The Internet Encyclopedia (Volume 3)

P1: c-143Braynov-2

Braynov2 WL040/Bidgoli-Vol III-Ch-05 July 11, 2003 11:43 Char Count= 0

60 PERSONALIZATION ANDCUSTOMIZATIONTECHNOLOGIES

Identification:

First name: Joe

Family name: Smith

Personal data:

Age: 20

Gender: male

Job: student

Preferences:

Quality:

Importance: 0.8

Values: low 0.1; medium 0.1; high 0.8

Price:

Importance: 0.7

Values: low 0.6; medium 0.3; high 0.1

Ease of use:

Importance: 0.7

Values: low 0.1; medium 0.1; high 0.8

Figure 3: An example of a user model.

font face and size) that best fits customer preferences and

perceptivity.

In order to provide personalized interaction, SETA em-

ploys user’s models. A user model consists of a fixed part,

containing a set of domain-independent user’s attributes,

and a configurable part, containing the user’s preferences

for domain-dependent product properties. An example of

a user model is given in Figure 3.

Data in a user model are represented as a list of

<feature, value> pairs. The value slot represents a prob-

ability distribution over the set of possible values for a

given feature. For example, the user presented in Figure 3

prefers high-quality products with probability 0.8 and

medium and low quality products with probability 0.1.

The importance slot describes the system’s estimate of the

relevance of a particular preference. Following the previ-

ous example, the user attaches an importance of 0.8 to

product quality.

SETA divides all customers into groups according to

the similarity of their preferences. All customers belong-

ing to one group are described by a group profile called

a stereotype. The classification into stereotypes is used

to evaluate how closely a customer visiting a Web store

matches a stereotypical description. The stereotype clos-

est to a customer’s properties is used for predicting his

preferences, his product selection, and the interface de-

sign. Every stereotype consists of a conditional part and

a prediction part. The conditional part describes the gen-

eral properties of a group of customers. For example, the

conditional part may assert that 30% of customers in a

group are below 30 years of age, 60% of customers are be-

tween 30 and 50, and 10% are above 50 years of age. The

prediction part of a stereotype is similar to a user model

and describes group preferences for product properties.

When a user visits a Web store, the system tries to

predict the user preferences by computing the degree

of matching between the user’s profile and each stereo-

type. The predictions of all stereotypes are merged as a

weighted sum of predictions suggested by each stereotype,

where the weights represent the user’s degree of matching

with a stereotype.

The main advantage of SETA is that the system tai-

lors graphical design, product selection, page content and

structure, and terminology to customers’ receptivity, ex-

pertise, and interests. In addition, it maintains a model

of each customer and of large groups of customers. On

the other hand, SETA depends essentially on customer

registration. The system is unable to observe customers’

behavior in order to automatically build individual pro-

files. Another drawback of SETA is that stereotypes (group

profiles) are prepared manually and, therefore, may not

reflect the actual dynamic properties of customer popula-

tion. An alternative approach is to use data mining tech-

niques that automatically build and dynamically update

group profiles.

Customer Relationship Management

Customer relationship management (CRM) refers to pro-

viding quality service and information by addressing cus-

tomer needs, problems, and preferences. CRM can in-

clude sales tracking, transaction support, and many other

activities. A CRM system usually consists of a database

of customer information and tools for analyzing, aggre-

gating, and visualizing customer information. To achieve

its goal, CRM makes extensive use of personalization and

customization technologies such as log-file analysis, data

mining, and intelligent agents. For example, many CRM

systems store and analyze consumer profiles in order to

develop new products, increase product utilization, opti-

mize delivery costs, etc.

The market for integrated solutions for online CRM

is growing at a rapid pace. BroadVision (http://www.

broadvision.com), for example, provides solutions for

contextual personalization with its one-to-one portal and

one-to-one commerce center. BroadVision combines rule-

based personalization with intelligent agent matching

to dynamically tailor relevant information to customers.