Analysing Cross-Country Protest Dynamics: A Transformer-based Approach to Newspaper Content. EPSA (Cologne)

This study employs Large Language Models (LLMs) to analyse protest events across seven European countries using multilingual newspaper articles. Through supervised machine learning, we examine a subset of over 4,000 manually annotated texts from the Far-Right Protest Observatory dataset, describing public events by far-right collective actors. Our research has two complementary aims. First, we train a classifier to identify pertinent articles for protest event analysis (i.e. whether the texts focus on protest action). Second, within this subset, we employ two other classifiers to annotate nuanced event characteristics, such as the issue focus (i.e., the motivation of protest action) and forms of action (i.e., whether the events were confrontational/violent). This comprehensive approach aims to enhance cross-country protest event analysis by automating coding using LLMs, improving efficiency and accuracy. Our models will be made available open source.