Hierarchical model improves simulation of individual social behavior
Socio-Foundation: A Model for Generalizable Individual Behavior Simulation via Hierarchical Capability Distillation
Artificial Intelligence
Summary
Simulating how people behave in social settings is hard because models often ignore unique personality traits or struggle to adapt to different situations. The authors organize behavior simulation into five key types of skills and create a large training collection combining many datasets. They build a model called Socio-Foundation that learns expert skills separately and then combines them to simulate people more realistically. Tests show their model works better than a strong base model and is good at handling new, unseen situations.
What this means in practice
- •For game developers: Create more realistic non-player characters by simulating individual social behaviors with preserved personalities and context awareness.
- •For customer service chatbot teams: Improve chatbot responses by modeling diverse user personas and social interactions for natural and coherent conversations.
Authors
Liang Wang, Wenxuan Xie, Xinyi Mou, Yixin Luo, Zhongyu Wei
Abstract
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.