Convergence Accelerator Phase I (RAISE): Spatially-Explicit Models, Methods, and Services for Open Knowledge Networks

NSF融合加速器支持基于团队的多学科努力，以应对国家重要性的挑战，并在不久的将来显示出交付成果的潜力。这一融合加速器第一阶段项目的更广泛影响和潜在社会效益将有助于将业界、政府机构、学术界和更广泛的公众中的广大地理信息系统用户社区连接到知识图谱。知识图谱由关于我们这个世界的相互关联的信息组成。通过连接不同领域、多媒体格式和视角的数据，知识图使用户能够提出更复杂的问题，并对复杂的物理和社会过程达成更全面的理解。该项目侧重于确定知识图谱的空间和时间并将其联系起来所需的工具。这些工具很重要，因为每件事都在某个地方和某个时间发生，而且知道事情发生的地点和时间对于理解它们为什么和如何发生至关重要。目前，学术界、工业界和政府使用的地理信息系统还没有与知识图谱很好地集成。该项目的高度跨学科团队(代表四所主要大学、三个行业合作伙伴和两个政府机构)将演示如何开发通用模型、方法和服务，以实现跨域边界知识图的空间和时间数据的发布、检索、重用、分析和推断。该团队计划将开发的技术应用于土壤健康、水文学和城市规划等领域。预期的项目结果将使领域专家和更广泛的公众能够自由和开放地访问数据和分析功能，这些功能将把有关极端事件、土壤健康、智能农业和城市规划的原本互不相连的知识联系在一起。这一努力设想的开放知识网络有可能使人们更容易获得知识图谱数据，这可能会增加依赖连接地理信息系统数据的许多部门的机会。地点将人、实体和事件连接在一起，因此，它们是迄今为止所有通用知识图谱中紧密相连的一部分。尽管如此，个别领域已经发展出自己的方式(通常是不兼容的)来表示地点以及更一般的空间和时间。例如，没有达成一致的方法来在时空上限制诸如改变国界之类的声明。同样，用于推断新知识图陈述或总结现有知识图谱陈述的现有机器学习技术通常忽略空间和时间方面，因此无法充分发挥其潜力。最后，现有的努力无法处理卫星图像等数字或多模式数据。这一第一阶段项目的总体主题是在如何有效地表示、检索和分析几乎所有数据集中存在的空间和时间方面达成一致，而无论它们来自哪个单独的学科，从而产生互操作性，而不是特定于领域的和可能不兼容的解决方案。在第一阶段，该团队将审查最先进的方法、模型和服务；跨领域边界协调它们；开发空间和时间显式的机器学习方法和工具，以更好地表示、分析和推断空间和时间数据；并为其他开放知识网络提供最佳实践。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The NSF Convergence Accelerator supports team-based, multidisciplinary efforts that address challenges of national importance and show potential for deliverables in the near future. The broader impact and potential societal benefit of this Convergence Accelerator Phase I project will be to help connect the vast community of geographic information systems users in industry, government agencies, academia, and the broader public, to knowledge graphs. Knowledge graphs consist of interlinked pieces of information about our world. By connecting data across different domains, multimedia formats, and perspectives, knowledge graphs enable users to ask more complicated questions and arrive at a more holistic understanding of complex physical and social processes. This project focuses on the tools needed to identify and link space and time for knowledge graphs. These tools are important because everything happens somewhere and at some time and because knowing where and when things happen is critical to understanding why and how they happen. Currently geographic information systems in use in academia, industry, and governments are not yet well integrated with knowledge graphs. This project's highly interdisciplinary team (representing four major universities, three industry partners, and two government agencies) will demonstrate how to develop common models, methods, and services to enable the publication, retrieval, reuse, analysis, and inference of spatial and temporal data for knowledge graphs across domain boundaries. The team plans to will apply the techniques developed to applications such as soil health, hydrology, and urban planning. Expected project results will give domain experts and the broader public free and open access to data and analysis functions that will link otherwise disconnected knowledge about extreme events, soil health, smart farming, and urban planning together. The open knowledge network envisioned in this effort has the potential to provide easier access to knowledge graph data which may improve opportunities for many sectors that rely on connecting to geographic information system data. Places connect people, entities, and events together and, thus, are a densely interconnected part of all general-purpose knowledge graphs to date. Nonetheless, individual domains have developed their own (often incompatible) ways to represent places as well as space and time more generally. For instance, there is no agreed upon method to spatio-temporally restrict statements such as changing national boundaries. Similarly, established machine-learning techniques used to infer new knowledge graph statements or to summarize existing ones typically ignore spatial and temporal aspects and, thus, fall short of their full potential. Finally, existing efforts cannot handle numerical or multi-modal data such as satellite imagery. The overarching theme of this Phase I project is to reach convergence on how to efficiently represent, retrieve, and analyze the spatial and temporal aspects present in almost all datasets irrespective of the individual disciplines they originate from, resulting in interoperability instead of domain-specific and potentially incompatible solutions. During phase I the team will review state-of-the-art methods, models, and services; harmonize them across domain boundaries; develop spatially and temporally explicit machine-learning methods and tools to better represent, analyze, and infer spatial and temporal data; and provide best practices for other open knowledge networks.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.