GeoAgent: An Agentic AI Framework for Spatial Query Understanding and Interactive Geospatial Intelligence

作者：Jinghong Hu, Ligang Sun, Xiliang Liu · 年份：2025 · DOI：10.1145/3764915.3770719 · 被引用次数：1 · 研究领域：Multimodal Machine Learning Applications、Geographic Information Systems Studies、Data Visualization and Analytics

Abstract—Generating actionable intelligence from unstructured geospatial data is a critical challenge in high-stakes domains like disaster management and urban security. We introduce GeoAgent, an agentic AI framework designed to synthesize fragmented reports and multimodal data to generate an interactive knowledge space for decision-making. The framework orchestrates multiple AI agents that leverage novel semantic segmentation and a two-stage retrieval mechanism to generate context-aware answers, geospatial visualizations, and decision-support narratives in response to complex spatial queries. A key innovation is its localized deployment capability. Optimized via GPTQ quantization, GeoAgent operates effectively offline on consumer-grade hardware, making it a viable solution for generative AI at the edge. Validated on over 7,300 maritime incident reports, our system substantially reduces harmful hallucinations compared to baseline models, demonstrating its value as a trustworthy tool for geospatial content generation. GeoAgent provides a foundational blueprint for generative geospatial assistants, enabling the on-demand synthesis of multimodal intelligence and supporting dynamic risk assessment directly at the point of need.