1. médialab Sciences Po
  2. News
  3. Computational content analysis : how, when, and why ?

Computational content analysis : how, when, and why ?

Salomé Do will defend her thesis on computational content analysis on October 17, 2024

Notice, PhD Defense

On Thursday, October 17, 2024, Salomé Do will present her thesis titled "Computational content analysis : how, when, and why ? Measuring the prevalence of strategic framing in political journalism." This research was conducted under the supervision of Thierry Poibeau and Jean-Philippe Cointet, and was prepared at Lattice (École Normale Supérieure-PSL / CNRS) and at the médialab (Sciences Po), as part of the Doctoral School in Letters, Arts, Humanities, and Social Sciences (ED540) of ENS-PSL.

Abstract

This thesis explores theories of media framing and their implementation in computational studies, highlighting the potential and challenges of automating the detection of these frames. The empirical analysis focuses on detecting strategic framing in political articles from the newspaper Le Monde (1945-2018). Two specific indicators were developed, and various annotation methods were tested, showing that modern language models can achieve performance comparable to that of human annotators. The thesis examines the importance of expert annotations to improve model accuracy and then proposes a quantitative framework for comparing automated and traditional content analysis methods.

This framework is essential for informing the choice between computational and manual approaches for social science research. The study emphasizes the importance of rigorous error analysis and uncertainty estimation in supervised learning applications, suggesting avenues for future research to enhance the reliability and applicability of machine learning in content analysis. The findings advocate for an informed use of automated methods in media studies.

Composition of the jury

François YVON - CNRS, Sorbonne University - Rapporteur
Baldwin VAN GORP - KU Leuven - Rapporteur
Sophie MÜTZEL - University of Lucerne - Examiner
Dallas CARD - University of Michigan - Examiner
Chloé CLAVEL - INRIA - Examiner
Thierry POIBEAU - CNRS, École Normale Supérieure - Thesis Supervisor
Jean-Philippe COINTET - Sciences Po Paris - Thesis Supervisor

Practical information

The thesis presentation will take place on Thursday, October 17, 2024, at 2 PM in the Évariste Galois amphitheater, École Normale Supérieure, 45 rue d’Ulm, Paris.