COD: An adaptive utility learning method for composite recommendations

Abstract

This paper studies and proposes a method for learning the user's preferences and propsing recommendations on composite bundles of products and services. The user preferences are learned using a regression analysis on the historical purchase information. These learned preferences are used to infer a utility axis in the multidimensional utility space. Subsequently, the standard utility axes are adaptivelly adjusted towards the inferred utility axis to generate initial utility axes for the utility elicitation process. The amount by which the axes are adjusted is proportional to the confidence degree. An experimental study is conducted on real data which shows that the proposed method significantly outperforms the standard utility elicitation method in terms of precision of the recommendation set.