CRPO: Character-centric Group Relative Policy Optimization for Role-aware Reasoning in Role-playing Agents
🌟 About Me
Since March 2025, I have been pursuing my Ph.D. degree in Computer Technology at Harbin Institute of Technology, Shenzhen (HITSZ), supervised by Prof. Kehai Chen. Since September 2025, I have also been conducting research at the Shenzhen Loop Area Institute (SLAI). Prior to this, I received my Master's degree from Tianjin University, under the supervision of Prof. Bo Wang.
My primary research interests center on Natural Language Processing (NLP), Large Language Models (LLMs), Agents, and Role-Playing. I am dedicated to exploring the potential of LLMs in simulating human behaviors and interactions. Specifically, I focus on building Role-Playing Agents with enhanced reasoning capabilities and personalization, while investigating their robustness in complex interactive scenarios. My goal is to evolve agents from mere tools into more human-like interactive partners.
I am always open to academic discussions and collaborations. Please feel free to reach out!
🎓 Education
- 2025.09 - Present: Research training in Computer Technology, Shenzhen Loop Area Institute, Shenzhen, China. Advisor: Prof. Min Zhang.
- 2025.03 - Present: Ph.D. in Computer Technology, Harbin Institute of Technology (Shenzhen), Shenzhen, China. Advisor: Prof. Kehai Chen.
- 2022.06 - 2025.01, M.Eng. in Computer Technology, Tianjin University, Tianjin, China. Advisor: Prof. Bo Wang.
- 2018.09 - 2022.06, B.Eng. in Computer Science and Technology, Northeastern University at Qinhuangdao ,Qinhuangdao, China.
🧭 Research Overview
Publication-driven constellation
Human-like Agents
Auto-built from titles, abstracts, and keywords
Human-like Agents
Goal: turn LLM agents from generic tools into personalized, reliable interactive partners.
Role & persona Reasoning Safety Dialogue Evaluation & applications
📚 Publication (* indicates co-first authors)
Selected Publications
5 papersThinking in Character: Advancing Role-Playing Agents with Role-Aware Reasoning
RoleBreak: Character Hallucination as a Jailbreak Attack in Role-Playing Systems
MORPHEUS: Modeling Role from Personalized Dialogue History by Exploring and Utilizing Latent Space
Enhancing Personalized Dialogue Generation with Contrastive Latent Variables: Combining Sparse and Dense Persona
Other Publications
8 papersREG: Retrieval via Emotion Similarity for Guiding Empathetic Dialogue Generation
QueryAligner: Customizing User Query to Match LLMs Preferences for Better Intent Recognition
MASTER: Enhancing Large Language Model via Multi-Agent Simulated Teaching
ORPP: Self-Optimizing Role-playing Prompts to Enhance Language Model Capabilities
Exploiting Prompt-Induced Confidence for Black-Box Attacks on LLMs
The Rise of Darkness: Safety-Utility Trade-Offs in Role-Playing Dialogue Agents
ECC: Synergizing Emotion, Cause and Commonsense for Empathetic Dialogue Generation
DialogBench: Evaluating LLMs as Human-like Dialogue Systems
Preprints
4 papersCharacter-R1: Enhancing Role-Aware Reasoning in Role-Playing Agents via RLVR
Through the Lens of Character: Resolving Modality-Role Interference in Multimodal Role-Playing Agent
Empowering Real-World: A Survey on the Technology, Practice, and Evaluation of LLM-driven Industry Agents
The Power of Personality: A Human Simulation Perspective to Investigate Large Language Model Agents
🔬 Research Contribution
- Chinese-medical-dialogue-data: High-quality Chinese medical question-answer dataset.
- Emotion-Enhanced Agent Based on Role-Playing, 2025/01 - 2025/12, CCF-Baidu Songguo Fund, Participant (Student Lead)
- Reviewer: ACL ARR