Intelligent Agents

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

E-commerce Intelligent Agents

Aleksander Pivk, Matja Gams

Jozef Stefan Institute, Department of Intelligent Systems,

Jamova 39, 1000 Ljubljana, Slovenia

E-mail: ,

Abstract

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved in buying and selling, followed by lists of Internet e-commerce agents. Several agent-mediated electronic commerce systems are analysed in the context of a general model of the buying process. Several lists of related Internet links should help readers to gather additional relevant information. We presented an intelligent employment agent and e-commerce related agent modules in the system. Our experience indicates that new agent capabilities offer advanced functions in e-commerce.

1 Introduction

In recent years the Internet (World Wide Web) due to its exponential growth enabled substantial progress in new information society functions [10,21] such as online commerce. Latest studies of online spending habits of consumers by Forrester have shown that the growth has been explosive, increasing from $2.4 billion in 1997 to $8.0 billion in 1998 and $20.2 billion in 1999 and still growing at a rapid pace.

Electronic commerce entails business-to-business, business-to-customer and customer-to-customer transactions. It encompasses a wide range of issues including security, trust, reputation, law, payment mechanisms, advertising, ontologies, electronic product catalogs, intermediaries, multimedia shopping experiences, and back office management. Agent technologies can be applied to any of these areas [15].

Still, the potential of the Internet for truly transforming commerce is

largely unrealised to date. Electronic purchases remain mostly non-automated. While information about different products and vendors is easily accessible and orders and payments can be dealt with electronically, a human is still in the loop in all stages of the buying process. Traditional shopping activities require a large effort from a human buyer collecting and interpreting information on

merchants, products and services, making an optimal purchase decisions and finally entering appropriate purchase and payment information [20].

Software agents help automate a variety of activities, mostly time

consuming ones, and thus lower the transaction costs. Software agents differ from “traditional” software in that they are personalized, social, continuously running and semi-autonomous [15]. In this way, e-commerce is becoming more user-friendly, semi-intelligent and human-like. These qualities are conducive for optimising the whole buying experience and revolutionizing commerce, as we know it today [20].

2 Intelligent Agents 2.1 Description

There are many definitions of what the term “agent” denotes based on different approaches, expectations and visions. As pointed out by Bradshaw [2], one person’s “intelligent agent” is another’s person “smart object”.

Shoham [25] describes a software agent as a software entity which

functions continuously and autonomously in a particular environment often inhabited by other agents and processes. The requirement for continuity and autonomy derives from human desire that an agent be able to perform activities in a flexible and intelligent manner responsive to changes in the environment without constant human supervision. An agent that functions over a long period of time should be able to adopt from its experience. Further, we expect an agent to inhabit an environment with other agents and processes, to be able to communicate and cooperate with them, and perhaps move from one place to another in doing so [2].

Consistent with the requirements of a particular problem, each agent

might possess to a greater or lesser degree the following attributes [2,6,8]: ??Reactivity: the ability to selectively sense and act.

??Autonomy: goal-directedness, proactive and self-starting behaviour.

??Collaborative behaviour: can work in collaboration with other agent to achieve a common goal.

??“Knowledge-level” communication ability: the ability to communicate with human and other agents with language more resembling human-like speech than symbol-level protocols.

??knowledge of general goals and preferred methods to achieve flexibility.

Inferential capability: can act on abstract task specification using prior ??Temporal continuity: persistence of identity and state over long periods of time.

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

??character such as emotion.

Personality: the capability of manifesting the attributes of a believable ??Adaptivity: being able to learn and improve with experience.

??Mobility: being able to migrate in a self-directed way from one host platform to another.

Four types of agents can be derived from the characteristics of primary attributes [2,22]: collaborative, collaborative learning, interface and smart (see Figure 1). According to Jennigns and Wooldridge, and Nwana [18,22], more common classification of agents entails eight categories: collaborative agents, interface agents, mobile agents, information/Internet agents, reactive agents, hybrid agents, heterogeneous agent systems, and smart agents.

Figure 1. Typology based on Nwana’s primary attribute dimension

In summary, an intelligent agent is a software entity that carries out

some set of operations on behalf of a human user or another program with some degree of independence or autonomy. In doing so, it employs some knowledge or representation of the user’s goals or desires [13,14]. Intelligence represents the agent’s ability to accept the user’s statement of goals and carry out the task delegated to it. Highest level on the intelligence scale include systems that learn

and adopt to their environment, both in terms of the user’s objectives, and in terms of the resources available to the agent [2].

