Industry Day

Michel Manago
(founder kaidara, France)

Stefan Wess
(empolis, Germany)

Main Conference

David McSherry
(University of Ulster, UK)

Often in practice, a recommender system query may include constraints that must be satisfied. Ensuring the retrieval of a product that satisfies any hard constraints in a given query, if such a product exists, is one benefit of a retrieval criterion we refer to as completeness. Other benefits include the ease with which the non-existence of an acceptable product can often be recognized from the results for a given query, and the ability to justify the exclusion of any product from the retrieval set on the basis that one of the retrieved products satisfies at least the same constraints. We show that in contrast to most retrieval strategies, compromise driven retrieval (CDR) is complete. Another important benefit of CDR is its ability to ensure the retrieval of the most similar product, if any, which satisfies all the hard constraints in a given query, a criterion we refer to as optimal completeness.

Gholamreza Nakhaeizadeh
(DaimlerChrysler AG, Research and Technology, Ulm, Germany)

In Data Driven Solutions using CBR or Data Mining approaches the optimal results can be achieved if one can consider and use, besides available dataand texts, all other available information sources like general andbackground knowledge. Formalization and integration of such kind ofknowledge in the knowledge extracted from data and texts is not, however, asimple task. For this reason, a lot of approaches, among them Bayesian Networks and Inductive Logic Programming, have been suggested in theliterature to solve this problem. In the talk, this topic will be discussed pragmatically by reviewing thepersonal experiences of the speaker in the last 20 years using concreteexamples from the automotive industry.

Edwina Rissland
(National Science Foundation, North Arlington, USA)

This talk discusses the importance of the post-retrieval steps of CBR, that is, the steps that occur after relevant cases have been retrieved. Ex- planations and arguments, for instance, require much to be done post-retrieval. I also discuss both the importance of explanation to CBR and the use of CBR to generate explanations.