schlenker/moderngbert-parl-german-stance-detection

Detection of Group Appeals

This model serves as a baseline of the LREC 2026 submission Appeal, Align, Divide? Stance Detection for Group-Directed Messages in German Parliamentary Debates.

moderngbert-parl-german-stance-detection is trained to detect group appeals in a given text sequence, i.e., to assign a stance (favour, against, neither) to a previously identified social group. Refer to our paper for further details on model training and evaluation.

---

🔍 Model Overview

• Task Type: Single-label, multi-class sequence classification
• Base Model: ModernGBERT 134M
• Fine-tuning method: full fine-tuning
• Language: German