Understanding Generalization in Role-Playing Models via Information Theory

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Yongqi, Lang, Hao, Huang, Fei, Qian, Tieyun, Li, Yongbin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913023422103552
author Li, Yongqi
Lang, Hao
Huang, Fei
Qian, Tieyun
Li, Yongbin
author_facet Li, Yongqi
Lang, Hao
Huang, Fei
Qian, Tieyun
Li, Yongbin
contents Role-playing models (RPMs) are widely used in real-world applications but underperform when deployed in the wild. This degradation can be attributed to distribution shifts, including user, character, and dialogue compositional shifts. Existing methods like LLM-as-a-judge fall short in providing a fine-grained diagnosis of how these shifts affect RPM generalization, and thus there lack formal frameworks to characterize RPM generalization behaviors. To bridge these gaps, we introduce an information-theoretic metric, named reasoning-based effective mutual information difference (R-EMID), to measure RPM performance degradation in an interpretable way. We also derive an upper bound on R-EMID to predict the worst-case generalization performance of RPMs and theoretically reveal how various shifts contribute to the RPM performance degradation. Moreover, we propose a co-evolving reinforcement learning framework to adaptively model the connection among user, character, and dialogue context and thus enhance the estimation of dialogue response generation probability, which is critical for calculating R-EMID. Finally, we evaluate the generalization performance of various RPMs using R-EMID, finding that user shift poses the highest risk among all shifts and reinforcement learning is the most effective approach for enhancing RPM generalization.
format Preprint
id arxiv_https___arxiv_org_abs_2512_17270
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Understanding Generalization in Role-Playing Models via Information Theory
Li, Yongqi
Lang, Hao
Huang, Fei
Qian, Tieyun
Li, Yongbin
Machine Learning
Artificial Intelligence
Computation and Language
Role-playing models (RPMs) are widely used in real-world applications but underperform when deployed in the wild. This degradation can be attributed to distribution shifts, including user, character, and dialogue compositional shifts. Existing methods like LLM-as-a-judge fall short in providing a fine-grained diagnosis of how these shifts affect RPM generalization, and thus there lack formal frameworks to characterize RPM generalization behaviors. To bridge these gaps, we introduce an information-theoretic metric, named reasoning-based effective mutual information difference (R-EMID), to measure RPM performance degradation in an interpretable way. We also derive an upper bound on R-EMID to predict the worst-case generalization performance of RPMs and theoretically reveal how various shifts contribute to the RPM performance degradation. Moreover, we propose a co-evolving reinforcement learning framework to adaptively model the connection among user, character, and dialogue context and thus enhance the estimation of dialogue response generation probability, which is critical for calculating R-EMID. Finally, we evaluate the generalization performance of various RPMs using R-EMID, finding that user shift poses the highest risk among all shifts and reinforcement learning is the most effective approach for enhancing RPM generalization.
title Understanding Generalization in Role-Playing Models via Information Theory
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2512.17270