发表机构
Ivannikov Institute for System Programming of the Russian Academy of Sciences; Russian Presidential Academy of National Economy and Public Administration（俄罗斯科学院伊万尼科夫系统编程研究所; 俄罗斯总统国民经济与公共管理学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究跨大语言模型对后苏联实体情感框架差异，用目标导向的效价-唤醒-优势剖析，向19个模型询问500个特定区域目标并评分验证，聚类得出行为聚类，发现效价-唤醒-优势剖析能捕捉情感强度，补充传统情感评估。
AI 中文摘要
在不同语言和区域生态系统中训练和对齐的大语言模型，可能会以不同方式构建相同的政治、文化和地缘政治实体。此类差异常通过情感、好感度或立场来评估，将模型态度简化为单一的正负轴。我们引入REGARD，一项使用目标导向的效价-唤醒-优势剖析来研究跨大语言模型对后苏联实体情感框架差异驱动因素的研究。我们向19个模型询问500个特定区域目标，由两名独立的大语言模型评判员GPT-4o-mini和Qwen3.6-35B-A3B对其回答评分，并在一个300项人工标注子集中验证测量结果。通过情感和反应行为概况对所有19个模型进行事后沃德链接聚类，得出三个跨越模型来源、家族和参数数量的行为聚类。通用答案率与较低唤醒度（r = -0.81）以及聚类位置密切相关：无论来源如何，那些用模板化回答转移评价性提示的模型在低唤醒度下聚集在一起。这些发现表明，效价-唤醒-优势剖析捕捉到了情感强度，这是情感框架的一个维度，在传统的基于情感的评估中基本不可见。
英文摘要
Large language models trained and aligned within different linguistic and regional ecosystems may frame the same political, cultural, and geopolitical entities in different ways. Such differences are often evaluated through sentiment, favorability, or stance, reducing model attitudes to a single positive-negative axis. We introduce REGARD, a study of what drives affective framing differences across LLMs on post-Soviet entities using target-directed Valence-Arousal-Dominance profiling. We query 19 models on 500 region-specific targets, score their responses with two independent LLM judges, GPT-4o-mini and Qwen3.6-35B-A3B, and validate the measurements on a 300-item human-annotated subset. Post-hoc Ward-linkage clustering of all 19 models by affective and response-behavior profiles yields three behavioral clusters that cut across model origin, family, and parameter count. Generic-answer rate is strongly associated with lower arousal (r = -0.81) and with cluster placement: models that deflect evaluative prompts with templated responses cluster together at low arousal regardless of origin. These findings show that VAD profiling captures emotional intensity, a dimension of affective framing that is largely invisible to conventional sentiment-based evaluation.
Comments17 pages, 11 figures, 3 tables. Includes evaluation of 19 language models, two independent VAD judges, and human validation on 300 items
