arXiv：cs.AI· Yuan Yuan·· 20 小时前
超越文化人格偏差：响应几何作为语言模型与人类的度量方法
Beyond Cultural Persona Bias: Response Geometry as a Measure for Language Models and People
AI 导读
研究提出用响应几何（维度间相关矩阵构成的对称正定矩阵）衡量语言模型与人类的文化偏好。在 DeepSeek-V3 和 GPT-4o 中，固定题目顺序时文化偏好同时出现在 Big Five 得分和响应几何两个层面；随机化题目顺序后响应几何降至随机水平，Big Five 仍能以 95–97% 区分两种人格。
正文
Abstract:Researchers are increasingly prompting language models with cultural personas so that they answer questionnaires as diverse human groups would. However, studies often report that these persona responses do not predict the models' behavior. We argue that whether the persona is Western or non-Western, the model's answers carry a cultural preference, and that this preference masks the variables that could predict its behavior.
We examine the prompted preference at two levels: aggregate Big Five scores and response geometry, the correlation matrix between dimensions within each respondent, represented as a symmetric positive-definite (SPD) matrix. In DeepSeek-V3 and GPT-4o, when the item order is fixed, the prompted preference appears at both levels. Randomizing item order reduces it in the response geometry, while leaving it in the Big Five scores: the Big Five still separate the two personas at $95$--$97\%$, while response geometry falls to chance.
In randomized administration, the response geometry still contains information about the respondent. In DeepSeek-V4-Flash, SPD features are associated with whether the same instance sorts Asian American public figures as ``foreign'' in a text IAT ($p = .047$), while the Big Five do not. In human studies conducted using randomized item order, the response geometry identifies people from test-retest sessions in $81$ -- $85\%$, against chance $0.4\%$, and predicts sex and, weakly, age.
In addition, SPD features also predict complex human behavior (e.g., cumulative GPA, $R^2 = 0.28$, permutation $p = .003$), where the Big Five do not.
| Subjects: | Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Differential Geometry (math.DG) |
| Cite as: | arXiv:2607.02368 [stat.ML] |
| (or arXiv:2607.02368v4 [stat.ML] for this version) | |
| https://doi.org/10.48550/arXiv.2607.02368 arXiv-issued DOI via DataCite |
Submission history
From: Yuan Yuan [view email]
[v1]
Thu, 2 Jul 2026 16:11:44 UTC (646 KB)
[v2]
Fri, 21 Aug 2026 14:37:51 UTC (147 KB)
[v3]
Sun, 30 Aug 2026 02:08:52 UTC (152 KB)
[v4]
Thu, 8 Oct 2026 00:19:03 UTC (98 KB)
来源：arXiv：cs.AI · arxiv.org