A case-based reasoning (CBR) system is only as good as the cases within its Case Base and its ability to retrieve those cases in response to a new situation. In this paper we focus on the case retrieval problem and on the computation of similarity measures between cases. We illustrate this problem by showing an application of our CBR system, named CARS, Combining Approximate Reasoning Systems, in the domain of Mergers and Acquisitions.
We define a case as a situation/solution pair, indexed by surface (observed) and abstract (derived) features. The mapping from surface to abstract features, based on fuzzy predicates and plausible rules implemented in Plausible Reasoning MOdule (PRIMO), allows us to represent the situation descriptor of the case in a more robust feature space.
The similarity of each abstract feature is computed as the complement of the distance between the fuzzy numbers representing the feature values. The abstract features similarities are aggregated hierarchically, according to a semantic taxonomy. The aggregation is based on T-norms, averaging operators, and T-conorms.