Xilong Cheng

Xilong Cheng is a third-year Master’s student at the School of Information and Communication Engineering, Communication University of China, advised by Yunxiao Qin. He has a strong interest in streaming media narratives, cognitive psychology, and behavioral psychology. His research focuses on leveraging AI technologies to build more generalizable role-playing systems, aiming to construct highly realistic social simulation environments.

His current research interests center on anthropomorphic generative agents, character personas, and social simulation, including but not limited to:

1. Role-playing agents: Exploring the role-playing capabilities of large language models (LLMs) with the goal of developing generalizable role-playing systems. This includes: (1) accurate and quantitative evaluation of LLM role-playing abilities; (2) constructing high-quality character datasets based on existing narrative materials; (3) advancing role-playing capabilities from text-based to multimodal foundation models; and (4) building realistic social simulation environments based on generalized role-playing systems.
2. Psychology in language models: Enhancing the anthropomorphic capacity of LLMs from a psychological perspective (e.g., self-awareness, personality, memory). The goal is to enable LLMs not only to perform well when modeling well-known characters but also to generalize to modeling and representing ordinary individual roles.
3. More realistic social simulation systems: Investigating social simulation systems grounded in role-playing agents, with an emphasis on equipping models with both “System 1” (intuitive decision-making) and “System 2” (deliberative planning) capabilities, thereby enabling more authentic simulation of social behaviors and dynamic interactions.

In parallel, he is also working on human-inspired self-supervised learning for long-form first-person videos, aiming to develop richer video representations that capture temporal dynamics and embodied perspectives.

Education

• 🎓 M.S. in Information & Communication Engineering, 2023–2026 (estimated)
  Communication University of China · Advisor: Yunxiao Qin
• 🎓 B.S. in Information & Communication Engineering, 2018–2022
  Communication University of China

Publications

1. PsyMem: Fine-grained psychological alignment and Explicit Memory Control for Advanced Role-Playing LLMs
   Xilong Cheng^★, Yunxiao Qin^★,†, Yuting Tan^★, Zhengnan Li, Ye Wang, Hongjiang Xiao^†, Yuan Zhang
   Transactions of the Association for Computational Linguistics

2. Towards Stable Self-Supervised Object Representations in Unconstrained Egocentric Video
   Yuting Tan^★, Xilong Cheng^★, Yunxiao Qin, Zhengnan Li, Jingjing Zhang^†

3. Structured Guidance or Open-Ended Generation? Boundaries of Co-Creative AI for Screenwriting
   Ruihan Dong^★, Xilong Cheng^★,§, Qi Zhou^★, Yuting Tan, Yuhao Zhou, Jingbo Ji, Yunxiao Qin^†

4. FSMLP: Modelling Channel Dependencies With Simplex Theory Based Multi-Layer Perceptions In Frequency Domain
   Zhengnan Li, Haoxuan Li, Hao Wang, Jun Fang, Zhichao Chen, Yuting Tan, Xilong Cheng, Yunxiao Qin

5. Meta-Tuner: Meta-Trained Node-Specific Transformations for Graph Few-Shot Class-Incremental Learning
   Zhengnan Li, Jun Fang, Junbo Wang, Xilong Cheng, Yuting Tan, Yunxiao Qin
   Expert Systems with Applications

6. FTMixer: Frequency and Time Domain Representations Fusion for Time Series Forecasting
   Zhengnan Li, Yuting Tan, Xilong Cheng, Yunxiao Qin
   IEEE Signal Processing Letters

_{^★ Equal contribution · ^† Corresponding author · ^§ Project leader}

Interests

Beyond my academic focus on Large Language Models and Role-Playing Language Agents, I also have a strong passion for sports, particularly football and, more recently, tennis. While I see sports as an essential form of relaxation, I equally enjoy streaming media and reading literature on modern psychology, which help me unwind and broaden my perspective.