Generating Concept Hierarchies/Networks: Mining Additional Semantics in Relational Data

Abstract

In relational theory, attribute domains are classical sets; no interactions among attribute values are modeled. So the concept hierarchies, which are additional semantics, used in data mining have to be input by users. In this paper, “real world” data model - relational model with additional semantics specified by binary relational structures (adopt from first order logic) - are explored; in such model, concept hierarchies/networks can be generated automatically. In fact, there are two families of concepts. One family forms a traditional hierarchy. Another forms a hierarchy syntactically, but semantically the hierarchy is a network; this is due to the fact that distinct concepts may be semantically related. A simple example is illustrated.Keywordsattribute oriented generalizationbinary relationconcept hierarchies/networksdata miningpartitiongranulationneighborhood systems