Analyzing Public Opinion on International Conflict Through YouTube Comments Using NLP Techniques

Abstract

The proliferation of user-generated content across social media platforms has fundamentally reshaped the landscape through which public sentiment toward geopolitical events is expressed, contested, and amplified. This study examines public opinion formation and polarization surrounding international armed conflicts by analyzing a corpus of 2.4 million YouTube comments harvested from 847 conflict-related video uploads spanning the 2022–2024 period. Drawing on a multi-layered natural language processing pipeline that integrates transformer-based sentiment classification, topic modeling via Latent Dirichlet Allocation, and graph-theoretic opinion propagation analysis, we demonstrate that comment sections exhibit statistically measurable echo-chamber dynamics — with opinion clusters exhibiting inter-group Jaccard similarity scores below 0.18, markedly lower than intra-cluster baselines. A fine-tuned RoBERTa architecture achieved 91.4% macro-averaged F1 in six-class sentiment categorization, outperforming prior multilingual baselines by a margin of 6.3 percentage points. Temporal analysis further reveals that sentiment volatility spikes within 72 hours of breaking news cycles, before regressing toward regionally distinct equilibria. These findings carry implications for conflict communication scholars, platform governance researchers, and humanitarian information actors seeking to understand the digital public sphere during crises.