Multi-scale Dynamical Community Detection for the Digital Economy: from analyzing to influencing policy through Open Government data

数字时代带来了前所未有的详细、实时的日常生活数据收集，从手机使用到专门的医院传感器。从丰富的物理和数字基础设施中获得这种真实世界的数据，再加上计算能力的增强，提供了一个独特的机会来审问社会行为，从个人的层面到不同层面的群体动态和特征的出现。最近，政府举措(特别是在美国和英国)旨在向更广泛的公众提供此类数据集。这些举措提供了定量审查政策对社会动态不同方面的影响和效力的可能性，并为公民参与和反馈提供了途径。这可能会改善医疗保健、交通、安全方面的生活质量，或设计从个人层面到集体层面的公共支出和资源使用政策。这些诱人的可能性导致了去年的一系列宣言，甚至宣布需要一个新的领域--计算社会科学。虽然这些贡献来自不同的学科，但它们都认为，目前缺乏用于分析这类数据集的数学工具构成了根本挑战，以便将多模式、动态数据集整合的前景转化为真正的解释性结果。特别是，有必要超越纯粹的(静态)统计方法，克服缺乏能够将数据正规化、审问和分析的数学和最终计算方法，以便能够以严格的数据驱动方式检验假设和得出结论。然而，这项建议超越了数据的可获得性和表示形式的问题，并侧重于开发数学工具，用于分两步分析数据：(1)根据多标签、可能是动态的网络找到数据的真实表示；(2)根据能够揭示数据中交互模式的多层次动态社区等级，生成这种网络的简化、可理解的简化。这项提议的目的是制定强有力的方法，分析目前通过开放政府倡议向公众提供的大型、复杂的社会数据集所衍生的网络。我们的数学工具将解决从数据创建代表性网络以及对这种网络的多尺度和多标签分析，从而在没有任何先验规范的情况下，根据从数据得出的动态社区结构来减少描述。所选择的数据集将具有当前的社会意义，但也体现了社交数据集的三个基本特征，这些特征与其分析的具体数学挑战有关：(I)社交网络的多尺度性质；(Ii)社交数据集的多标签表征；以及(Iii)动态和流动在社会描述中的重要性。数学工具将具体应用于以下三个对数字经济高度感兴趣的领域：邻里统计数据、重新划分选区的问题和最近公布的预算支出数据。

The digital age has brought with it an unprecedented gathering of detailed, real-time data from our daily lives, from mobile phone usage to specialized hospital sensors. The availability of such real-world data from a wealth of physical and digital infrastructures coupled with increased computational power offers a unique opportunity to interrogate social behaviour from the level of the individual to the emergence of group dynamics and traits at different levels. Recently, governmental initiatives (specifically in the US and the UK) have been designed to make such datasets available to the wider public. These initiatives offer the possibility to examine quantitatively the influence and effectiveness of policies on different aspects of social dynamics, as well as providing a route for the exercise of citizen participation and feedback. This could lead to improved quality of life in healthcare, traffic, security, or to the design of policies for public spending and usage of resources from the individual level to the collective of groups. These tantalising possibilities have led in the last year to a series of manifesto and even the declaration of the need for a new field, Computational Social Science.. Although those contributions have arisen from different disciplines, they share the belief that the lack of mathematical tools at present for the analysis of such datasets constitutes the fundamental challenge so that the promise of the integration of multi-modal, dynamic datasets can translate into real interpretative results. In particular, there is a need to go beyond the purely (static) statistical methods and to overcome the lack of mathematical, and eventually computational, methodologies that can formalise, interrogate and analyse the data such that hypotheses can be tested and conclusions can be drawn in a rigorous data-driven manner. This proposal, however, goes beyond issues of accessibility and presentation of data and focuses on the development of mathematical tools for the analysis of data in two steps: (1) finding a faithful representation of the data in terms of multi-label, possibly dynamic, networks, and (2) the generation of simplified, intelligible reductions of such networks in terms of a multi-level dynamical hierarchy of communities that can uncover patterns of interaction in the data. The aim of this proposal is to develop robust methodologies for the analysis of networks derived from large, complex social datasets currently made available to the public through the Open Government initiative. Our mathematical tools will address the creation of representative networks from the data and the multi-scale and multi-label analysis of such networks leading to reduced descriptions in terms of dynamical community structures derived from the data without any a priori specification. The datasets chosen will be of current social interest but also exemplify three fundamental characteristics of social datasets that are linked to specific mathematical challenges for their analysis: (i) the multi-scale nature of social networks; (ii) the multi-label characterisation of social datasets; and (iii) the importance of dynamics and flows in social descriptions. The mathematical tools will be specifically applied to the following three areas of high interest for the Digital Economy: Neighbourhood statistics data, the redistricting problem and the recently released budget expenditure data.