IP-BERT Evaluated Corpus: Predictions from Final Model (Model 11)

This dataset contains the output of applying the final fine-tuned IP-BERT model (Model 11) to the full corpus of German Bundestag plenary speeches from the Open Discourse corpus (1949–2021). Each row represents a 3-sentence passage extracted from a Bundestag speech, together with the model's prediction of whether the passage addresses economic inequality. The dataset was generated by running inference.ipynb from the accompanying code repository (see Related Identifiers) on the full corpus of 3-sentence units. IP-BERT is a German BERT model (based on dbmdz/bert-base-german-cased) fine-tuned via active learning to classify text passages as relating to economic inequality (label 1) or not (label 0).