MULTILEVEL HIERARCHICAL RETRIEVAL

As large databases of conceptual graphs are developed for complex domains, efficient retrieval techniques must be developed to manage the complexity of graph matching while reasonable space requirements are maintained. The paper describes a novel method, the 'multilevel hierarchical retrieval method', that exploits redundancy to improve both space and execution-time efficiency. The method involves search in multiple partially ordered (by 'more-general-than') hierarchies such that search in a simpler hierarchy reduces the search time in the hierarchy of next complexity. The specific hierarchies used are the traditional partial order over conceptual graphs, a partial order over node descriptors, and a partial order over 'descriptor units', and, finally, the simplest partial order is the traditional type hierarchy.