arXiv:2511.08017v1 Announce Type: new 
Abstract: Multi-character role-playing aims to equip models with the capability to simulate diverse roles. Existing methods either use one shared parameterized module across all roles or assign a separate parameterized module to each role. However, the role-shared module may ignore distinct traits of each role, weakening personality learning, while the role-specific module may overlook shared traits across multiple roles, hindering commonality modeling. In this paper, we propose a novel HyCoRA: Hyper-Contrastive Role-Adaptive learning framework, which efficiently improves multi-character role-playing ability by balancing the learning of distinct and shared traits. Specifically, we propose a Hyper-Half Low-Rank Adaptation structure, where one half is a role-specific module generated by a lightweight hyper-network, and the other half is a trainable role-shared module. The role-specific module is devised to represent distinct persona signatures, while the role-shared module serves to capture common traits. Moreover, to better reflect distinct personalities across different roles, we design a hyper-contrastive learning mechanism to help the hyper-network distinguish their unique characteristics. Extensive experimental results on both English and Chinese available benchmarks demonstrate the superiority of our framework. Further GPT-4 evaluations and visual analyses also verify the capability of HyCoRA to capture role characteristics.

تقدم الورقة HyCoRA، وهو إطار تعلم تكيفي فائق التباين يهدف إلى تحسين لعب الأدوار المتعددة الشخصيات في نماذج الذكاء الاصطناعي. من خلال تحقيق التوازن بين تعلم السمات المميزة والمشتركة، تتناول HyCoRA القيود المفروضة على الأساليب الحالية التي إما تبسط أو تعقد تمثيل الأدوار. هذه الخطوة مهمة لتحسين قدرة الذكاء الاصطناعي على محاكاة شخصيات متنوعة، كما يتضح من النتائج التجريبية الواسعة والتقييمات باستخدام GPT-4.

El artículo presenta HyCoRA, un nuevo marco de aprendizaje adaptativo hiper-contrastivo destinado a mejorar el juego de roles multi-personaje en modelos de IA. Al equilibrar el aprendizaje de rasgos distintos y compartidos, HyCoRA aborda las limitaciones de los métodos existentes que simplifican o complican la representación de roles. Este avance es significativo para mejorar la capacidad de la IA para simular diversas personalidades, como lo demuestran los extensos resultados experimentales y las evaluaciones utilizando GPT-4.

L'article présente HyCoRA, un nouveau cadre d'apprentissage adaptatif hyper-contrasté visant à améliorer le jeu de rôle multi-personnages dans les modèles d'IA. En équilibrant l'apprentissage des traits distincts et partagés, HyCoRA répond aux limites des méthodes existantes qui simplifient ou compliquent la représentation des rôles. Cette avancée est significative pour améliorer la capacité de l'IA à simuler des personnalités diverses, comme en témoignent des résultats expérimentaux étendus et des évaluations utilisant GPT-4.

The paper introduces HyCoRA, a novel Hyper-Contrastive Role-Adaptive learning framework aimed at enhancing multi-character role-playing in AI models. By balancing the learning of distinct and shared traits, HyCoRA addresses limitations of existing methods that either oversimplify or complicate role representation. This advancement is significant for improving AI's ability to simulate diverse personalities, as evidenced by extensive experimental results and evaluations using GPT-4.

HyCoRA: Hyper-Contrastive Role-Adaptive Learning for Role-Playing

Was this article worth reading? Share it

Ready to build your own newsroom?