09-10, 10:30–11:00 (America/Chicago), Grand G
Geospatial awareness can be brought to LLMs to address humanity’s most pressing challenges by using authoritative and cross-sectoral Spatial Knowledge Graphs that are automatically interoperable by location. We will present a technical approach using open-source software.
Advances in Artificial Intelligence are transforming industries, and we can expect geoinformatics to be no exception. Natural language processing (NLP) interfaces using Large Language Models (LLM) should not only be able to make finding data easier, but also should be able to reveal the semantic relationships between networks of features across domains. This could support cross-sectoral collaboration to address the multiple and interrelated challenges humanity faces (i.e., polycrises), including economic, public health, and climate.
Recent research indicates that spatial knowledge graphs (SKGs) can help overcome limitations of LLMs with domain-specific and evolving spatial awareness. However, to address polycrises at scale, SKGs across sectors need to be available from authoritative sources and interoperable by location using machine-to-machine readable interfaces as they change over time. This can be achieved using a Geospatial Knowledge Infrastructure (GKI) to allow SKGs to be integrated on-demand so that LLMs can answer ad-hoc questions.
We developed an approach using open-source software for managing dependencies and propagation of change between interlinked spatial knowledge graphs that was initially developed for the health sector but is being used to support national spatial infrastructures and climate disaster resiliency efforts, including the U.S. Army Corps of Engineers and the recent Open Geospatial Consortium (OGC) Climate Disaster and Resiliency Pilot (CDRP). Our presentation will provide a more in-depth and technical look than what was presented at the 129th Open Geospatial Consortium members meeting and will incorporate recent learnings. This will include a metamodel abstraction for making features and their semantic spatial relationships with other features findable, accessible, interoperable, and reusable as a part of a GKI architecture and how they can bring geospatial awareness to LLMs.