MUCnoHARM@GermEval Shared Task 2026: Retrieval-based In-Context Learning for Defamatory Offences, and Where It Falls Short

2026-09-09

AI Summary

研究了多种基于检索的上下文学习方法来检测诽谤性言论，发现少样本提示优于零样本，但基于检索的方法仅提供边际收益。模型选择最为关键，但过预测犯罪相关性。

Abstract

With hate speech being ubiquitous online, automatic detection is crucial, in particular when it comes to criminally relevant social media posts. We study a variety of retrieval-based in-context learning (RetICL) strategies for detecting defamatory offences under §§ 185-187 StGB (the subject of GermEval 2026 Subtask 4). Few-shot prompting beats zero-shot, but retrieval-based approaches offer only marginal gains over random demonstrations, and even fall behind an optimised static set of demonstrations. Providing concrete legal knowledge helps, yet model choice outweighs every other system choice. Models over-predict criminal relevance while still missing 26-57% of criminally relevant posts, suiting them for triage rather than autonomous moderation.

Problem

• hate speech
• defamatory offences
• retrieval-based in-context learning
• criminally relevant
• social media

Innovation

• retrieval-based in-context learning
• defamatory offences detection
• few-shot prompting