Case-based reasoning retrieval and reuse using case resemblance hypergraphs

This work presents a similarity case-based reasoning approach in which clustering and similarity relations plays a central role in the retrieval and reuse processes. A set of cases will form a cluster when the similarity of the case in the solution space is at least as large as their similarity in the problem space. Our approach is composed of four steps: preparation of cases in the case base, creation of the sets of (eventually intersecting) clusters of cases in the case base, selection of the cluster whose case descriptions reach the highest overall similarity with the new case description, and computation of the solution for the new problem as a function of the solutions yielded by the individual cases in the selected cluster. Preliminary results obtained in a classification task shows that our approach is promising.