Extracting the backbone of weighted networks

在这个项目中，研究了提取加权网络主干的方法，并开发了使用这些方法的计算机软件。社会网络通常是复杂的，参与者（如人、组织、城市）之间的互动强度各不相同，这可以用加权网络来表示。因为加权网络很难分析和可视化，所以关注它们的主干通常是有用的，主干只包含最重要的连接。存在许多提取骨干的方法，但我们对它们是否有效或如何选择一种方法而不是另一种方法知之甚少。通过为网络主干提取提供指导和工具，该项目将使研究人员能够在广泛的社会重要背景下更好地分析信息丰富的网络数据，并通过可视化更容易地将他们的发现传达给不同的受众。该项目的目标是促进研究人员正确提取加权网络骨干的能力。实现这一目标涉及六项活动。首先，对二部集成方法进行了改进，使其更快、更灵活。其次，R主干包被扩展到允许从所有类型的加权网络中提取主干，以适应更大的数据集，并与其他网络分析包互操作。第三，利用基准数据集对骨干方法进行实证比较，探讨其在实践中的异同。第四，对骨干方法进行了分析比较，确定了它们的计算复杂度和零边权重分布的函数形式。第五，使用合成加权网络数据对骨干方法进行数值比较，允许识别每种方法有效再现已知基础真值的条件。最后，开发了培训材料，指导研究人员选择骨干方法和使用骨干包进行提取。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

In this project, methods to extract the backbone of weighted networks are investigated and computer software to use these methods is developed. Social networks are often complex, with the interactions between actors (e.g., people, organizations, cities) varying in strength, which can be represented by weighted networks. Because weighted networks are challenging to analyze and visualize, it is often useful to focus on their backbones, which contain only the most significant connections. Many approaches to backbone extraction exist, but we know little about whether they work or how to choose one approach over another. By providing guidance on and tools for network backbone extraction, this project will enable researchers to analyze better information-rich network data in a wide range of socially significant contexts, and to communicate their findings more easily to diverse audiences through visualization. The goal of this project is to facilitate researchers’ ability to correctly extract the backbone of weighted networks. Achieving this goal involves six activities. First, bipartite ensemble methods are refined to make them faster and more flexible. Second, the R backbone package is extended to allow the extraction of backbones from all types of weighted networks, to accommodate larger datasets, and to be interoperable with other network analysis packages. Third, backbone methods are compared empirically using benchmark datasets to explore their similarities and differences in practice. Fourth, backbone methods are compared analytically to determine their computational complexities and the functional form of their null edge weight distributions. Fifth, backbone methods are compared numerically using synthetic weighted network data, allowing identification of the conditions under which each method validly reproduces a known ground truth. Finally, training materials are developed to instruct researchers on the selection of backbone methods and on the use of the backbone package for their extraction.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.