发表机构
School of Data Science, Fudan University; Shanghai Innovation Institute; Tencent（复旦大学数据科学学院; 上海创新研究院; 腾讯）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对通用LLM模拟个体行为时人物特质扁平化与任务微调泛化不足的问题，提出基于FONTS分类法及三阶段分层能力蒸馏的Socio-Foundation模型，在IndiEval上超越基础模型11.0分并接近前沿模型。
AI 中文摘要
模拟个体行为需要大型语言模型（LLM）在保持人物特质的同时适应动态社会情境。然而，通用LLM往往会抹平不同人物的独特性，而任务特定微调则面临碎片化和泛化能力不足的问题。为克服这些挑战，我们将个体模拟组织为FONTS分类法，涵盖五个互补的能力维度：人物保真度（F）、结果实现（O）、行为自然度（N）、轨迹连贯性（T）和社会接地（S）。基于该分类法，我们构建了一个包含约1000万条实例、覆盖14个代表性数据集的标准训练语料库，并提出了Socio-Foundation模型。Socio-Foundation通过三阶段流程将专业化与整合解耦：首先通过DAPO学习任务专家，然后通过离策略蒸馏将其整合为能力专家，最后通过多教师在线策略蒸馏（MOPD）进行统一。我们还建立了IndiEval，整合了FONTS各维度上的29项指标。实验表明，Socio-Foundation比其Qwen3-8B基础模型高出11.0分，并接近GLM-5.2等前沿模型，消融实验和分布外评估进一步证明了我们模型的有效性和泛化能力。
英文摘要
Simulating individual behavior requires large language models (LLMs) to preserve persona traits while adapting to dynamic social contexts. However, general-purpose LLMs often flatten distinct personas, while task-specific tuning suffers from fragmentation and generalization. To overcome these challenges, we organize individual simulation into the \textbf{FONTS Taxonomy}, comprising five complementary capability dimensions: \emph{persona fidelity} (\textbf{F}), \emph{outcome realization} (\textbf{O}), \emph{behavioral naturalness} (\textbf{N}), \emph{trajectory coherence} (\textbf{T}), and \emph{social grounding} (\textbf{S}). Grounded in this taxonomy, we curate a standardized training corpus library of approximately 10 million instances across 14 representative datasets and present \textbf{Socio-Foundation}. Socio-Foundation decouples specialization from integration via a three-stage pipeline: learning task experts via DAPO, consolidating them into capability experts via off-policy distillation, and unifying them via multi-teacher on-policy distillation (MOPD). We also establish \textbf{IndiEval}, consolidating 29 metrics across the FONTS dimensions. Experiments show that Socio-Foundation outperforms its \textit{Qwen3-8B} base by 11.0 points and approaches frontier models such as \textit{GLM-5.2}, with ablations and out-of-distribution evaluations further demonstrating the effectiveness and generalization of our model.
