P-graph-a graph model for anomaly checking of knowledge bases

Abstract

The authors present a graph model, P-graph, which supports the checking of knowledge bases for anomalies such as deadends, unreachability, cycles, inconsistency, redundancy, subsumption, and missing rules. P-graph captures the essential information needed for anomaly checks. The proposed approach differs from existing research as follows: it checks on groups of problem instances rather than on individual problem instances; it uses empirical knowledge to generate problem instances realizable in practice (only these problem instances need to be checked); and it considers the fact base as part of the knowledge base to be checked.<>