Introduction
In Geographic Information Systems (GIS), data interoperability is often hindered by "semantic rigidity." Different systems may use varying terminology for the same geographic features—for instance, labeling a location as a "harbor" versus a "seaport"—without a shared logical framework to reconcile them. While traditional digital gazetteers provide massive repositories of location names and coordinates, they typically lack the formal structure required for machine reasoning. They cannot inherently "understand" that a "river" is a subclass of a "flowing body of water" or that "Florence" is a specific instance of a "city."
To address this, researchers have developed a methodology for constructing high-quality, large-scale geospatial ontologies by adapting principles of "faceted classification" from 20th-century library science.
Methodology: The Faceted Approach
The core innovation lies in applying the Faceted Approach, originally proposed by S.R. Ranganathan in the 1930s. Rather than a single, rigid hierarchy, the domain of Space is decomposed into "facets"—homogeneous groups of terms sharing a common characteristic (e.g., "Bodies of Water" vs. "Administrative Divisions").
The construction process follows five rigorous steps:
- Identification: Aggregating terms from diverse sources.
- Analysis: Deconstructing terms using genus et differentia (e.g., a "river" is a "body of water" [genus] that is "flowing and confined to a bed" [differentia]).
- Synthesis: Organizing analyzed terms into facets.
- Standardization: Selecting preferred terms to eliminate ambiguity.
- Ordering: Arranging terms within facets based on logical principles like increasing size or complexity.
The resulting ontology architecture separates information into three distinct levels: a Formal language level for logical concepts, a Knowledge level for specific entity facts, and a Natural language level for linguistic representation.
Data Integration and Scale
The methodology was used to integrate and "upcycle" several massive existing resources, primarily GeoNames and WordNet. GeoNames provided the bulk of the spatial data, while WordNet contributed the linguistic structure and initial class hierarchies.
By mapping GeoNames feature codes to a refined faceted hierarchy, the researchers resolved common data issues, such as the "null class" problem where entities lack a specific parent category. In this system, every entity is forced into a logical hierarchy, ensuring no data point remains "floating" without context.
Key Results and Insights
The resulting ontology, part of the GeoWordNet project, represents one of the most comprehensive geospatial resources available:
- 845 Classes: A significant expansion over the original 663 in GeoNames.
- 6,907,417 Entities: Nearly 7 million specific locations populated automatically.
- 70 Relations & 31 Attributes: Including complex spatial relations like part-of, overlaps, and near-to.
- Multilingual Support: Full mapping between English and Italian concepts via MultiWordNet.
A significant insight of this work is the convergence of library science and computer science; the classification challenges solved by librarians decades ago are remarkably similar to modern data interoperability problems. By using facets, the system can disambiguate terms based on context—for example, distinguishing a "bank" (landform) from a "bank" (financial facility) based on its position in the hierarchy.
Conclusion and Future Work
This facet-based methodology successfully bridges the gap between massive, unstructured gazetteers and formal, reason-capable ontologies. While the initial categorization of facets requires manual expert intervention, the subsequent population of entities is highly automated. Future efforts will focus on expanding multilingual support and further automating the analysis phase to scale the methodology to other complex domains.
Summary updated