온라인 정치 담론에서의 극단화 위험 측정 : 유튜브 댓글의 자동 담론 분류와 사건 기반 위험도 분석 = Measuring Radicalization Risk in Online Political Discourse : Automated Discourse Classification and Event-Driven Risk Analysis of YouTube Comments

다국어 초록 (Multilingual Abstract)

This study aims to quantitatively assess radicalization risk in online political discourse and to examine how such risk intensifies in response to political events. Focusing on videos and comments from the most-subscribed far-right YouTube channel in South Korea, the study proposes an automated analytical framework that integrates functional discourse classification, a behavioral stage scale, and a continuous risk index.
The results show that the majority of comments fall into the low-risk category, and the average risk level remains relatively low. However, despite this overall stability, a markedly asymmetric risk structure emerges in the upper tail of the distribution. In particular, following the issuance of an arrest warrant for a former president, both the 95th percentile of the risk distribution and the proportion of explicit violent discourse increased sharply on the following day.
Discourse-level analysis reveals that mobilization-related discourse appears more frequently than explicit violent discourse, while dehumanization and conspiratorial narratives tend to co-occur with mobilization or violence, facilitating escalation to higher-risk stages. In addition, reclassification of symbolic expressions demonstrates that removing non- actionable symbolic-only content reduces false positives without diminishing the detection of genuinely high-risk discourse.
Overall, the findings suggest that online political radicalization is not characterized by a uniform shift in average discourse intensity, but rather by selective escalation within attitude-aligned groups in response to salient events. This study highlights the importance of tail-focused risk indicators and stage-based analysis for effective assessment of online extremism, offering practical implications for platform governance and policy interventions.