Diversity-Conscious Retrieval from Generalized Cases: A Branch and Bound Algorithm

Recommendation systems offer the most similar point cases to a target query. Among those cases similar to the query, some may be similar and others dissimilar to each other. Offering only the most similar cases wrt. the query leads to the well known problem that the customers may have only a few number of choices. To address the problem of offering a diverse set of cases, several approaches have been proposed. In a different line of CBR research, the concept of generalized cases has been systematically studied, which can be applied to represent parameterizable products. First approaches to retrieving the most similar point cases from a case base of generalized cases have been proposed. However, until now no algorithm is known to retrieve a diverse set of point cases from a case base of generalized cases. This is the topic of this paper. We present a new branch and bound method to build a retrieval set of point cases such that its diversity is sufficient and each case in the retrieval set is a representative for a set of similar point cases.