, and normalizing for parameter selection, to fit our formulation of argumentation polarization. This model was evaluated using an empirical study conducted using our cyber argumentation platform, the intelligent cyber argumentation system (ICAS). This model was evaluated with two other distribution-based opinion polarization models that were applied in online discussion contexts, both analytically and empirically, since there are no existing argumentation polarization models. The analytical and empirical evaluations indicate that our model performs more effectively in terms of the definition of argumentation polarization and the four attributes.