# N"urnberg NLP @ GermEval Shared Task 2026: Harmful Content Detection in German Social Media through Error-Independent LLM Voters

arXiv: https://arxiv.org/abs/2608.22246 · HTML: https://mlsanity.com/paper/2608.22246/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security.
- Builds on 4 of its 38 references that are in the map.
- First posted on arXiv 23 Aug 2026.

## Placed in

- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)

## Builds on (most cited first, 4 of 4)

- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [LL"aMmlein: Transparent, Compact and Competitive German-Only Language Models from Scratch](https://mlsanity.com/paper/2411.11171/) (2024)
- [AI Systems in Text-Based Online Counselling: Ethical Considerations Across Three Implementation Approaches](https://mlsanity.com/paper/2601.08878/) (2026)
- [N"urnberg NLP at PsyDefDetect: Multi-Axis Voter Ensembles for Psychological Defence Mechanism Classification](https://mlsanity.com/paper/2605.07606/) (2026)

## Built on by (most cited first, 1 of 1)

- [N"urnberg NLP at ChildSafeAds 2026: Structurally Dissimilar Voter Ensembles under Four Levels of Data Access](https://mlsanity.com/paper/2609.34986/) (2026)

## Abstract (arXiv)

Harmful content in German social media does real-world damage, from calls to action to criminal defamation. The GermEval 2026 shared task scores its detection in four subtasks. The technical challenge is a severe class imbalance. The harmful classes are rare and share surface language with the dominant majority class, yet under macro-F1 they decide the score. The decisive lever is then not a stronger single model but error independence. This insight becomes a per-subtask nine-voter ensemble spanning three orthogonal axes: LLM, training method and class scope. Selected mainly on internal cross-validation, the system reaches macro-F1 of 89.56 (C2A), 71.63 (DBO), 54.84 (VIO) and 83.02 (DEF) on the hidden test set, placing first on all four subtasks.