Paper #12

F. A. Lisi, D. Malerba "Towards Object-Relational Data Mining"

Keywords: data mining and knowledge discovery, logic and databases

Description logics (DLs) are general-purpose languages for knowledge representation and reasoning. They have been considered especially effective for those domains where the knowledge could be easily organized along a hierarchical structure, based on the is-a relationship. Recently DLs contaminated Datalog, thus yielding hybrid languages, such as AL-log, which can deal with both the structural and relational features of data. In this paper we propose to use AL-log as the starting point for the definition of a general framework for object-relational data mining (ORDM). Such a framework allows us to formulate data mining tasks in applications domains characterized by objects, properties of objects, relations between objects and concept hierarchies. Frequent pattern discovery at multiple levels of description granularity is taken as a showcase of ORDM.