Structured Cases, Trees and Efficient Retrieval

A set of efficient algorithms for case retrieval from a case base of trees labeled on both nodes and edges is proposed. They extend the best known algorithm for solving the subtree-isomorphism problem. A branch and bound technique and a general definition of similarity between trees is considered. Both the case structure, i.e. the graph structure, and his semantic part, the labels, is taken into account when evaluating similarity. The comparison with a conventional state-space search algorithm on randomly generated case bases, shows that significant speed up can be obtained.