Fast computation of query relaxations for knowledge-based recommenders

Article type: Research Article

Authors: Jannach, Dietmar

Affiliations: Department of Computer Science, TU Dortmund, Dortmund, Germany. E-mail: [email protected]

Abstract: “No matching product found” is an undesirable message for the user of an online product finder application or interactive recommender system. Query Relaxation is an approach to recovery from such retrieval failures and works by eliminating individual parts of the original query in order to find products that satisfy as many of the user's constraints as possible. In this paper, new techniques for the fast computation of “user-optimal” query relaxations are proposed. We show how the number of costly catalog queries can be minimized with the help of a query pre-processing approach, how we can compute relaxations that contain at least n items in the recommendation, and finally, how a recent conflict–detection algorithm can be applied for fast determination of preferred conflicts in interactive recovery scenarios.

Keywords: Knowledge-based and constraint-based recommender systems, query relaxation, user interaction

DOI: 10.3233/AIC-2009-0454

Journal: AI Communications, vol. 22, no. 4, pp. 235-248, 2009