Elamur Research⌕
☰
home / papers / 2025 / PROPER: A Progressive Learning Framework for Personalized Large Language Models with Group-Level Adaptation

PROPER: A Progressive Learning Framework for Personalized Large Language Models with Group-Level Adaptation

Paper · 2025 · Linhai Zhang, Jialong Wu, Deyu Zhou et al.

Summary

Personalizing large language models by fine-tuning separate parameters for each user is limited because most users supply very little data. The authors propose PROPER, a framework built on the social-science idea of a meso level. It places a group-level model between the population-wide model and individual user models. Personalization becomes a sequence of residual refinements, trained in three LoRA-based stages:

  1. A task-level LoRA is learned from all data and merged into the backbone.
  2. A LoRAMoE layer is trained, where each expert represents a user group. A user-aware router combines text and learnable user embeddings to place users into groups automatically.
  3. A LoRA is trained for each user and combined with the group experts through a LoRA-aware router.

Each stage freezes the parameters learned before it. Instead of a standard load-balancing loss, a constraint loss pushes different users toward distinct expert weightings. On the LaMP benchmark, PROPER outperforms earlier fine-tuning-based personalization methods on every task. This suggests that knowledge shared within a group can compensate for sparse individual histories.

Key points

  • The motivation is data sparsity in LaMP. Task-adaptation data averages over 1,000k tokens, while each user has only about 48k. The top 10% of users produce roughly 85% of the data.
  • Training proceeds in three stages: population, group, then user. Each stage learns only the residual that earlier frozen stages miss, and LoRA is applied to the FFN layers.
  • Group modeling uses LoRAMoE, where each expert corresponds to one user group. The user-aware router sums softmax scores from a text router and a user-embedding router.
  • A similarity-penalizing constraint loss, inspired by P-Tailor, replaces the usual expert-balancing loss so that groups stay differentiated.
  • The main baselines are OPPU and PER-PCS. The authors list three limitations: evaluation covers single tasks only, user preferences are assumed static, and training is costlier. Code is available at github.com/callanwu/PROPER.

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

🔬The Alan Turing InstituteLondon · United Kingdom29 works · 4,368 citations🎓King's College LondonLondon · United Kingdom42 works · 1,617 citations🎓Southeast UniversityNanjing · China28 works · 354 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksInformation Systems · 1,437 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Linhai Zhang, Jialong Wu, Deyu Zhou and Yulan He. 2025. PROPER: A Progressive Learning Framework for Personalized Large Language Models with Group-Level Adaptation. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.acl-long.800

@inproceedings{zhang2025proper,
  title = {{PROPER: A Progressive Learning Framework for Personalized Large Language Models with Group-Level Adaptation}},
  author = {Linhai Zhang, Jialong Wu, Deyu Zhou and Yulan He},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2025},
  doi = {10.18653/v1/2025.acl-long.800},
  url = {https://doi.org/10.18653/v1/2025.acl-long.800}
}

