arXiv cs.CL· Mumin Jia, Yilin Chen, Divya Sharma, Jairo Diaz-Rodriguez·· 3 天前AI 评分33
LLM 数字人能否可靠近似人类调查结果？基于 LISS 面板的研究
When Can Digital Personas Reliably Approximate Human Survey Findings?
AI 导读
研究基于 LISS 面板构建 LLM 数字人，用受访者背景变量和 2023 年前的调查历史预测其被留出的后续回答，覆盖四种 persona 架构、三个 LLM 和两项预测任务。数字人能改善与人类回答分布的对齐，尤其在稳定属性与价值观相关领域，但在个体预测上仍受限，且无法还原受访者的多变量结构。检索增强架构增益最明显，表现更多取决于人类回答结构而非模型选择。
正文 · 原文
Abstract:Digital personas powered by Large Language Models (LLMs) are increasingly proposed as substitutes for human survey respondents, yet it remains unclear when they can reliably approximate human survey findings. We answer this question using the LISS panel, constructing personas from respondents' background variables and pre-2023 survey histories, then testing them against the same respondents' held-out post-cutoff answers. Across four persona architectures, three LLMs, and two prediction tasks, we assess performance at the question, respondent, distributional, equity, and clustering levels. Digital personas improve alignment with human response distributions, especially in domains tied to stable attributes and values, but remain limited for individual prediction and fail to recover multivariate respondent structure. Retrieval-augmented architectures provide the clearest gains, but performance depends more on human response structure than on model choice: personas perform best for low-variability questions and common respondent patterns, and worst for subjective, heterogeneous, or rare responses. Our results provide practical guidance on when digital personas could be appropriate for survey research and when human validation remains necessary.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Social and Information Networks (cs.SI); Machine Learning (stat.ML) |
| Cite as: | arXiv:2605.10659 [cs.CL] |
| (or arXiv:2605.10659v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.10659 arXiv-issued DOI via DataCite |
Submission history
From: Mumin Jia [view email]
[v1]
Mon, 11 May 2026 14:41:11 UTC (6,334 KB)
[v2]
Mon, 5 Oct 2026 01:29:11 UTC (6,349 KB)
[v3]
Tue, 13 Oct 2026 20:12:03 UTC (6,343 KB)
来源：arXiv cs.CL · arxiv.org