Models and Tools for Dynamic Health-Relevant Diffusion over Complex Networks

描述（由申请人提供）：健康和健康行为在很大程度上取决于社交网络上的扩散过程;传染病（如艾滋病毒）的传播以及影响健康行为的信念和做法（如饮食风险行为）的扩散都可以被理解为（有时非常复杂和动态的）社交网络上的广义扩散。然而，尽管有明确的需求，我们的正式分析工具很少可以用来模拟任何扩散，但这些系统中最简单的，作为数学建模扩散的动态多关系网络目前是棘手的，除非通过直接模拟;这往往是一个薄弱的基础，为推广知识;通过将动态网络分解为一组易处理的子集和关系特定的传输规则，我们提供了一种新的解决方案来建模动态网络上的复杂扩散过程。我们将开发改进的网络社区检测方法，专门针对理解扩散，并包括这样的社区在我们的一般框架，然后测试这些新模型对银行的建设和真实的社交网络，以评估模型的性能。使用模拟，新颖的新网络聚类技术和正式扩散模型的扩展，该项目将开发必要的工具，在最简单的直接扩散案例和最复杂的社会行为案例之间进行插值;进步将提供对健康相关扩散的基本见解，产生更强大的建模并突出可能的干预方向。

DESCRIPTION (provided by applicant): Health and health behavior depend crucially on diffusion processes over social networks; both the spread of communicable diseases, such as HIV, and the diffusion of beliefs and practices that shape health behavior, such as dietary risk behavior, can be understood as generalized diffusion over (sometimes very complex and dynamic) social networks. However, despite the clear need, few of our formal analytic tools can be used to model diffusion over any but the simplest of these systems, as the mathematics for modeling diffusion over dynamic multi- relational networks are currently intractable except by direct simulation; which is often a weak foundation for generalizable knowledge; we provide a novel solution to modeling complex diffusion processes over dynamic networks by decomposing the networks into a set of tractable subsets and relation specific transmission rules. We will develop improved methods for network community detection that specifically targets understanding diffusion, and include such communities in our general framework; then test these new models against a bank of constructed and real social networks to assess model performance. Using simulation, novel new network clustering techniques and extensions of formal diffusion models, this project will develop the tools necessary to interpolate between the simplest direct diffusion cases and the most complex social behavior cases; advances will provide fundamental insights into health- relevant diffusion, yielding more robust modeling and highlighting directions for possible intervention.