• 디지털인문사회과학부 스페셜세미나-(11/14 금. 10:30~12:00), Back to 'Data' Science in the Age of Al
  • 관리자 |
  • 2025-11-05 13:29:58|
  • 407

디지털인문사회과학부에서 특별 세미나를 개최합니다. 관심 있는 분들의 많은 참여 바랍니다.
 

Title: Back to 'Data' Science in the Age of Al
Speaker: Patrick Park (Carnegie Mellon University)
Date / Time: Nov 14th, 2025 (FRI) 10:30am - 12:00pm (Lunch will be provided.)
Venue: N4-1433
Inquiry: lanukim@kaist.ac.kr

 

Abstract:
Unconventional uses of data can stimulate creativity and innovation at scales that dwarf the creativity and innovation unlocked by unconventional applications of established knowledge. In this talk, I will present three studies, each motivated by separate questions of human behavior in social networks, yet collectively shed light on the benefits and challenges of unconventional uses of data. Using Twitter communication and tweet deletion data, the first study develops and tests a novel network mechanism through which network brokers' individual decisions to self-censor can collectively lead to online opinion polarization. The second study applies sociological theory of interaction rituals to operationalize higher-order group interactions in a simplicial complex representation of communication among Twitter users. Analysis reveals that users who interact in a shared context tend to exhibit ritualistic aspects of offline group interaction, such as markedly higher communication frequency, focus on the collective, and stronger affect, which would not have been discernible in conventional graph-based representations. The final study attempts to explain the puzzle of scientific disruption, disproportionately produced by small teams in the age of big science. Analysis of scholarly acknowledgements in sociology journal publications suggests that small teams, perhaps by necessity, may produce disruptive knowledge in the course of seeking intellectual resources from informal academic ties positioned in distant niches in knowledge space. The talk will briefly reflect on the challenges of repurposing and/or combining data in unconventional ways, including construct validity, generalizability, survivorship bias, and research ethics, then conclude with potential implications for Al research.