Disproportionate Voices: Participation Inequality and Hostile Engagement in News Comments

Abstract

Digital platforms were expected to foster broad participation in public discourse, yet online engagement remains highly unequal and underexplored. We ask whether who talks the most also shapes the tone that becomes visible in news comment sections. Using 260 million comments from 6.2 million users over 13 years on Naver News, we measure participation inequality with the Gini and Palma indexes and estimate hostility levels with a KC-Electra model, which outperformed other Korean pre-trained transformers in multi-label classification tasks. We test two hypotheses: (H1) a within-section, over-time association between participation concentration and hostile tone; and (H2) a decomposition of hostility changes into composition (who talks how much) versus behavior (how those same groups talk). We find that higher concentration is associated with higher hostile tone, and that changes in hostility are driven mainly by behavioral intensification among active users, rather than compositional shifts. These results quantify the relationship suggested by exploratory patterns and point to an amplification risk: when a few users dominate, visible discourse can skew sharper, potentially discouraging casual participation and reinforcing concentration.