jlattman/EU_speech_german_twitter

EU speech detection model

This model is trained to classify statements on the European Union. It finetuned a bert-base-german-cased model on 1700 sentences from the German parliament. It is trained to detect explicit and implicit mentionings of the European Union.

Code Book If a sentence mentions the EU in an explicit or implicit way, the model categorizes it as EU speech. Examples can be mentioning about policies, politicians, events, elections on the European level.

Examples:

1. This is what the #industry always wanted: EU emissions trading. This naturally leads to rising CO2 and electricity exchange prices. The best answer is a further EU renewable energy booster, which has a price-reducing effect. Fossil is expensive, RE is cheap
2. As part of #EuropeWeek, I'm discussing Europe with Söhre students today. May 8 (Liberation Day) is a perfect day for this.

Model Details Finetuned from model: google-bert/bert-based-german-cased Epochs: 3 Accuracy: 0.83

Lattmann, J. (2025, March 17). Detecting EU sentiment in texts: A LLM Machine Learning application for Euroscepticism research. https://doi.org/10.31219/osf.io/mravb_v1