Fine-tuned BERT models (sequence labeling, stance classification, embedding via CL) for the detection of social group appeals in political texts.