Automatic Concept-Based Query Expansion Using Weighted Transitive Part-Whole Relations

Ontologies as well as classification systems and thesauri consist of concepts and the paradigmatic semantic relations between them. Paradigmatic relations are document-independent relations used to model a domain of interest; in contrast to syntagmatic relations, which are merely based on the co-occurrence of concepts within single documents (Stock & Stock, 2008, pp. 68-70). Paradigmatic relations constitute the fundamental backbone of any controlled vocabulary. The “classical” semantic relations used in knowledge representation are synonymy, hierarchy and the unspecific association relation (Khoo & Na, 2006; Storey, 1993). Ontologies may define more specific relations. Still, the core semantic relation of every knowledge organization system is hierarchy. There are two kinds of hierarchic relations, which should be distinguished: hyponymy (is-a relation) (Cruse, 2002) and meronymy (part-of relation) (Winston, Chaffin, & Herrmann, 1987; Artale, Franconi, Guarino, & Pazzi, 1996; Iris, Litowitz, & Evens, 1989). Relations can furthermore obtain different properties, which allow drawing additional conclusions from the presence of certain relations. An important property of semantic relations is transitivity (others are mainly reflexivity or symmetry). Given x, y, z are concepts and ρ is a semantic relation, the relation ρ is transitive if