Springer’s Social Network Analysis and Mining (SNAM), a prestigious journal at the intersection of computational modeling and social systems, recently published our team’s study titled “Developing a Network-Centric Approach for Anomalous Behavior Detection on YouTube.” This research introduces a novel, scalable, and interpretable methodology rooted in social network analysis to identify and characterize coordinated anomalous behaviors specifically, commenter “mobs” that artificially manipulate engagement signals such as likes, shares, and comment volumes on YouTube.

The study leverages a comprehensive dataset of 47 YouTube channels, comprising 26,901 videos, nearly 1.4 million commenters, and over 2.49 million comments. Of these, 20 channels were associated with misleading narratives on U.S. military activities, while 27 channels served as a baseline control group. By constructing co–commenter networks, the research team extracted features of interaction and coordination that go beyond surface-level activity metrics.

The findings show a clear structural divide between anomalous and normal channels, with manipulated channels displaying tighter, denser clusters of activity that reflect coordinated efforts rather than organic audience participation. Trustworthiness assessments revealed that channels amplifying deceptive or militarized content could be reliably flagged using network-based feature representations combined with advanced dimensionality reduction methods (e.g., PCA, UMAP, Graph2Vec) and clustering algorithms such as K-means and hierarchical clustering.

The research also emphasizes the role of symbolic communication in driving coordinated behaviors. Misleading military narratives often employed provocative imagery, and emotionally charged language to draw viewers into orchestrated networks of engagement. These symbolic cues acted as catalysts, enabling anomalous communities to coalesce more rapidly and sustain manipulative campaigns at scale.

Beyond its immediate findings, the study advances both methodological and theoretical contributions. Methodologically, it demonstrates how combining network science, machine learning, and clustering yields interpretable and actionable frameworks for detecting manipulation in large-scale online ecosystems. Theoretically, it highlights how cultural and political symbols intersect with computational propaganda strategies, offering new perspectives on how trust, engagement, and information campaigns interact in algorithm-driven environments.

Prof. Agarwal reflected on the study’s impact, stating: “This research offers a robust, network-centric framework capable of detecting both subtle and large-scale coordinated commenter behaviors. It equips policymakers, platforms, and researchers with much-needed tools to safeguard the integrity of online discourse and mitigate the influence of manipulative engagement tactics.”

Ultimately, the study underscores the importance of integrating social theory with computational methods to better understand the evolving dynamics of online influence operations and to build resilient digital ecosystems in the face of coordinated manipulation.

Click here to read the full article.