— with Vihang Jumle (University of Bern)
When: October 10, 9am-5.30pm
Where: WI Kassenhalle; in-person only
Level: Intermediate
Category: Computational methods; text analysis
Seats: 20
Abstract: Text coding lies at the heart of many social science research projects – especially in communication studies, where identifying “frames” in media content reveals underlying meanings and narratives. Traditionally, this process is manual, time-consuming, and difficult to scale. This workshop introduces a powerful alternative: automating frame detection using transformer-based models like RoBERTa. Designed specifically for social scientists, the course teaches how to apply pre-trained language models to perform multi-class text classification, enabling scalable, reproducible, and rigorous content analysis. Intermediate experience in working with Python and its libraries is necessary. Experience with Natural Language Processing is advantageous
Participants will learn the fundamentals of transfer learning, model fine-tuning, and data preprocessing—without requiring deep mathematical or technical background. Through hands-on exercises, you will prepare real datasets, implement data augmentation strategies (e.g., synonym replacement, zero-shot sampling), configure hyperparameters (learning rate, batch size, epochs), and evaluate models using precision, recall, F1, and cross-validation. You will also learn how to log training runs, save models, and document your workflow for research transparency. The workshop follows a practical, project-based format, guided by real-world examples and troubleshooting support. Attendees are encouraged to bring their own research datasets and frame analysis questions, but it is not a requirement for registration and attendance.
There are required readings for this workshop:
Jumle, V., Makhortykh, M., Sydorova, M., & Vziatysheva, V. (2025). Finding Frames With BERT: A Transformer-Based Approach to Generic News Frame Detection. Social Science Computer Review.
Vihang Jumle is a doctoral candidate at the Institute of Communication and Media Studies, University of Bern. His research explores how digital media shape access to political information, using computational methods such as NLP and machine learning. His work has been published in journals including Social Science Computer Review, Nations and Nationalism, and Global Policy.
This workshop is organized together with Martha Stolze from the research group Platform Algorithms and Digital Propaganda at the Weizenbaum Institute.