Learning User Representations to Improve Simulation Fidelity

Computer Science > Computation and Language arXiv:2610.22518 (cs) [Submitted on 15 Oct 2026] Title: Learning User Representations to Improve Simulation Fidelity Authors: Shuhaib Mehri, Philippe Laban, Michel Galley, Dilek Tür, Jennifer Neville, Jianfeng Gao.

计算机科学 > 计算与语言 arXiv:2610.22518 (cs) [提交于 2026 年 10 月 15 日] 标题：学习用户表示以提高模拟保真度 作者：Shuhaib Mehri, Philippe Laban, Michel Galley, Dilek Tür, Jennifer Neville, Jianfeng Gao。

Abstract: User simulators enable interactive training and evaluation of AI assistants at scale. Yet current simulators struggle to reproduce the diverse behaviors of real users, leading to assistants that generalize poorly to underrepresented users. We identify user representations as an underexplored bottleneck in user simulation.

摘要：用户模拟器能够实现人工智能助手的规模化交互式训练与评估。然而，当前的模拟器难以重现真实用户多样化的行为，导致助手在面对代表性不足的用户时泛化能力较差。我们将用户表示识别为用户模拟中一个尚未被充分探索的瓶颈。

User representations are natural-language descriptions that condition simulators to reproduce a particular user’s behavior. A representation’s quality becomes apparent in hindsight, through the behavior it induces in a simulator. To evaluate and optimize user representations, we formulate a conversation reconstruction task: given an original conversation, we generate a representation, condition a simulator on it, simulate a conversation, and then measure fidelity by comparing user behavior in the original and simulated conversations.

用户表示是自然语言描述，用于调节模拟器以重现特定用户的行为。表示的质量在事后通过其在模拟器中诱导出的行为变得显而易见。为了评估和优化用户表示，我们制定了一项对话重构任务：给定一段原始对话，我们生成一个表示，将其作为模拟器的条件，模拟一段对话，然后通过比较原始对话和模拟对话中的用户行为来衡量保真度。

This comparison also yields hindsight feedback, which identifies where the representation falls short and how it can be improved. We introduce hindsight distillation, an online training procedure that distills feedback-conditioned improvements into the representation generator, enabling it to generate improved representations directly from the original conversation.

这种比较还会产生事后反馈，从而识别出表示的不足之处以及改进方法。我们引入了事后蒸馏（hindsight distillation），这是一种在线训练程序，将基于反馈的改进蒸馏到表示生成器中，使其能够直接从原始对话中生成改进后的表示。

We present UserRepGen-4B, a representation generator trained with hindsight distillation. Across 700 conversations spanning seven datasets, hindsight feedback helps improve user representations at test time, and UserRepGen-4B generates representations that enable more faithful simulation than those generated by GPT-5.5.

我们提出了 UserRepGen-4B，这是一个通过事后蒸馏训练的表示生成器。在涵盖七个数据集的 700 段对话中，事后反馈有助于在测试时改进用户表示，且 UserRepGen-4B 生成的表示比 GPT-5.5 生成的表示能实现更忠实的模拟。

Beyond conversation reconstruction, learned representations improve simulation fidelity for unseen tasks, better preserve the population-level diversity of real users, and better distinguish users across conversations.

除了对话重构之外，学习到的表示还提高了未见任务的模拟保真度，更好地保留了真实用户的群体级多样性，并能更好地在不同对话中区分用户。