Elamur Research⌕
☰
home / papers / 2026 / MENTOR: Mitigating Identity Drift in Dynamic Role-Playing via Dual-Chain Structured Memory

MENTOR: Mitigating Identity Drift in Dynamic Role-Playing via Dual-Chain Structured Memory

Paper · 2026 · Zhu Zhuoning, Xingyu Gao and Hailong Shi

Summary

The paper addresses "identity drift": when one LLM agent serves several users or personas in a single long session, it starts mixing their states. It may apply one user's constraints to another, speak from the wrong persona, or leak private facts across roles. These failures often hide behind fluent, locally coherent replies. The authors call this a "pseudo-alignment" effect.

To measure it, they build BEAM-SWITCH on top of the BEAM narrative framework. The benchmark cuts long traces into six-round windows with controllable switching density and injects contradictory facts that belong to only one role.

Their method, MENTOR, needs no fine-tuning. Each role gets its own isolated working memory (Role Chain), a Global Chain logs session events, and a semantic Knowledge Graph checks which information the active role may use before the model answers.

Across six LLM families, the average score rises from 0.46 to 0.75. This shows that structured, role-aware memory beats plain similarity retrieval for keeping roles separated in multi-user deployments.

Key points

  • Benchmark design: BEAM-SWITCH uses windows of W = 6 rounds and targets dense switching (S ≥ 4, ρ ≥ 0.7). It adds adversarial conflicts between roles, for example "strictly vegan" vs. "loves steak".
  • Metrics: an LLM-as-a-Judge rates responses from 0 to 1 along three axes, abbreviated IA (for Identity Adherence), KF (for Knowledge Fidelity), and CC (for Contextual Coherence). Long-context models such as Kimi-K2 and LongCat keep CC high while IA and KF fall.
  • Main result: MENTOR is compared with Naive RAG and MemoChat baselines. It lifts the average score by +0.29 absolute (0.46 → 0.75). For long-context models, IA climbs from 0.28 to 0.72.
  • Ablation on Qwen3-30B: the full system scores Avg 0.63. Dropping the Role Chain cuts IA to 0.25 (Avg 0.43), and dropping the Knowledge Graph cuts KF to 0.28. Mixing 20% distractor turns into the Role Chain still leaves KF at 0.50.
  • Limitations: the approach adds per-turn latency and relies on the base model's ability to follow instructions. It is also evaluated only on a synthetic benchmark.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏛️Chinese Academy of SciencesBeijing · China246 works · 24,951 citations🎓University of Chinese Academy of SciencesBeijing · China188 works · 8,743 citations🔬Institute of MicroelectronicsBeijing · China2 works

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

Cite

Cite as Zhu Zhuoning, Xingyu Gao and Hailong Shi. 2026. MENTOR: Mitigating Identity Drift in Dynamic Role-Playing via Dual-Chain Structured Memory. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1046

@inproceedings{zhuoning2026mentor,
  title = {{MENTOR: Mitigating Identity Drift in Dynamic Role-Playing via Dual-Chain Structured Memory}},
  author = {Zhu Zhuoning, Xingyu Gao and Hailong Shi},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1046},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1046}
}

