Beyond Raw Transcripts: Structured Persona Extraction for LLM-Based Digital Twins

超越原始文本：基于大语言模型的数字孪生结构化人格提取

Abstract: LLM-based “digital twins” aim to simulate how an individual would behave in new environments or respond to novel questions, given some representation of that individual’s prior responses. A common approach constructs this representation from survey transcripts or summaries responses. Prior work shows that compressing long transcripts into shorter LLM-generated summaries does not significantly reduce predictive accuracy, suggesting that information volume is not the primary bottleneck.

摘要： 基于大语言模型（LLM）的“数字孪生”旨在通过个体过往的响应表征，模拟其在特定新环境下的行为或对新问题的反应。目前常见的方法是利用调查访谈记录或总结性回复来构建这种表征。既往研究表明，将长篇访谈记录压缩为简短的 LLM 生成摘要并不会显著降低预测准确性，这表明信息量并非主要的瓶颈。

In this work, we argue that the key limitation is instead structural: how persona information is organized before being provided to the simulator model. We study this by comparing unstructured summaries with structured persona representations. First, we introduce a hand-crafted schema (BDE: Background, Decision procedure, Evaluation), grounded in consumer-behavior theory, and show that it improves predictive accuracy over raw transcripts by +1.91 percentage points on a homogeneous benchmark (Twin-2K-500), with similar gains on gpt-5.4-mini and Qwen3-8B as robustness checks.

在这项工作中，我们认为核心限制在于结构问题：即在将人格信息提供给模拟模型之前，这些信息是如何组织的。我们通过对比非结构化摘要与结构化人格表征来研究这一问题。首先，我们引入了一种基于消费者行为理论的手工架构（BDE：背景、决策程序、评估），并证明在同质化基准测试（Twin-2K-500）中，该架构比原始文本的预测准确率提高了 1.91 个百分点；在 gpt-5.4-mini 和 Qwen3-8B 上的稳健性检查也显示了类似的提升。

However, this fixed structure does not generalize across more heterogeneous tasks, where performance is statistically indistinguishable from the raw transcript baseline. To address this limitation, we propose an automatic structure-discovery pipeline in which an LLM iteratively proposes and refines task-specific persona structures and extraction prompts. On a benchmark of 13 diverse sub-studies, this approach restores performance, improving mean accuracy by +1.91 percentage points over the raw transcript baseline and eliminating significant losses observed with the fixed schema.

然而，这种固定结构无法推广到更多异构任务中，在这些任务中，其表现与原始文本基准在统计学上没有显著差异。为了解决这一局限性，我们提出了一种自动结构发现流程，通过该流程，LLM 可以迭代地提出并优化针对特定任务的人格结构及提取提示词。在包含 13 项不同子研究的基准测试中，该方法恢复了性能，平均准确率较原始文本基准提高了 1.91 个百分点，并消除了固定架构中观察到的显著性能损失。

Overall, our results suggest that the main constraint in LLM-based digital twins is not how much information is provided, but how it is structured — and that the optimal structure depends on the task.

总而言之，我们的研究结果表明，基于 LLM 的数字孪生的主要制约因素不在于提供了多少信息，而在于信息是如何被结构化的——且最优结构取决于具体任务。