Séminaire

Text, Culture, Time: Computational Explorations of Collective Information

Zackary Dunivin (Indiana University)

2 avril 2024, 14h00–15h15

Auditorium 3 JJL

Salle Auditorium 3 JJL

SBS recruitment seminar

Résumé

Zackary Dunivin is a candidate in Complex Systems and Sociology at Indiana University. His work draws on various computational tools to study culture and cultural change from the perspectives of dyadic interactions, group dynamics, social movements and formal organizations. This talk presents findings from "Black Lives Matter protests shift public discourse" (PNAS 2022), which shows how BLM leveraged street protests to change the ways we talk about race and politics. Zackary will also report results from "Scalable qualitative coding with LLMs: Chain-of-thought reasoning matches human performance for some hermeneutic tasks" (arXiv 2024), which suggests that GPT-4 and subsequent generative models are capable of sophisticated social-scientific analysis of paragraph-long passages. State-of-the-art tools are far more powerful and accessible than preceding ML models, and allow for much larger samples than human-coding while also freeing up researchers to focus on more creative research aspects.