References

35 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. T. Abuelsaad, D. Akkil, P. Dey, A. Jagmohan, A. Vempaty, and R. Kokku. 2024. Agent-e: From autonomous web navigation to foundational design principles in agentic systems. arXiv preprint arXiv:2407.13032. — [arXiv:2407.13032](https://arxiv.org/abs/2407.13032) find
  2. Muhammad Arslan, Hussam Ghanem, Saba Munawar, and Christophe Cruz. 2024. A survey on rag with llms. Procedia computer science, 246:3781–3790. find
  3. Sanghwan Bae, Donghyun Kwak, Soyoung Kang, Min Young Lee, Sungdong Kim, Yuin Jeong, Hyeri Kim, Sang-Woo Lee, Woomyoung Park, and Nako Sung. 2022. Keep me updated! memory management in long-term conversations. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 3769–3787. find
  4. Jiangjie Chen, Xintao Wang, Rui Xu, Siyu Yuan, Yikai Zhang, Wei Shi, Jian Xie, Shuang Li, Ruihan Yang, Tinghui Zhu, Aili Chen, Nianqi Li, Lida Chen, Caiyu Hu, Siye Wu, Scott Ren, Ziquan Fu, and Yanghua Xiao. 2024a. From persona to personalization: A survey on role-playing language agents. Preprint,…
  5. Jiawei Chen, Xinyan Guan, Qianhao Yuan, Mo Guozhao, Weixiang Zhou, Yaojie Lu, Hongyu Lin, Ben He, Le Sun, and Xianpei Han. 2025. Consistentchat: Building skeleton-guided consistent multi-turn dialogues for large language models from scratch. In Proceedings of the 2025 Conference on Empirical Methods… find
  6. Nuo Chen, Hongguang Li, Juhua Huang, Baoyuan Wang, and Jia Li. 2024b. Compress to impress: Unleashing the potential of compressive memory in real-world long-term conversations. https://arxiv.org/abs/2402.11975. ArXiv preprint. ↗
  7. DeepSeek AI. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via pure rl. arXiv preprint. find
  8. X. Deng, Y. Gu, B. Zheng, S. Chen, S. Stevens, B. Wang, H. Sun, and Y. Su. 2023. Mind2web: Towards a generalist agent for the web. In Advances in Neural Information Processing Systems, volume 36, pages 28091–28114.
  9. H. He, W. Yao, K. Ma, W. Yu, Y. Dai, H. Zhang, Z. Lan, and D. Yu. 2024. Webvoyager: Building an end-to-end web agent with large multimodal models. arXiv preprint arXiv:2401.13919.
  10. Bowen Jin, Jinsung Yoon, Jiawei Han, and Sercan O Arik. 2024. Long-context llms meet rag: Overcoming challenges for long inputs in rag. arXiv preprint arXiv:2410.05983. — [arXiv:2410.05983](https://arxiv.org/abs/2410.05983) find
  11. Jiazheng Kang, Mingming Ji, Zhe Zhao, and Ting Bai. 2025. Memory os of ai agent. arXiv preprint arXiv:2506.06326.
  12. Hana Kim, Kai Ong, Seoyeon Kim, Dongha Lee, and Jinyoung Yeo. 2024. Commonsense-augmented memory construction and management in long-term conversations via context-aware persona refinement. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics… find
  13. Hao Li, Chenghao Yang, An Zhang, Yang Deng, Xiang Wang, and Tat-Seng Chua. 2024a. Hello again! llm-powered personalized agent for long-term dialogue. https://arxiv.org/abs/2406.05925. ArXiv preprint. ↗
  14. Yuanchun Li, Hao Wen, Weijun Wang, Xiangyu Li, Yizhen Yuan, Guohong Liu, Jiacheng Liu, Wenxing Xu, Xiang Wang, Yi Sun, Rui Kong, Yile Wang, Hanfei Geng, Jian Luan, Xuefeng Jin, Zilong Ye, Guanjing Xiong, Fan Zhang, Xiang Li, and 6 others. 2024b. Personal llm agents: Insights and survey about the cap…
  15. Yucheng Li, Huiqiang Jiang, Qianhui Wu, Xufang Luo, Surin Ahn, Chengruidong Zhang, Amir H. Abdi, Dongsheng Li, Jianfeng Gao, Yuqing Yang, and Lili Qiu. 2025. Scbench: A kv cache-centric analysis of long-context methods. Preprint, arXiv:2412.10319. — [arXiv:2412.10319](https://arxiv.org/abs/2412.10319) find
  16. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024a. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  17. Nelson F Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2024b. Lost in the middle: How language models use long contexts. Transactions of the Association for Computational Linguistics, 12:157–173.
