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

2025-11-11

AI Summary

Existing methods for multi-role playing struggle to balance role-specific individuality and cross-role commonality: shared-parameter approaches overlook inter-role differences, while independent-parameter strategies neglect shared semantic patterns. To address this, we propose Hyper-Half—a lightweight hypernetwork-based framework that generates role-specific low-rank adaptation modules while coupling them with a trainable shared backbone, establishing a dual-path modeling architecture for both role-shared and role-specific representation learning. Furthermore, we introduce hyper-contrastive learning to explicitly pull together embeddings of semantically similar roles and push apart those of dissimilar ones. Evaluated on bilingual (Chinese and English) multi-role benchmarks, Hyper-Half achieves significant improvements over state-of-the-art methods. Human evaluation via GPT-4 and visualization analysis further confirm its superior accuracy and robustness in capturing both role个性 and role commonality.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Knowledge Representation and Reasoning: Common-Sense Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

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.

Problem

• Existing role-playing methods fail to balance distinct and shared character traits
• Role-shared modules ignore unique personalities while role-specific overlook commonalities
• Need improved framework for multi-character role-playing with adaptive learning

Innovation

• Hyper-Half Low-Rank Adaptation structure balances role-specific and shared modules
• Lightweight hyper-network generates role-specific modules for distinct personas
• Hyper-contrastive learning mechanism enhances unique role characteristic differentiation