Artificial Intelligence Review (2005)24: 319-338 C Springer 2005 DOI10.1007/10462-005-9000-z Retrieval Failure and Recovery in Recommender Systems DAVID MCSHERRY School of Computing and Information Engineering. University of Ulster, Coleraine BT52 ISA, Northern Ireland, UK(Tel: +44(0)28 7032 4130, Fax:+44(0)28 7032 Abstract. In case-based reasoning(CBR) approaches to product recommendation, descriptions of the available products are stored in a case library and retrieved in response to a query representing the users requirements. We present an approach to recovery from the retrieval failures that often occur when the user's requirements are treated as constraints that must be satisfied. Failure to retrieve a matching case trig- gers a recovery process in which the user is invited to select from a recovery set of relaxations (or sub-queries) of her query that are guaranteed to succeed. The sug gested relaxations are ranked according to a simple measure of recovery cost defined n terms of the importance weights assigned to the query attributes. The recovery set for an unsuccessful query also serves as a guide to continued exploration of the prod uct space when none of the cases initially recommended by the system is acceptable to the user Keywords: case-based reasoning, recommender systems, retrieval failure, query 1. Introduction n spite of limitations such as lack of diversity in the recommended cases(e.g. McGinty and Smyth, 2003; McSherry, 2003c), similar y-based retrieval remains the dominant CBR approach to retrieval in recommender systems. However, it is increasingly common for the user's requirements to be treated initially as constraints that the recommended products must satisfy(Goker and Thompson, 2000: Bridge, 2002; Ricci et al., 2002; McSherry, 2004b: Thompson et al 2004). Typically these approaches rely on query relaxation to recover from the retrieval failures that occur when there is no product that satisfies all the user's requirements. We focus here on approaches in which relaxing a query means eliminating one or more constraints from the query rather than requiring the user to revise individual con- straints, for example as in Sermo(Bridge, 2002) In Ricci et al.s (2002)Intelligent Travel Recommender, the user is told how many results she will get, if any, by eliminating each ofDOI 10.1007/s10462-005-9000-z Artificial Intelligence Review (2005) 24:319–338 © Springer 2005 Retrieval Failure and Recovery in Recommender Systems DAVID MCSHERRY School of Computing and Information Engineering, University of Ulster, Coleraine BT52 1SA, Northern Ireland, UK (Tel: +44 (0)28 7032 4130; Fax: +44 (0)28 7032 4916; e-mail: dmg.mcsherry@ulster.ac.uk) Abstract. In case-based reasoning (CBR) approaches to product recommendation, descriptions of the available products are stored in a case library and retrieved in response to a query representing the user’s requirements. We present an approach to recovery from the retrieval failures that often occur when the user’s requirements are treated as constraints that must be satisfied. Failure to retrieve a matching case trig￾gers a recovery process in which the user is invited to select from a recovery set of relaxations (or sub-queries) of her query that are guaranteed to succeed. The sug￾gested relaxations are ranked according to a simple measure of recovery cost defined in terms of the importance weights assigned to the query attributes. The recovery set for an unsuccessful query also serves as a guide to continued exploration of the prod￾uct space when none of the cases initially recommended by the system is acceptable to the user. Keywords: case-based reasoning, recommender systems, retrieval failure, query relaxation 1. Introduction In spite of limitations such as lack of diversity in the recommended cases (e.g. McGinty and Smyth, 2003; McSherry, 2003c), similar￾ity-based retrieval remains the dominant CBR approach to retrieval in recommender systems. However, it is increasingly common for the user’s requirements to be treated initially as constraints that the recommended products must satisfy (Goker and Thompson, 2000; ¨ Bridge, 2002; Ricci et al., 2002; McSherry, 2004b; Thompson et al., 2004). Typically these approaches rely on query relaxation to recover from the retrieval failures that occur when there is no product that satisfies all the user’s requirements. We focus here on approaches in which relaxing a query means eliminating one or more constraints from the query rather than requiring the user to revise individual con￾straints, for example as in Sermo (Bridge, 2002). In Ricci et al.’s (2002) Intelligent Travel Recommender, the user is told how many results she will get, if any, by eliminating each of