PsyMem: Fine-grained psychological alignment and Explicit Memory Control for Advanced Role-Playing LLMs

2025-05-19

AI Summary

Current role-playing LLMs suffer from two critical limitations: superficial character modeling—relying solely on surface-level textual descriptions—and insufficient memory consistency—employing implicit knowledge or weak retrieval augmentation without explicit alignment—thereby undermining reliability in high-stakes applications such as trustworthy social simulation. To address these, we propose PsyMem, the first framework to explicitly model characters using 26 fine-grained psychological dimensions and introduce a memory alignment training mechanism that enables dynamic, controllable memory retrieval during inference. Built upon Qwen2.5-7B-Instruct, PsyMem is trained via psychometric feature injection and memory consistency supervision on a novel, self-constructed novel-derived dataset comprising 5,414 characters and 38,962 dialogues. Experiments demonstrate that PsyMem-Qwen significantly outperforms baselines in character fidelity and anthropomorphism metrics, while exhibiting markedly improved reliability in trustworthy social simulation tasks.

Technology Category

• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Humans and AI: Game Design — Virtual Humans, NPCs and Autonomous Characters
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: Psychology-informed user models and recommender systems
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Existing LLM-based role-playing methods often rely on superficial textual descriptions or simplistic metrics, inadequately modeling both intrinsic and extrinsic character dimensions. Additionally, they typically simulate character memory with implicit model knowledge or basic retrieval augment generation without explicit memory alignment, compromising memory consistency. The two issues weaken reliability of role-playing LLMs in several applications, such as trustworthy social simulation. To address these limitations, we propose PsyMem, a novel framework integrating fine-grained psychological attributes and explicit memory control for role-playing. PsyMem supplements textual descriptions with 26 psychological indicators to detailed model character. Additionally, PsyMem implements memory alignment training, explicitly trains the model to align character's response with memory, thereby enabling dynamic memory-controlled responding during inference. By training Qwen2.5-7B-Instruct on our specially designed dataset (including 5,414 characters and 38,962 dialogues extracted from novels), the resulting model, termed as PsyMem-Qwen, outperforms baseline models in role-playing, achieving the best performance in human-likeness and character fidelity.

Problem

• Inadequate modeling of intrinsic and extrinsic character dimensions in role-playing LLMs
• Lack of explicit memory alignment compromising memory consistency
• Weak reliability in applications like trustworthy social simulation

Innovation

• Integrates 26 psychological indicators for character modeling
• Implements explicit memory alignment training
• Uses a custom dataset for dynamic memory-controlled responding