Detecting EU sentiment in texts: A LLM Machine Learning application for Euroscepticism research

This article presents a supervised classification model for measuring mentionsand sentiments about the European Union. This research provides new ways tostudy and explain variation in Euroscepticism. Specifically, this research presentsa binary classification algorithm for detecting speech about the European Union.In addition, it introduces a second classifier consisting of a multi-class classificationmodel for EU praise, hard eurosceptic sentiment, policy scepticism and demand forfurther EU integration. These are widely used concepts in Euroscepticism research.Based on a pre-trained BERT-based model, I fine tune my classification models on2000 human annotated tweets of German election candidates between 2007-2021. Iam further refining this by training additional models on parliamentary speeches,in order to expand the potential texts to be analysed with these models. All modelsare available in German and English. This new classification application can expandthe possibilities for researchers to study how politicians write and speak about theEU.