Spark Talk: The Augmented Social Scientist: Using Sequential Transfer Learning to annotate Millions of Texts with Human-Level Accuracy by Salomé Do (Jan 12, 2024)

Description: The talk presents an automated content analysis method based on supervised machine learning (SML) and recent advances in transfer learning. The method will be presented through a case study on strategic framing in the press. Attendees will first be introduced to SML basics. They will then be shown how classic content analysis work can be ‘augmented” through SML, enabling them to annotate millions of texts following a traditional coding scheme without having to painstakingly implement a complex set of rules. Ultimately, factors leading to the human-level accuracy of SML modules (transfer learning, annotation quantity and quality) will be discussed.

Bio: Salomé Do is a PhD Student at ENS and Sciences Po. Her research focuses on methodological challenges faced using large language models for automated content analysis. To do so, she studies news framing analysis and explores how framing could in turn be a key concept for computational approaches to text.