Variable Selection as an Instance-Based Ontology Mapping Strategy.
Konstantin Todorov, Peter Geibel · 2009
Abstract — The paper presents a novel instance-based approach to aligning concepts taken from two heterogeneous ontologies populated with text documents. We introduce a concept similarity measure based on the size of the intersection of the sets of variables which are most important for the class separation of the instances in both input ontologies. We suggest a VC dimension variable selection criterion elaborated for Support Vector Machines (SVMs), which is novel in the SVMs literature. The study contains results from experiments on real-world text data, where variables are selected using a discriminant analysis framework and standard feature selection techniques for text categorization. 1.