Completeness criteria for retrieval in recommender systems

Abstract

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. © Springer-Verlag Berlin Heidelberg 2006.