Modeling The Price-Relevance Relationship To Drive Users’ Decision Making

Very little research has studied how price affects users' behavior when using a Recommender System (RS). This research argues that the way the relationship between price and relevance is modeled in extant RS is appropriate to provide high accuracy, but may not be appropriate for driving users' decision making and providing high business performance, such as sales. We propose to model this relationship in each item category as an alternative to building a model independently of item categories, which is the model implicitly used by extant RS. We compare the accuracy and business performance of these two models through an offline experiment. We find that changing the way the price-relevance relationship is modeled in a RS may dramatically contribute to drive users' decision making and keep accuracy high. This suggests important implications on the diffusion of RS in industrial applications