We are thrilled to share one of our studies that was recently published in the Journal of Social Network Analysis and Mining (SNAM) titled “Safeguarding YouTube Discussions: A Framework for Detecting Anomalous Commenter and Engagement Behaviors”. The study addresses one of the most pressing challenges in today’s digital landscape: the manipulation of online discussions through coordinated and deceptive user behaviors. Using an extensive dataset of over 12 million commenters and 123 million comments across 71 YouTube channels, the authors developed a robust analytical framework to detect and measure anomalous activity within video comment sections. Their study focuses on identifying irregular patterns of engagement, such as artificially inflated comments, coordinated posting, and sudden spikes in user activity, which can distort public discourse and amplify information campaigns.
To achieve this, the research employs advanced unsupervised learning techniques, including Kernel Density Estimation (KDE) and Gaussian Mixture Models (GMM), to identify statistical deviations from normal engagement patterns. The model integrates multiple dimensions of analysis through an innovative scoring system that combines feature-level metrics such as cosine similarity and Principal Component Analysis (PCA) with output-level aggregation methods, including harmonic mean (HM), weighted average with interaction term (WAIT), and agreement-weighted maximum (AWM). The resulting unified anomaly score effectively distinguished organic channels from those channels that deployed inorganic means to boost engagements and comments. Such channels were later suspended by YouTube, validating the model’s practical accuracy and real-world relevance.
Beyond its technical contributions, the paper underscores the rising sophistication of influence operations and coordinated networks seeking to manipulate public sentiment, particularly in geopolitically sensitive regions like the Indo-Pacific, where information warfare and narrative shaping can have tangible political and societal consequences. The study emphasizes that such deceptive engagement not only erodes trust in digital platforms but also threatens the integrity of democratic dialogue and informed decision-making.
From a practical perspective, the proposed framework provides a scalable, language-independent, and platform-neutral solution for identifying anomalous engagement behaviors across diverse social media ecosystems. Its adaptability enables researchers, policymakers, and platform administrators to apply the methodology beyond YouTube to detect coordinated information campaigns, safeguard civic discourse, and strengthen the resilience of online information environments. The paper further calls for interdisciplinary collaboration between computer scientists, social scientists, and policymakers to establish standardized benchmarks for anomaly detection and promote transparent, ethical AI practices in digital governance.
Prof. Agarwal said, “Overall, this publication marks a significant milestone in the effort to preserve the authenticity of social media interactions and combat coordinated online manipulation. It reinforces COSMOS’s leadership in advancing computational social science research that bridges technology, ethics, and policy. Through studies like this, we continue to demonstrate how data-driven innovation can be harnessed to protect digital ecosystems, promote responsible AI, and foster trust in the global information landscape.” Click here to read the full article.