References

33 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jürgen Backhaus. 1980. The pareto principle. Analyse & Kritik, 2(2):146–171. find
  2. Jin Chen, Zheng Liu, Xu Huang, Chenwang Wu, Qi Liu, Gangwei Jiang, Yuanhao Pu, Yuxuan Lei, Xiaolong Chen, Xingmei Wang, et al. 2024. When large language models meet personalization: Perspectives of challenges and opportunities. World Wide Web, 27(4):42. find
  3. Sunhao Dai, Ninglu Shao, Haiyuan Zhao, Weijie Yu, Zihua Si, Chen Xu, Zhongxiang Sun, Xiao Zhang, and Jun Xu. 2023. Uncovering chatgpt’s capabilities in recommender systems. In Proceedings of the 17th ACM Conference on Recommender Systems, pages 1126–1132. find
  4. Yuhao Dan, Jie Zhou, Qin Chen, Junfeng Tian, and Liang He. 2024. P-tailor: Customizing personality traits for language models via mixture of specialized lora experts. arXiv preprint arXiv:2406.12548. — [arXiv:2406.12548](https://arxiv.org/abs/2406.12548) find
  5. Shihan Dou, Enyu Zhou, Yan Liu, Songyang Gao, Wei Shen, Limao Xiong, Yuhao Zhou, Xiao Wang, Zhiheng Xi, Xiaoran Fan, Shiliang Pu, Jiang Zhu, Rui Zheng, Tao Gui, Qi Zhang, and Xuanjing Huang. 2024. LoRAMoE: Alleviating world knowledge forgetting in large language models via MoE-style plugin. In Proce…
  6. Nan Du, Yanping Huang, Andrew M Dai, Simon Tong, Dmitry Lepikhin, Yuanzhong Xu, Maxim Krikun, Yanqi Zhou, Adams Wei Yu, Orhan Firat, et al. 2022. Glam: Efficient scaling of language models with mixture-of-experts. In International Conference on Machine Learning, pages 5547–5569. PMLR.
  7. Thomas Faist. 2021. The crucial meso-level. In International migration, immobility and development, pages 187–217. Routledge. find
  8. Haytham M Fayek, Lawrence Cavedon, and Hong Ren Wu. 2020. Progressive learning: A deep learning framework for continual learning. Neural Networks, 128:345–357. find
  9. Gary Alan Fine. 2012. Group culture and the interaction order: Local sociology on the meso-level. Annual Review of Sociology, 38(1):159–179. find
  10. Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations.
  11. Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. 1991. Adaptive mixtures of local experts. Neural computation, 3(1):79–87. find
  12. Wang-Cheng Kang, Jianmo Ni, Nikhil Mehta, Maheswaran Sathiamoorthy, Lichan Hong, Ed Chi, and Derek Zhiyuan Cheng. 2023. Do llms understand user preferences? evaluating llms on user rating prediction. arXiv preprint arXiv:2305.06474. — [arXiv:2305.06474](https://arxiv.org/abs/2305.06474) find
  13. Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. 2021. {GS}hard: Scaling giant models with conditional computation and automatic sharding. In International Conference on Learning Representations. find
  14. Qidong Liu, Xian Wu, Xiangyu Zhao, Yuanshao Zhu, Derong Xu, Feng Tian, and Yefeng Zheng. 2024. When moe meets llms: Parameter efficient fine-tuning for multi-task medical applications. SIGIR ’24, page 1104–1114, New York, NY, USA. Association for Computing Machinery. find
  15. Ilya Loshchilov and Frank Hutter. 2018. Decoupled weight decay regularization. In International Conference on Learning Representations. find
  16. Tongxu Luo, Jiahe Lei, Fangyu Lei, Weihao Liu, Shizhu He, Jun Zhao, and Kang Liu. 2024. Moelora: Contrastive learning guided mixture of experts on parameter-efficient fine-tuning for large language models. arXiv preprint arXiv:2402.12851. — [arXiv:2402.12851](https://arxiv.org/abs/2402.12851) find
  17. William McConnell, Emilio F Moran, et al. 2002. Meeting in the middle: the challenge of meso-level integration. Land Use Policy, 19(1):99–101. find
  18. Sheshera Mysore, Zhuoran Lu, Mengting Wan, Longqi Yang, Bahareh Sarrafzadeh, Steve Menezes, Tina Baghaee, Emmanuel Barajas Gonzalez, Jennifer Neville, and Tara Safavi. 2024. Pearl: Personalizing large language model writing assistants with generation-calibrated retrievers. In Proceedings of the 1st… find
  19. Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence embeddings using Siamese BERT-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992,…
  20. Chris Richardson, Yao Zhang, Kellen Gillespie, Sudipta Kar, Arshdeep Singh, Zeynab Raeesy, Omar Zia Khan, and Abhinav Sethy. 2023. Integrating summarization and retrieval for enhanced personalization via large language models. arXiv preprint arXiv:2310.20081. — [arXiv:2310.20081](https://arxiv.org/abs/2310.20081) find
  21. Carlos Riquelme, Joan Puigcerver, Basil Mustafa, Maxim Neumann, Rodolphe Jenatton, André Susano Pinto, Daniel Keysers, and Neil Houlsby. 2021. Scaling vision with sparse mixture of experts. Advances in Neural Information Processing Systems, 34:8583–8595. find
  22. Alireza Salemi, Sheshera Mysore, Michael Bendersky, and Hamed Zamani. 2024. LaMP: When large language models meet personalization. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7370–7392, Bangkok, Thailand. Association for C…
  23. Noam Shazeer. 2020. Glu variants improve transformer. arXiv preprint arXiv:2002.05202. find
  24. Noam Shazeer, *Azalia Mirhoseini, *Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. 2017. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. In International Conference on Learning Representations. find
  25. Zhaoxuan Tan, Zheyuan Liu, and Meng Jiang. 2024a. Personalized pieces: Efficient personalized large language models through collaborative efforts. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6459– 6475, Miami, Florida, USA. Association for Computa…
  26. Zhaoxuan Tan, Qingkai Zeng, Yijun Tian, Zheyuan Liu, Bing Yin, and Meng Jiang. 2024b. Democratizing large language models via personalized parameter-efficient fine-tuning. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6476–6491, Miami, Florida, USA.…
  27. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288.
  28. Laurens Van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-sne. Journal of machine learning research, 9(11). find
  29. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lho… — Transformers: State-of-the-Art Natural Language Processing
  30. Joanne V Wood. 1989. Theory and research concerning social comparisons of personal attributes. Psychological bulletin, 106(2):231. find
  31. Linhai Zhang, Jialong Wu, Deyu Zhou, and Guoqiang Xu. 2024. STAR: Constraint LoRA with dynamic active learning for data-efficient fine-tuning of large language models. In Findings of the Association for Computational Linguistics: ACL 2024, pages 3519– 3532, Bangkok, Thailand. Association for Computa… find
  32. Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, et al. 2023. A survey of large language models. arXiv preprint arXiv:2303.18223.
  33. Yuchen Zhuang, Haotian Sun, Yue Yu, Rushi Qiang, Qifan Wang, Chao Zhang, and Bo Dai. 2024. HYDRA: Model factorization framework for black-box LLM personalization. In The Thirty-eighth Annual Conference on Neural Information Processing Systems. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2025-proper-progressive-learning-framework-personalized/