A tremendous amount of geospatial knowledge remains hidden in text-based data sources such as news reports, research papers, and social media.

To address this, we propose a novel architecture, the geospatial knowledge hypercube. The hypercube is a multi-scale, multidimensional knowledge structure that integrates information from multiple geospatial dimensions, themes, and application semantics. This integration is achieved by analyzing and extracting relevant data from large volumes of spatially-relevant text data. The hypercube can then serves as a foundation for various knowledge discovery processes, in-depth statistical analysis, and other advanced applications.

Constructing the geospatial knowledge hypercube requires the development of weakly supervised machine learning approaches. These approaches must be capable of automatically, dynamically, and incrementally combining information from unstructured text, structured tables, maps, and image data to extract geospatial data, themes, latent connections and relationships, and application semantics. Moreover, constructing the cube requires the development of new methods for recognizing geospatial entities and inferring geospatial relationships.

This VCO will introduce and demonstrate a knowledge extraction system designed for the automatic construction of the hypercube. The system has already been deployed for public access.