EpiPersona: Persona Projection and Episode Coupling for Pluralistic Preference Modeling

arXiv cs.AI / 3/31/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper introduces EpiPersona, a framework for pluralistic alignment that separates enduring personal traits from episode-specific factors in LLM preference modeling.
EpiPersona projects noisy preference feedback into a low-dimensional persona space and aggregates similar personas into shared discrete codes to avoid relying on predefined preference dimensions.
It then couples the inferred persona representation with the current episode to enable episode-aware preference prediction.
Experiments indicate EpiPersona outperforms existing baselines, delivering especially strong gains in difficult episodic-shift settings and still performing well under sparse preference data.

Abstract

Pluralistic alignment is essential for adapting large language models (LLMs) to the diverse preferences of individuals and minority groups. However, existing approaches often mix stable personal traits with episode-specific factors, limiting their ability to generalize across episodes. To address this challenge, we introduce 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 codes. 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 consistently outperforms the baselines. It achieves notable performance gains in hard episodic-shift scenarios, while remaining effective with sparse preference data.