Similarity measures for concepts written in Description Logics (DLs) are often devised based on the syntax of concepts or simply by adjusting them to a set of instance data. These measures do not take the semantics of the concepts into account and can thus lead to unintuitive results. It even remains unclear how these measures behave if applied to new domains or new sets of instance data. In this paper we develop a framework for similarity measures for εLH-concept descriptions based on the semantics of the DL εLH. We show that our framework ensures that the measures resulting from instantiations fulfill fundamental properties, such as equivalence invariance, yet the framework provides the flexibility to adjust measures to specifics of the modelled domain.
展开▼
