Mining Complex Interconnected Data

联系在不同的领域无处不在；从通过电子邮件交流的员工，到相互作用以执行生物功能的蛋白质和基因。这个项目旨在通过设计模型和算法来研究相互关联的数据，这些模型和算法在制定和解决现实世界的问题时整合了两个互补的信息源：连接和内容。例如，在调查在线护送广告中的人口贩运活动时，我们希望考虑从每个广告内容中提取的指标，例如对未成年受害者的引用；以及不同广告如何相互联系，例如使用共享电话号码，这可能会揭示有组织的活动。为了聚合连接和内容，这一建议借鉴了网络科学和机器学习的想法；后者传统上侧重于内容(单个数据点的特征)，而前一个领域主要研究它们之间的联系。虽然两者之间存在多条研究路线，但仍需要付出大量努力才能对我们在现实世界中遇到的复杂数据进行具有表现力的、通用的、可伸缩的和依赖时间的建模。更具体地说，这项建议追求三个目标，以建立更丰富的互联数据模型：(I)在真实世界特征丰富的图表中找到特定于领域的通用模式；(Ii)重新定义和统一机器学习和网络科学中用于互联数据分析的经典问题和技术；以及(Iii)形式化和解决现有框架之外的新任务。我计划通过调查它们在三个应用领域的更广泛的影响来实现这三个目标：数据驱动的反人口贩运、计算社会科学和在线学习生态系统。在这三个应用中，相互关联的数据分析是理解和解决手头问题的关键。这一研究计划提供了一个及时的计划，通过结合网络科学、机器学习和数据挖掘的最新进展的跨学科方法，推进研究在线社会等新兴现象所需的理论基础，并解决当今世界的问题，如在线人口贩运。

Connections are ubiquitous in different domains; from employees communicating by emails, to proteins and genes interacting to carry out a biological function. This program aims to study interconnected data by designing models and algorithms that integrate two complementary sources of information when formulating and solving real-world problems: connection and content. For example, when investigating human trafficking activity in online escort advertisement, we want to consider indicators extracted from the content of each ad, e.g., references to an underage victim; as well as how different ads are connected to each other, e.g., using a shared phone number, which may reveal an organized activity. To aggregate connection and content, this proposal draws on ideas from network science and machine learning; where the latter domain has been traditionally focused on the content (features of individual datapoints), and the former area primarily studies the connections between them. While there exist multiple lines of research in between, substantial effort is still needed to enable expressive, general, scalable, and time-dependent modelling of the complex data we encounter in real-world settings. In more concrete terms, this proposal pursues three objectives towards richer models for interconnected data: (i) finding domain-specific and universal patterns in real-world feature-rich graphs; (ii) redefining and unifying classical problems and techniques in machine learning and network science for interconnected data analysis, and (iii) formalizing and addressing novel tasks that are beyond the existing frameworks. I plan to navigate these three objectives by investigating their broader impacts within three application domains: data-driven counter human trafficking, computational social science, and online learning ecosystems. In these three applications, interconnected data analysis holds the key to understanding and solving the problems at hand. This research program provides a timely plan to advance the theoretical foundations needed for the study of emerging phenomena, such as online societies, and addressing today's world's problems, such as online human trafficking, through an interdisciplinary approach that unifies the recent progress in network science, machine learning and data mining.