The 19th International AAAI Conference on Web and Social Media (ICWSM 2025), held from June 23–26 in Copenhagen, Denmark, served as a key gathering for scholars in social computing, AI, and digital behavior. The event brought together experts from around the world to present cutting-edge interdisciplinary research on the evolving dynamics of online platforms.
This year, COSMOS researchers contributed three significant papers, with two featured in the 6th International Workshop on Cyber Social Threats (CySoc), a co-located workshop focused on emerging sociotechnical risks.
In the first study, entitled “The Viral Nature of Symbols,” cosmographers used AI and epidemiological modeling to explore how visual symbols—such as flags, logos, and hand gestures—impact online engagement on multimedia-centric social platforms, such as TikTok and YouTube. The study sheds light on the symbolic influence of imagery in shaping engagement and sustaining momentum in digital information campaigns. “The research demonstrates how symbolic content functions as a vehicle for emotional resonance and collective identity. By applying (Susceptible Infected Recovered) SIR and (Susceptible-Exposed-Infected-Skeptic) SEIZ models traditionally used in disease modeling, the study reveals how certain visuals spread virally across networks, acting as markers of ideology, affiliation, or resistance. This innovative approach bridges computational modeling and media semiotics, offering a new lens through which to understand symbolic amplification in online discourse.” said Prof. Agarwal.
In the second study, entitled “PRISM: Perceptual Recognition for Identifying Standout Moments in Human-Centric Keyframe Extraction,” cosmographers proposed a perception-driven framework to identify key frames in videos. Their work enhances understanding of how visual media attracts attention and guides narrative construction in a scalable manner, leveraging advances in AI. Prof. Agarwal said, “PRISM framework utilizes perceptual cues—such as emotion, motion, and gaze—to extract moments that resonate most with human viewers. This system can significantly benefit platforms, journalists, and analysts by reducing vast video data into digestible, emotionally salient snapshots. It also advances computational storytelling by prioritizing human-centered interpretation, offering a pathway toward more intuitive and impactful multimedia summarization.”
In the third study, entitled “Telegram as a Battlefield, cosmographers in collaboration with researchers from Georgia State University, University of Massachusetts, and Wright State University introduce a comprehensive Telegram dataset capturing over 5 million posts from both pro- and anti-Kremlin channels, offering a rare longitudinal view of information dissemination surrounding the Russia-Ukraine conflict. “By documenting the platform’s dual role in spreading information and resistance content, the dataset provides a critical foundation for future research on digital warfare, manipulation, and platform governance in high-stakes geopolitical crises.” said Prof. Agarwal.
Together, these studies reflect COSMOS’s ongoing commitment to exploring the intersections of influence operations, visual perception, and global digital conflict.