Applying Inductive Logic Programming and Rule Relaxation for the Generation of Metadata
Andreas D. Lattner, Jan D. Gehrke · 2004
Activities towards Knowledge Management have recently become popular in enterprises to have all relevant information easily accessible. As one of the major problems encountered here is the cost for the manual acquisition of metadata this task should be supported by some semi-automated generation of metadata. In this work we present an approach for applying Inductive Logic Pro-gramming to create metadata generation rules. These rules can later be applied to create values for attributes of information items. As learned rules might not be completely correct or might miss values for attributes of objects, we propose a rule relaxation algorithm while applying the learned rules. For this purpose we represent rules by lattices of their variables. As we show, this structure can be easily used to identify reason-able generalizations of an original rule. 1