Journal of the Association for Information Systems


Recommender systems are an increasingly important technology and researchers have recently argued for incorporating different kinds of data to improve recommendation quality. This paper presents a novel approach to generating recommendations and evaluates its effectiveness. First, we review evidence that item viewing time can reveal user preferences for items. Second, we model item preference as a weighted function of preferences for item attributes. We then propose a method for generating recommendations based on these two propositions. The results of a laboratory evaluation show that the proposed approach generated estimated item ratings consistent with explicit item ratings and assigned high ratings to products that reflect revealed preferences of users. We conclude by discussing implications and identifying areas for future research.