Automatic transformation and enlargement of similarity models for case-based reasoning
Mirjam Minor, Karsten Schmidt · 2006
In this paper, we present a new approach how similarity models for Case-Based Reasoning can be extended by means of operators for relations. A second contribution of the relations approach is that the mathematical model can be transformed into an OWL notation as well as in the proprietary format that is needed to build a Case Retrieval Net1 Inside a case-based system, a case c is represented by a set of information entities ei E: c = {e1, e2, ..., en}. It is compared with a query q that is as well just a set of information entities. A composite similarity function for q, c ⊆ E determines a numeric value for the degree of similarity SIM(q, c). SIM is computed by means of a local similarity function sim : E × E → R. A very simple example of SIM is the sum of the local values: SIM(q, c) = ∑ ei q ∑ ej c sim(ei, ej). To provide the values for sim, we need a similarity model. In small domains, this might be a table of manually specified similarity values for the pairs of information entities like, for instance, sim(MPEG − 4, XviD) = 0.3, sim(XviD, V ideocodec) = 0.7. In real world applications, the similarity model is often too complex to be clearly arranged in a table. We employ more abstract relations instead. A similarity type is a binary relation between two information entities S ⊆ E × E. Each similarity type has a qualitative or quantitative identifier, e.g. “IS SUCCESSOR”, “IS ABSTRACTION”, “LOW SIMILARITY”. A set S of similarity types is called a similarity dictionary. A weighting function g : S → R assigns a numeric value to a similarity ∗This work has been partly funded by the BMBF project URANOS (no. 01M3075) 1Mario Lenz and Hans-Dieter Burkhard. Lazy Propagation in Case Retrieval Nets. In Wolfgang Wahlster, editor, 12th European Conf. on Artificial Intelligence (ECAI96), pages 127 131. John Wiley & Sons, 1996.