通过多轴投票集成，提升心理防御机制分类的准确性。
Nürnberg NLP at PsyDefDetect: Multi-Axis Voter Ensembles for Psychological Defence Mechanism Classification
- 构建九人投票集成系统，覆盖类别粒度、训练方法和模型架构三轴。
- 在隐藏测试集上取得F1=0.420，位列21支队伍第一名。
- 适合需要高鲁棒性心理分析任务的研究者使用。
在支持性对话中识别心理防御机制的层级具有内在模糊性。在2026年BioNLP会议的PsyDefDetect共享任务中，八个正向防御类别表面语言相似，仅在语用功能上存在差异，且标注者间一致性仅为中等水平。在此类任务中，决定性优势并非更强的单模型，而是错误独立性，因为单一表示在重叠的防御边界上表现不稳定。本文将此洞察转化为一个9人投票集成系统，涵盖三个正交维度：类别粒度（门控者处理全部九类，专家仅处理八种防御类）、训练方法（生成式与判别式）及基础模型。该系统在隐藏测试集上获得F1_{test}=0.420，位列21个参赛团队首位。
原文摘要 · Abstract (English)
Detecting levels of psychological defence mechanisms in supportive conversations is inherently ambiguous. In the PsyDefDetect shared task at BioNLP 2026 the eight positive defence categories share surface language and differ only in pragmatic function and trained raters reach only moderate inter-annotator agreement. On such a task the decisive lever is not a stronger single model but error independence, since any single representation will waver on the overlapping defence boundaries. We translate this insight into a 9-voter ensemble spanning three orthogonal axes: class granularity (all nine classes for the gatekeeper, only the eight defence classes for the specialists), training method (generative and discriminative) and base model. The system reaches $F1_{test}{=}.420$ on the hidden test set, placing first among 21 registered teams.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。