Profit Mining

Profit Mining

Senqiang Zhou (Simon Fraser University, Canada)
Copyright: © 2009 |Pages: 5
DOI: 10.4018/978-1-60566-010-3.ch244
OnDemand PDF Download:
No Current Special Offers


A major obstacle in data mining applications is the gap between the statistic-based pattern extraction and the value-based decision-making. “Profit mining” aims to reduce this gap. In profit mining, given a set of past transactions and pre-determined target items, we like to build a model for recommending target items and promotion strategies to new customers, with the goal of maximizing profit. Though this problem is studied in the context of retailing environment, the concept and techniques are applicable to other applications under a general notion of “utility”. In this short article, we review existing techniques and briefly describe the profit mining approach recently proposed by the authors. The reader is referred to (Wang, Zhou & Han, 2002) for the details.
Chapter Preview


It is a very complicated issue whether a customer buys a recommended item. Consideration includes items stocked, prices or promotions, competitors’ offers, recommendation by friends or customers, psychological issues, conveniences, etc. For on-line retailing, it also depends on security consideration. It is unrealistic to model all such factors in a single system. In this article, we focus on one type of information available in most retailing applications, namely past transactions. The belief is that shopping behaviors in the past may shed some light on what customers like. We try to use patterns of such behaviors to recommend items and prices.

Consider an on-line store that is promoting a set of target items. At the cashier counter, the store likes to recommend one target and a promotion strategy (such as a price) to the customer based on non-target items purchased. The challenge is determining an item interesting to the customer at a price affordable to the customer and profitable to the store. We call this problem profit mining (Wang, Zhou & Han, 2002).

Most statistics-based rule mining, such as association rules (Agrawal, Imilienski & Swami, 1993; Agrawal & Srikant, 1994), considers a rule as “interesting” if it passes certain statistical tests such as support/confidence. To an enterprise, however, it remains unclear how such rules can be used to maximize a given business object. For example, knowing “Perfume→Lipstick” and “Perfume→Diamond”, a store manager still cannot tell which of Lipstick and Diamond, and what price should be recommended to a customer who buys Perfume. Simply recommending the most profitable item, say Diamond, or the most likely item, say Lipstick, does not maximize the profit because there is often an inverse correlation between the likelihood to buy and the dollar amount to spend. This inverse correlation reflects the general trend that the more dollar amount is involved, the more cautious the buyer is when making a purchase decision.

Complete Chapter List

Search this Book: