High-utility rule mining for cross-selling

Abstract

In association rule mining, utility has recently been regarded as a practical measure for a rule's usefulness in that it can reflect the actual amount of output achieved by applying each rule. Even the same rule may have different utilities depending on how well the rule fits a specific business purpose. However, most recent studies have tried to apply a uniform standard to assessing rules disregarding this. This paper introduces high-utility rule mining (HURM) as an alternative approach. HURM proposes rule utility as a new measure for rules' usefulness. Rule utility, expressed in the form of a rule utility function (RUF), can be developed from three elements (opportunity, effectiveness, and probability) that are designed by considering a rule's fitness to business purposes. HURM algorithms were developed to meet each specific purpose by replacing RUFs. A cross-selling case was chosen to show how HURM can be applied to a particular business.