Social World Models

arXiv 25.8 2025 49 method

TLDR

Introduces Social World Models and S3AP representation to improve LLM social reasoning, achieving significant gains on benchmarks like FANToM and SOTOPIA.

Reasoning

The paper presents a novel concept (SWMs) and a structured representation (S3AP) that demonstrably improves LLM performance on social reasoning benchmarks. Strengths include clear empirical gains and ablation studies; weaknesses include narrow focus on social dynamics without comparison to other world model paradigms.

Read-first score

Read-first score 49, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 24.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=ablation,baseline,benchmark,experiment

Topical relevance 42%

34.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 442.

Keyword Scores

world model

world dynamics prediction

interactive world model

world simulator

generative world model

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

Introduction of Social World Models (SWMs) concept
Structured social world representation formalism (S3AP)
Algorithm for social world models using S3AP

Methodology

The paper introduces S3AP, a structured representation capturing evolving states, actions, and mental states of agents. Experiments are conducted across five social reasoning benchmarks, with baselines including OpenAI's o1. Ablations are performed to isolate the effect of explicit modeling of hidden mental states.

Key Results

S3AP enhances LLM performance by +51% on FANToM over OpenAI's o1, and yields up to +18% improvement on the SOTOPIA multi-turn social interaction benchmark.

Resources

Paper