EpiPersona：人格投影与情景耦合实现多元化偏好建模
EpiPersona: Persona Projection and Episode Coupling for Pluralistic Preference Modeling
AI 导读
EpiPersona 是一个用于大语言模型多元化对齐的显式人格—情景耦合框架。它把含噪偏好反馈投影到低维人格空间，聚合为离散码字，将稳定个人特质与情景信号分离，再与当前情景耦合进行情境感知的偏好预测。实验显示其在情景剧变场景下性能明显优于基线，且在稀疏偏好数据下仍有效。
正文 · 原文
Abstract:Pluralistic alignment is essential for adapting large language models (LLMs) to the diverse preferences of individuals and minority groups. However, existing personalized reward models often mix stable personal traits with episode-specific factors, limiting their ability to generalize across episodes. To address this challenge, we propose EpiPersona, a framework for explicit persona-episode coupling. EpiPersona first projects noisy preference feedback into a low-dimensional persona space, where similar personas are aggregated into shared discrete codewords. This process separates enduring personal characteristics from situational signals without relying on predefined preference dimensions. The inferred persona representation is then coupled with the current episode, enabling episode-aware preference prediction. Extensive experiments show that EpiPersona outperforms the baselines. It achieves notable performance gains in hard episodic-shift scenarios, while remaining effective with sparse preference data.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2603.28197 [cs.AI] |
| (or arXiv:2603.28197v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2603.28197 arXiv-issued DOI via DataCite |
Submission history
From: Yujie Zhang [view email]
[v1]
Mon, 30 Mar 2026 09:04:24 UTC (687 KB)
[v2]
Sat, 24 Oct 2026 02:43:10 UTC (774 KB)
来源：arXiv cs.AI · arxiv.org