The Internet Encyclopedia (Volume 3)

P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML

WL040-Schalkoff WL040/Bidgolio-Vol I WL040-Sample.cls July 16, 2003 15:52 Char Count= 0

SELECTEDINTERNETAPPLICATIONS 243

Electronic Commerce: Recommender

Systems

E-commerce provides many opportunities for advanced

automation, including gathering and using information

on potential customers, buying trends, and product pur-

chases. This information may be used to suggest fu-

ture purchases to previous customers. A knowledge-

based recommender system (KBRS) generates purchase

recommendations by consulting a product information

knowledge base and then reasoning what products will

best satisfy perceived user requirements (Burke, 1999).

One technique for building a recommender system is a

knowledge-based approach called the PersonalLogic sys-

tem. This system helps users make decisions on wide

range of product choices. The system first acquires user re-

quirements in a particular product domain (e.g., automo-

biles) and then consults its knowledge base to find suitable

items that satisfy the users’ requirements. An automobile

shopper could provide requirements such as automobile

type, size, features, and price range, and the system would

then search its knowledge base for automobiles that best

satisfy these requirements.

Online Portfolio Selection

PORSEL (PORtfolio SELector) (Zargham & Mohammad,

1999) is a system for analysis and selection of stocks.

PORSEL allows for fundamental analysis and uses an ex-

pert system. PORSEL is a Web-based client–server system,

with the stock analysis programs and an associated infor-

mation database residing on the server. The Web allows

remote access to the system.

PORSEL uses a fuzzy rule-based system to perform

fundamental analysis. Fundamental analysis primarily

uses information about a company to predict future move-

ments of the company’s stock. This is the approach artic-

ulated by a number of famous investors, thus a large body

of knowledge is available for use in creating rules. Most

rules are derived from well-known rules, such as Graham’s

rule:

If a stock has a price-to-earnings ratio of less than 40%

of the stock’s highest price-to-earnings ratio over the

past 5 years, then the stock’s rating is good.

PORSEL has shown excellent performance when com-

pared with the Standard and Poor’s 500. Sample results

are shown in Table 1. (Note that these results represent ret-

rospective, not prospective, performance.) Here, all shares

were purchased at the beginning of the year, held for the

entire year, and sold at the year’s end. PORSEL then se-

lected 20 new stocks for the the next year. “Equal propor-

tion” means that the same amount was invested in each

of the selected stocks, whereas “Variable proportion” in-

dicates that PORSEL also optimized the relative amount

invested in each selected stock.

Network Monitoring

ExperNet (Vlahavas, 2002) is a multiagent system for

monitoring computer networks, detecting problems, and

Table 1Sample PORSEL Performance

EQUAL VARIABLE S&P 500

YEAR PROPORTION PROPORTION INDEX

1989 55.5 58.8 31.49

1990 73.06 89.5 −3.17

1991 109.3 226.5 30.55

1992 29.1 −4.0 7.67

1993 567.4 573.5 9.99

1994 38.1 44.4 −1.50

Average 145.41 164.78 12.5

S&P=Standard and Poor.

diagnosing the source of problems. Each agent in the sys-

tem is responsible for managing a portion of the network

(e.g., a single agent manages each subnet). Using multi-

ple agents has many advantages, including fault tolerance

and a reduction in the amount of monitoring information

transmitted over the network.

Each agent has a modular structure. The “Device”

knowledge base system comprises the expert system shell

in which rules are implemented. Device is implemented

on top of CS-Prolog II. CS-Prolog II uses an extension of

HNMS network management software, called HNMS+, to

acquire information about the network. The system also

uses a computer monitoring program called BigBrother

to gather information about the computer to which the

agent is attached. This information is used to infer infor-

mation about the network’s performance. Device provides

many interesting features, such as support for multiple

rule types (deductive, production, and event-driven rules)

and object orientation.

OSHA Compliance Monitoring and Advising

OSHA compliance monitoring, for many companies, is a

complex, time-consuming, and important issue. Rather

than providing a static database, OSHA, using the Corvid

ES development tools from EXSYS (http://www.exsys.

com/exsys.html), developed eTools for this purpose.

These tools are available at http://www.osha.gov/dts/osta/

oshasoft/. According to OSHA,

eTools are “stand-alone,” interactive, Web-based

training tools on occupational safety and health

topics. They are highly illustrated and utilize

graphical menus as well as expert system mod-

ules. These modules enable the user to answer

questions, and receive reliable advice on how

OSHA regulations apply to their work site.

Corvid allows the development of interactive experts

that deliver individualized decision-making knowledge

from a Web page or as stand-alone systems. The Exsys

CORVID developer interface provides two ways to view

the underlying system logic through Logic Blocks and in-

dividual rules. The Rule View window shows the full text

of each rule in English, making the rule simple to read

and interpret. A wide range of interfaces and controls are

available for how questions are asked of the system user

and for integration with other software.