Browse Prior Art Database

An information theoretic approach for customer modeling

IP.com Disclosure Number: IPCOM000015832D
Original Publication Date: 2002-Jul-20
Included in the Prior Art Database: 2003-Jun-21

Publishing Venue

IBM

Abstract

Abstract In this invention, we introduce a method for dynamically updating the customer profile. Individual customer profile is central to the personalization of B2C commerce. The merchant can derive a number of attributes from the customer’s demographics, purchase pattern and interests, e.g., whether or not a customer is loyal, flamboyant, extravagant and/or price-conscious. The derived attributes enable merchant to incorporate his/her business knowledge in the personalization system. Given the abstract nature of some of the derived attributes, it is difficult for a merchant to explicitly define the derived attributes. In this invention, we introduce a new adaptive method to model customer behavior, using information-theoretic criteria. Field of Invention: The invention, in general, is related to the field of personalization, and in particular, to the B2C (business to consumer) e-commerce. In this invention, a way of modeling the individual customer profile is introduced based on an information-theoretic measure. We propose a combination of supervised and unsupervised learning techniques to model the customer behavior.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 12% of the total text.

Page 1 of 13

An information theoretic approach for customer modeling

Abstract

In this invention, we introduce a method for dynamically updating the customer profile. Individual customer profile is central to the personalization of B2C commerce. The merchant can derive a number of attributes from the customer's demographics, purchase pattern and interests, e.g., whether or not a customer is loyal, flamboyant, extravagant and/or price-conscious. The derived attributes enable merchant to incorporate his/her business knowledge in the personalization system. Given the abstract nature of some of the derived attributes, it is difficult for a merchant to explicitly define the derived attributes. In this invention, we introduce a new adaptive method to model customer behavior, using information-theoretic criteria.

Field of Invention:

The invention, in general, is related to the field of personalization, and in particular, to the B2C (business to consumer) e-commerce. In this invention, a way of modeling the individual customer profile is introduced based on an information-theoretic measure. We propose a combination of supervised and unsupervised learning techniques to model the customer behavior.

1. Description of Related Art

In personalization of B2C commerce, merchant's targeting decision depends on the individual customer's demographics and behavior (patent reference [2] and [3]). Constructing profiles from customer's behavior for targeting is important.

Depending on the basic demographic attributes and the record of customer activity, e.g., purchase history, merchant may consider it important to define certain derived attributes (e.g. loyalty, flamboyance, extravagance, quality-consciousness) for better targeting of discounts, coupons, products and advertisements. However, it is often hard to provide a proper definition of such derived attributes. Marketing professionals base their decisions on some of these attributes and have developed heuristic measures for these derived attributes. Heilmann [4] describe a number of definitions of loyalty for modeling of the market for new products. They choose the last brand purchased as the representation of loyalty, an imperfect realization of a household preference that could change over time.

The merchant may have an approximate idea about these attributes. For example, a customer may be called loyal if his purchase frequency exceeds certain threshold. The merchant may use an explicit definition or cite suitable examples to express these ideas. Even if the merchant specifies an approximate model of a derived attribute, he/she may not be able to specify the degree to which the customer possesses that attribute. In some applications, it may be important to specify the degree of loyalty of a customer. Moreover, the real-life customer may not behave

1

Page 2 of 13

according to the merchant's view. Therefore, it is necessary to model the derived attribute by taking into account both the merchant's view and the...