Characterizing Suspicious Commenter Behaviors

Abstract

YouTube has revolutionized content consumption and global user interaction. It has become a central hub for video sharing, entertainment, and information dissemination. However, as the user base continues to expand and actively engage with the platform, concerns have arisen regarding the presence of suspicious behavior among commenters. This study presents an approach based on social network analysis to detect suspicious commenter behaviors and identify similarities across various YouTube channels in relation to such behaviors. The analysis involves 20 YouTube channels that disseminated false views about the U.S. Military. The dataset included 7,782 videos, 294,199 commenters, and 596,982 comments. We employ a combination of methods, including Graph2vec, UMAP, K-means, Hierarchical clustering, qualitative and quantitative analyses. The objective is to categorize channels based on the level of suspicious behavior and reveal common patterns exhibited among them. To assess the effectiveness of the proposed methodology, the outcomes revealed the presence of commenter mobs and significant similarities among these channels, providing valuable insights into the prevalence of suspicious commenter behavior.