Abstract

Observation-Driven Geo-Ontology Engineering
Track: Education and Training
Authors: Krzysztof Janowicz

While ontologies are crucial to reuse and integrate data from heterogeneous sources, existing ontology engineering methodologies are not well suited to bridge the gap between observation data, scientific models, and knowledge engineering. We propose an observation-driven approach to the engineering of geo-ontologies. Starting with the semantic description of sensors and their observations, we show how to combine semantics with geostatistics, data mining, and machine learning to arrive at ontological primitives. These primitives are integrated with geo-ontology design patterns acting as strategies to assist domain experts in becoming knowledge engineers. These patterns can be combined to crisp application-level ontologies. Mappings between them are inferred, e.g., via similarity reasoning and ontology alignment. Our methodology preserves semantic diversity without given up interoperability. Provenance information is maintained while going up the layers from observations to ontologies. Scientists exchanging data on the category level can track down how classes were constructed to uncover hidden incompatibilities.