  18. Junru Lu, Siyu An, Mingbao Lin, Gabriele Pergola, Yulan He, Di Yin, Xing Sun, and Yunsheng Wu. 2023. Memochat: Tuning llms to use memos for consistent long-range open-domain conversation. Preprint, arXiv:2308.08239. — [arXiv:2308.08239](https://arxiv.org/abs/2308.08239) find
  19. Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, and Yuwei Fang. 2024. Evaluating very long-term conversational memory of llm agents. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13851– 1387… find
  20. Charles Packer, Vivian Fang, Shishir_G Patil, Kevin Lin, Sarah Wooders, and Joseph_E Gonzalez. 2023. Memgpt: Towards llms as operating systems. find
  21. Zhuoshi Pan, Qianhui Wu, Huiqiang Jiang, Xufang Luo, Hao Cheng, Dongsheng Li, Yuqing Yang, Chin-Yew Lin, H Vicky Zhao, Lili Qiu, and 1 others. 2025. Secom: On memory construction and retrieval for personalized conversational agents. In The Thirteenth International Conference on Learning Representati… find
  22. Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, pages 1–22. find
  23. Kurt Shuster, Jack Urbanek, Arthur Szlam, and Jason Weston. 2022. Am i me or you? state-of-the-art dialogue models cannot maintain an identity. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 2367–2387. find
  24. Mohammad Tavakoli, Alireza Salemi, Carrie Ye, Mohamed Abdalla, Hamed Zamani, and J Ross Mitchell. 2025. Beyond a million tokens: Benchmarking and enhancing long-term memory in llms. arXiv preprint arXiv:2510.27246. — [arXiv:2510.27246](https://arxiv.org/abs/2510.27246) find
  25. Kimi Team, Yifan Bai, Yiping Bao, Guanduo Chen, Jiahao Chen, Ningxin Chen, Ruijue Chen, Yanru Chen, Yuankun Chen, Yutian Chen, and 1 others. 2025a. Kimi k2: Open agentic intelligence. arXiv preprint arXiv:2507.20534.
  26. Meituan LongCat Team, Bei Li, Bingye Lei, Bo Wang, Bolin Rong, Chao Wang, Chao Zhang, Chen Gao, Chen Zhang, Cheng Sun, and 1 others. 2025b. Longcat-flash technical report. arXiv preprint arXiv:2509.01322. — [arXiv:2509.01322](https://arxiv.org/abs/2509.01322) find
  27. Yu-Min Tseng, Yu-Chao Huang, Teng-Yun Hsiao, Yu-Ching Hsu, Jia-Yin Foo, Chao-Wei Huang, and Yun-Nung Chen. 2024. Two tales of persona in llms: A survey of role-playing and personalization. https://arxiv.org/abs/2406.01171. ArXiv preprint. ↗
  28. Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, and 1 others. 2024a. A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6):186345.
  29. Piaohong Wang, Motong Tian, Jiaxian Li, Yuan Liang, Yuqing Wang, Qianben Chen, Tiannan Wang, Zhicong Lu, Jiawei Ma, Yuchen Eleanor Jiang, and 1 others. 2025a. O-mem: Omni memory system for personalized, long horizon, self-evolving agents. arXiv preprint arXiv:2511.13593. — [arXiv:2511.13593](https://arxiv.org/abs/2511.13593) find
  30. Xiaohan Wang, Yuhui Zhang, Orr Zohar, and Serena Yeung-Levy. 2024b. Videoagent: Long-form video understanding with large language model as agent. In European Conference on Computer Vision, pages 58–76. Springer. find
  31. Zixuan Wang, Bo Yu, Junzhe Zhao, Wenhao Sun, Sai Hou, Shuai Liang, Xing Hu, Yinhe Han, and Yiming Gan. 2025b. Karma: Augmenting embodied ai agents with long-and-short term memory systems. In 2025 IEEE International Conference on Robotics and Automation (ICRA), pages 1–8. IEEE. find
  32. Xinchao Xu, Zhibin Gou, Wenquan Wu, Zheng-Yu Niu, Hua Wu, Haifeng Wang, and Shihang Wang. 2022. Long time no see! open-domain conversation with long-term persona memory. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2639–2650. find
  33. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.
  34. Zeyu Zhang, Xiaohe Bo, Chen Ma, Rui Li, Xu Chen, Quanyu Dai, Jieming Zhu, Zhenhua Dong, and Ji-Rong Wen. 2024. A survey on the memory mechanism of large language model based agents. arXiv preprint arXiv:2404.13501. find
  35. Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. 2024. Memorybank: Enhancing large language models with long-term memory. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19724–19731. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhuoning-2026-mentor-mitigating-identity-drift-dynamic/