Brought to you by EarthWeb
IT Library Logo

Click Here!
Click Here!

Search the site:
 
EXPERT SEARCH -----
Programming Languages
Databases
Security
Web Services
Network Services
Middleware
Components
Operating Systems
User Interfaces
Groupware & Collaboration
Content Management
Productivity Applications
Hardware
Fun & Games

EarthWeb Direct EarthWeb Direct Fatbrain Auctions Support Source Answers

EarthWeb sites
Crossnodes
Datamation
Developer.com
DICE
EarthWeb.com
EarthWeb Direct
ERP Hub
Gamelan
GoCertify.com
HTMLGoodies
Intranet Journal
IT Knowledge
IT Library
JavaGoodies
JARS
JavaScripts.com
open source IT
RoadCoders
Y2K Info

Previous Table of Contents Next


3. FROM DECISION AID TO TRUE MARKETING

In a most recent and comprehensive survey (Burke, 1994) of expert systems for marketing, the systems were classified into two categories: ones for everyday decisions and ones for strategic decisions. Previously, AI techniques were used only for marketing decisions rather than for marketing itself. In other words, the users of the intelligent marketing systems were not customers, but rather marketing managers, marketing analysts, advertising agents, and salesmen. While marketing managers were focusing on their own users, a new highway to customers was growing, and now, with the advent of the World Wide Web, marketers must be ready to utilize this revolutionary new medium (Hoffman and Novak, 1996; Quelch and Klein, 1996). An early response in expert system research to this new opportunity was UNIK-SES (Lee, Lee, and Lee, 1996), which will be described in this chapter. Lee, Lee, and Lee described an expert system with customized purchasing support; the intended user was not marketers, but the end-customers. This is an effort to integrate interactive marketing (Blattberg and Deighton, 1991) with database marketing (Cespedes and Smith, 1993) in the World Wide Web environment.

4. EXPERT SYSTEM FOR CUSTOMIZED PURCHASING SUPPORT

The existing two-way home shopping systems have so far been predominantly hypertext and database oriented. Thus, their major role is providing interactive retrieval of text and data (including figures) about product items. Since neither the hypertext and database approach nor the current expert systems can effectively support the selection of the customer's personalized preference, we propose to develop an expert system named UNIK-SES (Salesman Expert System) that can support the generation of a custom-tailored product configuration. To allow the natural interactive setting of the customer's preference and constraints, the Constraint and Rule Satisfaction Problems (CRSP) approach is adopted (Lee and Kwon, 1995). A particular domain that we have attempted to study is the process of purchasing menswear through an electronic market.

CRSP consists of a set of variables, an associated domain for each variable, and a set of constraints and rules to serve as relationships among the variables. The first advantage of adopting the CRSP is its representational richness. In UNIK-CRSP, four types of constraints are allowed: value compatibility (and incompatibility), algebraic inequality, mandatory equality, and functional assignment (Lee and Kwon, 1995).

The distinctive characteristics of the reasoning in UNIK-CRSP can be identified by three features: concurrent, integrated, and interactive reasoning. What we mean by concurrent reasoning is that a user can select the variables that he or she is most concerned about (goal variables or tightly constrained variables, so called "seed variables") as the starting point of the reasoning. Then each seed variable with a target value initiates the propagation concurrently until it reaches a variable(s) from different directions of propagation that has conflicting value requirements. For each individual propagation, the satisfaction of both the constraints and rules is pursued. This is why this reasoning process is called integrated reasoning. If a variable cannot simultaneously meet the requirements of both directional propagations, the customer is called on to decide which goal should be treated with a higher priority. Then the reasoning mechanism traces back to find out how much the level of lower prioritized goal must be degraded from the initial target. Thus, interactive reasoning is necessary to help in the tradeoff process, and the overall reasoning process devised for UNIK-CRSP matches very well with the human's natural decision-making process.

5. ARCHITECTURE OF SALESMAN EXPERT SYSTEM

The architecture of UNIK-SES is depicted in Figure 1. It is equipped with five types of knowledge: product database, customer database, constraints about customers and products, rules about customers and products, and sales strategic rules. UNIK-SES consists of two reasoning capabilities: a CRSP Engine that configures the specification of the products, and a Product Matcher that retrieves the product component instances that meet a given requirement. The product database matcher must first initialize the domain of values for each decision variable. Since the values should be updated whenever there is any change in the product database (e.g., introduction of new products, price change, and stock out), a data-driven, forward-chaining capability is called for. So, a forward-chaining subsystem UNIK-FWD is employed along with UNIK-CRSP.


FIGURE 1 Architecture of the UNIK-SES.

The product configuration process requires two classes of constraints among variables. One is an ordinary constraint between product types. For instance, the suitable color-matching between jacket and slacks belongs to this category. The other is a constraint within a product type. The latter constraint implies that the existence of such a product instance should be confirmed. Since imposing the within-product-type constraints for each product instance is very burdensome, it is better to store the instance items in an object-oriented database and go to confirm its existence. To operate in this manner, the domain of values should be dynamically arranged by retrieving the existing products that meet the user's requirement. For this reason, the product database matching capability should be closely embedded in the CRSP reasoning process for product specification.

The following steps summarize the CRSP reasoning process for the Salesman Expert System incorporating the product database matching routine.

Step 1. Goal setting. A customer selects the goal variables that he or she is interested in (e.g., types of products) and sets the desired target levels (e.g., available budget).
Step 2. Variable ordering. UNIK-SES computes the tightness of variables, and orders the variables according to the tightness measure and customer-dependent variable ordering rules.
Step 3. Seed variable selection. Customer selects seed variables from which the concurrent reasoning starts. The seed variables may be determined either by customer's choice or by selecting the variables whose tightness is above a threshold. Set the currency to each of the seed variables.
Step 4. Rule and constraint propagation. For each seed, propagate rules and constraints as described in steps 4a through 4d.
Step 4.1. Rule inference. Perform the backward-chaining inference taking the current variable (Xc) as a root node.
Step 4.2. Functional assignment. If there exists a functional assignment constraint that has Xc as a dependent variable, compute the value of Xc by asking the customer the values of independent variables, or by deriving the values of independent variables from the associated constraints and rules. Assign the computed value to Xc.
Step 4.3. Value ordering. To propagate the associated compatibility constraints with Xc, order the values of Xc by the least constraining order criterion and customer-dependent value ordering rules. Select a value of Xc to propagate with.
Step 4.4. Compatibility constraint propagation and product matching. With the selected value, propagate the associated compatibility constraints with Xc one by one, following to the constraint ordering criteria and considering importance of constraints.
Match the current requirement specification derived from the constraint propagation with the product database to retrieve the products that satisfy the requirements.
IF the propagation succeeds AND products satisfying the specification exist
IF the active values of the variables are changed by the current propagation,
THEN perform truth-maintenance for consistency among multiple paths of constraint propagation.
ELSE move the currency to one of propagated variables and go to step 4a.
ELSE
IF there exists a chance of backtracking,
THEN backtrack the variable by moving the currency via the traced path. Go to step 4a.
ELSE go to Step 5.
Step 5. Inconsistency elimination and goal tradeoffs.
Identify the reason of backtracking failure.
IF the failure is associated with a goal variable,
AND adjustment of the current goal is possible,
THEN relax the goal to resolve the backtracking failure.
IF the failure is caused by an inherent inconsistency between multiple goal variables,
THEN resolve the inconsistency by negotiating between associated goal variables.


Previous Table of Contents Next

footer nav
Use of this site is subject certain Terms & Conditions.
Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details.