aiqbench analysis
This paper tackles a practical bottleneck in personalized reward modeling: per-user preference labels are expensive, while large pools of unlabeled responses go unused. Its method, pair-mean score centering, aims to make pseudo-labels comparable across prompts for the same user, and the authors report top mean accuracy against seven baselines on four benchmarks. These are self-reported results, so the gains need independent replication, but the core idea of using unlabeled data for personalization is useful for anyone building preference-tuned models.
PerSPL: Personalized Semi-supervised Preference Learning
arXiv:2610.29219v1 Announce Type: new Abstract: Personalized reward models require user-specific preference comparisons, but collecting such feedback is costly and provides only limited supervision for each user. At the same time, large pools of responses can be generated without user identities or preference labels, yet existing personalized reward models leave this data unused. Directly learning from these responses is difficult because pairwise objectives constrain relative scores within a prompt, whereas pseudo-labeling individual responses requires scores that are comparable across prompts for the same user. We propose \textbf{PerSPL} (Personalized Semi-supervised Preference Learning), a backbone-agnostic framework that uses pair-mean score centering to stabilize this cross-prompt score coordinate and iteratively turns unlabeled responses into user-specific training signals. Against seven representative baselines across four benchmarks---PersonalLLM, HelpSteer2, Reddit TLDR, and Chatbot Arena---PerSPL achieves the highest reported mean accuracy among the evaluated methods for both seen and unseen users. The two latter benchmarks provide real-user preference comparisons in summarization and open-domain chatbot interaction, respectively. These results indicate that unlabeled responses can provide additional supervision for personalized reward modeling when per-user preference labels are scarce.
Abstract:Personalized reward models require user-specific preference comparisons, but collecting such feedback is costly and provides only limited supervision for each user. At the same time, large pools of responses can be generated without user identities or preference labels, yet existing personalized reward models leave this data unused. Directly learning from these responses is difficult because pairwise objectives constrain relative scores within a prompt, whereas pseudo-labeling individual responses requires scores that are comparable across prompts for the same user. We propose \textbf{PerSPL} (Personalized Semi-supervised Preference Learning), a backbone-agnostic framework that uses pair-mean score centering to stabilize this cross-prompt score coordinate and iteratively turns unlabeled responses into user-specific training signals. Against seven representative baselines across four benchmarks---PersonalLLM, HelpSteer2, Reddit TLDR, and Chatbot Arena---PerSPL achieves the highest reported mean accuracy among the evaluated methods for both seen and unseen users. The two latter benchmarks provide real-user preference comparisons in summarization and open-domain chatbot interaction, respectively. These results indicate that unlabeled responses can provide additional supervision for personalized reward modeling when per-user preference labels are scarce.
Machine Learning (cs.LG)
arXiv:2610.29219 [cs.LG]
(or arXiv:2610.29219v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.29219 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 266 words