[Paper Review] Universal Components of Real-world Diffusion Dynamics based on Point Processes

Minkyoung Kim, Raja Jurdak+1 more|arXiv (Cornell University)|Jun 20, 2017

Ecosystem dynamics and resilience90 references3 citations

TL;DR

This paper proposes a universal framework for real-world diffusion dynamics using point processes, identifying common components—such as endogenous and exogenous influences, temporal decay, and preferential attachment—across disciplines. It applies these components to model dengue spread, demonstrating their generalizability for transdisciplinary diffusion modeling with improved interpretability and predictive power.

ABSTRACT

Bursts in human and natural activities are highly clustered in time, suggesting that these activities are influenced by previous events within the social or natural system. Bursty behavior in the real world conveys information of underlying diffusion processes, which have been the focus of diverse scientific communities from online social media to criminology and epidemiology. However, universal components of real-world diffusion dynamics that cut across disciplines remain unexplored. Here, we introduce a wide range of diffusion processes across disciplines and propose universal components of diffusion frameworks. We apply these components to diffusion-based studies of human disease spread, through a case study of the vector-borne disease dengue. The proposed universality of diffusion can motivate transdisciplinary research and provide a fundamental framework for diffusion models.

Motivation & Objective

• To identify universal factors underlying bursty diffusion dynamics across diverse disciplines such as epidemiology, criminology, and social media.
• To develop a taxonomy of universal effects—like endogenous/exogenous influences and temporal decay—that govern real-world diffusion processes.
• To demonstrate the applicability of these universal components through a case study on dengue fever transmission using point process modeling.
• To establish a general, interpretable framework for diffusion modeling that transcends domain-specific limitations and supports transdisciplinary research.
• To guide the design of robust, scalable diffusion models by addressing key challenges in factor selection, balance, coverage, and dependency on prior knowledge.

Proposed method

• Uses point processes as a mathematical foundation to model discrete, bursty event sequences in time, enabling flexible representation of complex diffusion dynamics.
• Identifies four core universal components: endogenous influence (self-excitation), exogenous influence (external shocks), temporal decay (memory effects), and preferential attachment (preferential spreading).
• Applies these components to model dengue transmission by integrating human mobility, climate, and social interaction data as heterogeneous signals.
• Employs model-driven, parametric estimation to capture non-linear dynamics, contrasting with model-free, information-theoretic alternatives.
• Constructs a high-level diffusion framework where universal components serve as contextual building blocks, adaptable to domain-specific contexts.
• Compares existing point process models through the lens of the proposed taxonomy to evaluate their coverage, interpretability, and balance of factors.

Experimental results

Research questions

• RQ1 What universal components underlie bursty diffusion dynamics across diverse real-world domains such as disease spread, social media, and seismic activity?
• RQ2 How can these universal components be systematically organized into a generalizable framework for modeling diffusion processes?
• RQ3 To what extent can the proposed framework improve the interpretability and predictive performance of disease diffusion models, such as for dengue?
• RQ4 How do endogenous and exogenous factors interact in shaping the temporal clustering of diffusion events?
• RQ5 What design principles—such as factor targeting, balancing, and coverage—enable a robust and general diffusion framework?

Key findings

• The study identifies four universal components—endogenous influence, exogenous influence, temporal decay, and preferential attachment—that consistently shape bursty diffusion across disciplines.
• These components are validated in a dengue case study, where they enhance the interpretability and contextual richness of the diffusion model.
• The framework demonstrates that point process models incorporating these universal components can better capture complex, real-world diffusion dynamics than models relying on homogeneous signals or simplified assumptions.
• Factor balancing—particularly between positive (e.g., preferential attachment) and negative (e.g., time decay) effects—is critical for accurate modeling of event clustering.
• The framework shows strong potential for transdisciplinary application, as the same components can be adapted to model phenomena from social media virality to earthquake aftershocks.
• The approach reduces dependency on domain-specific prior knowledge while maintaining high coverage of real-world diffusion cases, improving generalizability and model robustness.