CSR : Applying the Classification-Based Learning of Subsumption Relations Method to the Oriented Matching Datasets of OAEI 2011 Vassilis Spiliopoulos, Konstantinos Kotis 1 and George A. Vouros 1,2 1 AI Lab, Information and Communication Systems Engineering Department, University of the Aegean, Samos, 83 200, Greece 2 Department of Digital Systems, University of Piraeus, Piraeus, Greece {vspiliop, georgev}@aegean.gr Abstract. This paper presents the results of the "Classification-Based Learning of Subsumption Relations” (CSR) method when applied to the Oriented Matching track dataset of OAEI 2011. The objective of the CSR method is to learn patterns of features that provide evidence for subsumption relations among classes, and thus, decide whether two classes belonging to two distinct ontologies are related via a subsumption relation. It must be emphasized that CSR does not discover equivalence relations between classes, and thus, it cannot infer subsumptions (for classes/properties) via equivalences. CSR addresses the discovery of subsumption relations as a classification task, using supervised machine learning methods. This paper describes briefly the method, provides experimental results over the OAEI 2011 Oriented Matching track, and discusses the potential of the method and possible improvements for the method. 1. Presentation of the system 2.1 State, purpose, general statement Subsumption relations are particularly useful when we deal with ontologies whose conceptualizations are at different “granularity levels”: in these cases, the elements (classes or properties) of an ontology are more generic than the corresponding elements of another ontology. Although subsumption relations between the elements of two ontologies may be deduced by exploiting equivalence relations between other elements (e.g. a class C 1 is subsumed by all subsumers of C 2 , if C 1 is equivalent to C 2 ), in the hard cases where no equivalence relations exist between classes/properties, or in cases where the assessed/provided equivalences are erroneous, the identification of subsumption relations between classes/properties cannot be done effectively. The "Classification-Based Learning of Subsumption Relations” (CSR) method has been motivated by the necessity to compute subsumption relations between elements of different ontologies in a direct way (i.e without relaying on equivalences). This can enhance the discovery/filtering of equivalence relations, and vise-versa, augmenting the effectiveness of any ontology alignment and merging method. Therefore, CSR addresses the problem of discovering subsumption relations between classes of two distinct ontologies, without relying on known equivalence relations among them. Specifically, given: (a) a source ontology O 1 =(S 1 , A 1 ) and a target ontology O 2 =(S 2 , A 2 ) where S is the ontological signature describing the vocabulary and A is the set of ontological axioms, restricting the intended meaning of the terms included in the signature, (b) the set W 1 ∪W 2 of distinct words that appear in both ontologies, and (c) optionally a morphism f:S 1 S 2 from the lexicalizations of ontology elements (properties or classes of the source ontology to the lexicalizations of the