Large language models for spreading dynamics in complex systems

Spreading dynamics is a central topic in the physics of complex systems and network science, providing a unified framework for understanding how information, behaviors, and diseases propagate through interactions among system units. In many propagation contexts, spreading processes are influenced by multiple interacting factors, such as information expression patterns, cultural contexts, living environments, cognitive preferences, and public policies, which are difficult to incorporate directly into classical modeling frameworks. Recently, large language models (LLMs) have exhibited strong capabilities in natural language understanding, reasoning, and generation, enabling explicit perception of semantic content and contextual cues in spreading processes, thereby supporting the analysis of the different influencing factors. Beyond serving as external analytical tools, LLMs can also act as interactive agents embedded in propagation systems, potentially influencing spreading pathways and feedback structures. Consequently, the roles and impacts of LLMs on spreading dynamics have become an active and rapidly growing research area across multiple research disciplines. This review provides a comprehensive overview of recent advances in applying LLMs to the study of spreading dynamics across two representative domains: digital epidemics, such as misinformation and rumors, and biological epidemics, including infectious disease outbreaks. We first examine the foundations of epidemic modeling from a complex-systems perspective and discuss how LLM-based approaches relate to traditional frameworks. We then systematically review recent studies from three key perspectives, which are epidemic modeling, epidemic detection and surveillance, and epidemic prediction and management, to clarify how LLMs enhance these areas. Finally, open challenges and potential research directions are discussed.

💡 Research Summary

The paper presents a comprehensive review of how large language models (LLMs) are reshaping the study of spreading dynamics in complex systems, covering both digital epidemics (misinformation, rumors) and biological epidemics (infectious disease outbreaks). It begins by situating spreading processes within the broader framework of network science and statistical physics, noting that classical approaches—compartmental models such as SIS/SIR and large‑scale agent‑based simulations—often simplify or omit crucial contextual factors like cultural norms, policy interventions, cognitive biases, and environmental conditions. These omissions limit the explanatory and predictive power of traditional models when applied to real‑world phenomena.

The authors then detail the technical foundations of LLMs, emphasizing their self‑supervised training on massive, heterogeneous text corpora, which endows them with rich semantic representations and cross‑domain knowledge. Two principal roles for LLMs are identified: (1) analytical tools that process, summarize, and reason over multimodal data streams (text, images, timestamps, network structures) and (2) generative agents that can actively participate in information ecosystems (e.g., chatbots, synthetic influencers). As analytical tools, LLMs can extract fine‑grained thematic and sentiment trajectories from social media, reconstruct diffusion networks by linking content to user interactions, and fuse expert knowledge with real‑time news to augment epidemiological forecasts. As generative agents, they can produce or amplify content, provide personalized health advice, and thereby alter contact patterns, behavioral responses, and feedback loops that drive the spread itself. This dual capacity introduces a new class of “human‑AI co‑evolutionary” dynamics, where model outputs influence human behavior, which in turn feeds back into future model training.

The review organizes recent literature into three methodological pillars. In epidemic modeling, LLM‑augmented agent‑based frameworks allow agents to make language‑driven decisions, turning the “payload” of contagion into a modelable semantic variable rather than a simple binary state. In detection and surveillance, LLM‑powered pipelines perform real‑time fact‑checking, early‑warning signal extraction, and multimodal network monitoring, outperforming rule‑based systems in speed and nuance. In prediction and management, LLMs synthesize historical literature, policy documents, and news narratives to generate scenario‑based forecasts, support decision‑makers in evaluating intervention strategies, and simulate the impact of AI‑generated misinformation on outbreak trajectories.

The authors also discuss practical challenges. Data privacy, model bias, hallucination risk, and the need for interpretability are highlighted as critical concerns when deploying LLMs in public‑health contexts. Technically, the integration of multimodal inputs, domain‑specific fine‑tuning, continual online learning, and scalable simulation‑to‑real‑time pipelines remain open research problems. From a governance perspective, the paper calls for regulatory frameworks that ensure transparency, accountability, and ethical use of LLM‑driven content creation, especially given the potential for AI‑generated misinformation to exacerbate crises.

In conclusion, the review argues that LLMs constitute a transformative “semantic‑contextual infrastructure” for spreading‑dynamics research. By bridging qualitative narrative information with quantitative network models, they enable a more faithful representation of the non‑linear, heterogeneous nature of real‑world diffusion processes. The paper positions LLMs not merely as auxiliary analysis tools but as active participants that can reshape the very pathways of information and disease spread, opening new avenues for interdisciplinary collaboration across physics, computer science, epidemiology, and social science.