On the optimization of rules in knowledge base systems

The problem of inferring rules describing concepts in terms of values of attributes is considered. It is assumed that the rules describing concepts from a set C are already stored in the (initial) knowledge base. However, these rules might contain superfluous attributes and might refer to concepts not necessarily taught before. To take care of this problem, a heuristic procedure is suggested that transforms rules of the initial knowledge base to a set of rules with terms in DNF. Some disjuncts are dropped and some replaced by other terms. However, the main goal of this procedure is to minimize the number of occurrences of values of attributes and the number of concepts in every rule.<>