Research

Short-video platforms such as TikTok, Instagram, and YouTube Shorts have become a prominent venue for self-expression, entertainment, sensemaking, political communication, and cultural transmission. The complexity of video content and its computational barriers have held back research on how these platforms work. This project uses vision-language models to study how visual and semantic conventions emerge in video-creation choices, in screenplay and structure, in technical cinematography standards, and in meaning, tracing how online culture evolves and how new trends emerge. Using calibrated patterns detected by vision language models, we develop an AI social simulation of a video creation platform to examine the mechanisms of imitation and emergence of conventions.

My dissertation asks how digital platforms enable voluntary crowds to coordinate physical work, such as aid distribution or neighborhood repair after a disaster, without formal roles or organizational workflows. Information Systems research has studied open collaboration in online settings such as open-source software and wikis. This project extends that phenomenon to physical work, where the materiality of the task changes what coordination requires. I developed a theory of technology-enabled physical crowd coordination and tested it in a group-level experiment with 512 participants on a purpose-built online platform that simulates disaster relief. Making the associations between physical elements of work more visible improves coordination performance by building shared situational awareness, and the effect grows with task interdependence. The stream continues with computational models of uncertainty and communication breakdowns, and with algorithmic coordination.

Marketing research mostly treats individuals as the units of consumption, yet in cultural and entertainment markets people usually attend as groups. With marketing colleagues, I modeled moviegoing as joint consumption, involving group choice processes, heterogeneous preferences, and social networks. The role of advertising and negative word of mouth shifts substantially when groups are treated as the unit of consumption, which departs from classical individual choice models. The agent-based model has 50,000 agents and replicates a full year of box office sales with week-level accuracy, for over 130 movies screened in thousands of theaters. It informs theory and models of social choice and advances granular validation of agent-based models against real-world competition patterns. The findings extend to other cultural markets, such as concerts, theaters, restaurants, travel, and streaming platforms.

This work examines how distinct emotions propagate on social media following celebrity suicides. We traced the resharing cascades of over one million tweets and detected emotions with a BERT-based model. Distinct emotions follow different propagation dynamics in cascade size, lifetime, speed, and burstiness, and these differences cannot be explained by valence or arousal levels. Celebrity suicides carry detrimental consequences for societal mental health and raise the risk of copycat suicide behavior, so the findings inform public policy on tracking distress and self-harm risk in the wake of public crises.