Ontology Matching Enhanced with Similarity Measures for Georeferenced Datasets

In this work I present a technique to improve the capability of the current data management systems to deal with geospatial data. In particular, I focus on enhancing ontology matching algorithms in order to make them more effective when identifying similarities between geospatial ontologies. This work is meant to define the basic techniques for creating a framework capable of identifying any kind of relationships between geospatial datasets. We proceed following two steps: first, we define similarity measures for comparing the instances of geospatial ontologies; second, we integrate the result into a matcher. To compare the datasets we create a tessellation to reduce them to a common format. Maximizing the spatial autocorrelation among the cells we are able to identify the tessellation that best expresses the degree of clustering of the data. Finally, Person's R is used as similarity measure to compare the distributions. I propose a few different ways to integrate the obtained similarity measure into an ontology matching algorithm. I show the effectiveness of each of the used techniques with tests performed both on synthetic and real datasets. We also suggests how to compare datasets collected in different places in different time intervals. Our approach allows to address the MAUP problem and to integrate datasets having different resolutions.