Graph World Model

arXiv 25.7 2025 55.6 method

TLDR

Proposes Graph World Model integrating graph-structured data via message-passing and action nodes, outperforming baselines on six diverse tasks.

Reasoning

Strengths include novel integration of graph-structured data into world models and strong empirical results across multiple domains. Weaknesses are lack of explicit limitations discussion and absence of video or interactive aspects.

Read-first score

Read-first score 55.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 22.

Recency 8%

86.7

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

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=code,github

Methodology quality 25%

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

Topical relevance 42%

31.4

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

Field roles

FrontierReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 519.

Keyword Scores

world model

generative world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

interactive world model

video world model

Deep Analysis

Innovations

Unified world model supporting both unstructured and graph-structured states with multi-modal information
Representation of diverse tasks as actions via action nodes linked by direct reference or similarity
Generic message-passing algorithm for aggregating structured information over unified multi-modal token space (GWM-T) or embedding space (GWM-E)
Strong zero-shot/few-shot capabilities on unseen new tasks demonstrated across six diverse domains

Methodology

GWM employs a generic message-passing algorithm to aggregate structured information over either a unified multi-modal token space (GWM-T) by converting multi-modal data to text, or a unified multi-modal embedding space (GWM-E) using modality-specific encoders. It introduces action nodes linked to other nodes via direct reference or similarity to support diverse tasks. The model is evaluated on six tasks across multiple domains, comparing against domain-specific baselines.

Key Results

The same GWM outperforms or matches domain-specific baselines on six tasks including multi-modal generation and matching, recommendation, graph prediction, multi-agent, retrieval-augmented generation, and planning and optimization. It benefits from multi-hop structures and demonstrates strong zero-shot/few-shot capabilities on unseen new tasks.

Resources

Paper Code