Evaluating ontology mapping techniques: An experiment in public safety information sharing

The public safety community in the United States consists of thousands of local, state, and federal agencies, each with its own information system. In the past few years, there has been a thrust on the seamless interoperability of systems in these agencies. Ontology-based interoperability approaches in the public safety domain need to rely on mapping between ontologies as each agency has its own representation of information. However, there has been little study of ontology mapping techniques in this domain. We evaluate current mapping techniques with real-world data representations from law-enforcement and public safety data sources. In addition, we implement an information theory based tool called MIMapper that uses WordNet and mutual information between data instances to map ontologies. We find that three tools: PROMPT, Chimaera, and LOM, have average F-measures of 0.46, 0.49, and 0.68 when matching pairs of ontologies with the number of classes ranging from 13-73. MIMapper performs better with an average F-measure of 0.84 in performing the same task. We conclude that the tools that use secondary sources (like WordNet) and data instances to establish mappings between ontologies are likely to perform better in this application domain.