Heuristics for constructing Bayesian Network based geospatial ontologies

Bayesian Network based ontologies enable specification of partial relations between concepts as an advantage over conventional ontologies, based on description logic. In the context of geospatial ontologies such specifications facilitate encoding relations between action and entitiy concepts. This paper presents a case study of transportation ontologies based on traffic code texts of two different countries. We construct ontologies of both geospatial entities and actions using the BayesOWL approach. Thereafter we employ heuristics based on verb-noun co-occurence evidences, available from analysis of formal texts, to construct linkages between the two types of concepts. This approach enables high recall and precission for querries on concepts and enables rich inferences such as most similar and disimilar concepts. The results of our experiments are verified with human subjects testing. Such heuristics-based-probablisitic approaches to geospatial ontology specification and reasoning can be utilized for concept mapping within and across geospatial ontologies as well as to quanitfy the naming hetrogeinities in two given ontologies.