Mining the implicit complete graph of knowledge bases

Association rule mining identifies patterns in transaction data that are not explicit. In the area of Knowledge Representation, cycle mining algorithms identify meta-patterns of these associations depicting inferences forming feedback chains of positive and negative rule dependencies. This paper presents one previously developed and two new algorithms in the domain of cycle mining. Using a variation of the Apriori association rulemining algorithm [AGRA93], we build our cycle mining formalism applicable to any relational database. Second, we show that every relation contains an implicit complete graph representing intrinsic causal relationships within the relation. Third, we describe a framework for cyclic meta-pattern extraction containing a new data structure and algorithm for the discovery of all cycles in a complete graph. Last, we describe our current system implementation of these algorithms.