Cooperative Query Rewriting for Decision Making Support and Recommender Systems

Abstract

& This article presents a new technology called interactive query management (IQM), designed for supporting flexible query management in decision support systems and recommender systems. IQM aims at guiding a user to refine a query to a structured repository of items when it fails to return a manageable set of products. Two failure conditions are considered here, when a query returns either too many products or no product at all. In the former case, IQM uses feature selection methods to suggest some features that, if used to further constrain the current query, would greatly reduce the result set size. In the latter case, the culprits of the failure are determined by a relaxation algorithm and explained to the user, enumerating the constraints that, if relaxed, would solve the ''no results'' problem. As a consequence, the user can understand the causes of the failure and decide what is the best query relaxation. After having presented IQM, we illustrate its empirical evaluation. We have conducted two types of experiments, with real users and offline simulations. Both validation procedures show that IQM can repair a large percentage of user queries and keep alive the human computer interaction until the user information goals are satisfied.

The performance of recommender systems in online shopping: A user-centric study

“…We show that the use of preference relaxation in product search is a valuable approach to product recommendation. Our user experiment demonstrates that classical approaches to preference relaxation (Mirzadeh & Ricci, 2007) fail to increase consumers' decision-making performance despite promising results achieved in previous work (Dabrowski et al, 2012). We show that although the results of previous studies suggest positive effects of the Standard Preference Relaxation (see Section 3) on decision performance, the strong negative effects on decision effort compromise the potential gain in decision quality.…”

“…In the example above, a consumer using such a product search tool, and who provided preferences on price ($7000 to $8000) and mileage (25000mi to 75000mi) would be presented with a result set with only those offers that fully satisfy all the stated criteria, that is, are both within the price and mileage range. This approach is often referred to as product filtration using hard-constraints or logical filtering (Mirzadeh & Ricci, 2007) and has many limitations acknowledged in the literature (Dabrowski & Acton, 2010b;Felfernig, Mairitsch, Mandl, Schubert, & Teppan, 2009;Mirzadeh, Ricci, & Bansal, 2004), addressed with recommendation method, for example through preference relaxation.…”

Recommendation Technologies for Configurable Products

“…It is a major requirement that state-ofthe-art intelligent systems are flexible and proactive in the way they support users in specifying their requirements and selecting relevant features (Pu and Chen 2008). Various approaches to feature recommendation have already been developedsee, for example Mirzadeh and Ricci (2007) and Thompson, Göker, and Langley (2004). We will now sketch the approaches of collaborative, popularity-based, entropy-based, and utility-based feature recommendation.…”

“…( 1) Both collaborative and popularity-based feature selection help to identify questions of relevance for the user but do not take into account minimality in terms of the number of questions needed (Mirzadeh and Ricci 2007). We will now sketch how to reduce the number of questions and at the same time take into account feature relevance.…”

“…(3) An empirical evaluation by Mirzadeh and Ricci (2007) shows that utility-based feature selection outperforms the entropy and popularity-based approaches in terms of minimizing the number of questions needed in a dialogue.…”