Katrin Weller & Isabella Peters & Wolfgang G. Stock - Generalizable Semantic Relations in Knowledge Representation

Methods of knowledge representation traditionally consist of a set of the defined concepts of a domain, and of paradigmatic semantic relations which are used for structuring these concepts. For years classical knowledge representation methods have been successfully working with established generalized relations such as synonymy, hierarchy (meronymy, hyponymy) and unspecified associations. Yet, these relations are still rather vague and somehow restricted; we expect that more relations can be specified and expediently applied to knowledge models. Recent developments like ontologies and folksonomies show new forms of collaboration, indexing and knowledge representation and thus encourage the reconsideration of standard knowledge relationships for practical use. In a summarizing overview we show which relations are currently utilized in elaborated knowledge representation methods (mainly thesauri and classifications) and which relations may be inherently hidden in folksonomies and ontologies. We believe in the necessity of reconsidering relationships regarding their classification and generalizability for the use in knowledge representation and information retrieval. New methods will have to be developed for structuring relationship types and identifying generalizable ones. The already existing theoretical considerations on specifying relationship types will have to be put into the context of practical knowledge representation and content indexing. We may then for example distinguish different subtypes of hierarchies and precisely specify associations. Different subtypes of relations may also possess specific properties and characteristics. Among them, one aspect of major importance is transitivity, which has strong effects on practical applications within information retrieval systems.  A better understanding of the semantics of relationships will lead to more accurate knowledge representation systems, which again will improve methods of knowledge management and information retrieval.