Decision diagrams: Fast and flexible support for case retrieval and recommendation

Abstract. We show how case bases can be compiled into Decision Di-agrams, which represent the cases with reduced redundancy. Numerous computations can be performed efficiently on the Decision Diagrams. The ones we illustrate are: counting characteristics of the case base; comput-ing the distance between a user query and all cases in the case base; and retrieving the k best cases from the case base. Through empirical investigation on four case bases, we confirm that Decision Diagrams are more efficient than a conventional algorithm. Finally, we argue that De-cision Diagrams are also flexible in that they support a wide range of computations, additional to the retrieval of the k nearest neighbours.