2.2 Agents and agent repositories

Here one can find links, descriptions and additional information on intelligent agents in practice:

?? ??

?? ?? ??

?? ?? ??

?? ??

Figure 2. Our intelligent agent repository contains lists and short descriptions of

each link ()

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

3 E-commerce 3.1 Description

Electronic commerce is sharing business information, maintaining business relationships, and conducting business transactions by means of communication networks. E-commerce includes the relationship between companies (business-to-business), between customers (customer-to customer), as well as between companies and customers (business-to customer). Business-to-business segment currently dominates the E-commerce while consumer-oriented segment is significantly lagging behind and current estimates place it at less than 10 percent of the total volume, even tough they are all experiencing an exponential growth.

To be successful in e-commerce, a company must build a site that

holds the following properties. It must be innovative, must add value, and provides information and interaction not otherwise available, it must create forums for opinion-building activities. In short, the site must build community and become the “port of entry” for commerce. A good community involves determination of customer interests, where two choices are available: having the customer fill out an interest profile information or having the system analyse a user's activity trails through a web site (e.g., personalization, collaborative information filtering, analysis of consumer buying behaviour, etc.).

On the other hand, company must understand the structure of the e-commerce industry and the roles of its many and varied players. A useful classification scheme for e-commerce providers of products and services is also required. Although several models of the e-commerce industry have been proposed, no validated classification scheme for the industry has appeared. Storey et al. [26] have introduced the structure, which includes the following e-commerce industry classes (describe roles and players in the industry): Network Access Services Providers, E-Commerce-related Hardware Manufacturers, E-Commerce Management Providers, E-Commerce Payment/Billing Services Providers, E-Commerce Payment/Billing Software Providers, E-Commerce Security Providers, E-Commerce Designers/Installers, Server-Side Software Providers, Client-Side Software Providers, Web Integrating Software Providers. It is useful for technology forecasting, trend analysis, planning, and strategizing [26].

The most highly touted applications of E-commerce are consumer-oriented. They include activities related to buying and selling goods or services,

banking, and stock brokerage, accompanied by on-line advertising. Limiting factor for lagging behind the other segments is the settlement phase of

transacting on the Web. The consumer should be able to pay for a purchase on the Web easily and with a perception of security. Customer can also gain a lot from e-commerce such as investing little effort to find a product or service, the best (lowest) price, the latest updates etc. Another advantage is that the stores on the Internet are open 24 hours a day, 7 days a week to allow customers to do business whenever they want. Not only customers gain advantages. Merchants are visible to a global market that represents a great potential, but on down side increases the competition that lowers the prices and profit. With automatic negotiation, merchants could use differential pricing, and products that used to be non-profitable could suddenly become profitable to sell [5].

3.2 Related e-commerce links

Here is a list of most visited e-commerce sites on the Internet: ?? - shopping mall

??more

- offers the biggest selection of books, CDs, and ?? - on-line travel agency ?? - provides a broad range of thoughtful gift products including flowers, gourmet foods, candies, gift baskets and other unique items

?? - offers computer hardware and software, videos, DVDs, games, etc.

?? - offers services involved in car business

?? - one of the world's largest computer manufacturers

??health solutions

- a leading Web site for custom-tailored natural ??games carefully selected by educators - sells educational books, software programs, and

??community. They created a new market: efficient one-to-one trading in an - the world's largest personal online trading auction format on the Web

Relevant Slovenian e-commerce sites:

?? - offers a great collection of Slovenian literature, textbooks, audio and videocassettes, and CDs

?? - offers the biggest collection of various dictionaries, textbooks, handbooks, teaching aids, manuals, printed forms, and numerous musical educations

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

??Slovenia

- the first and the biggest electronic shopping mall in ?? - another shopping mall

?? - offers frozen food and other frozen goods

??local region

- offers customer service and support for ??for the domestic market, services and maintenance

- offers comprehensive supply computer equipment ?? - offers solutions for e-banking, e-commerce, voice-systems, and secure Web applications. In 2000, Zaslon has joined Hermes Softlab.

4 E-commerce Intelligent Agents 4.1 Description

Artificial intelligence (AI) continues to play a significant role in many leading information systems. In the past, its use has been limited due to its complexity, monolithic designs and lack of knowledgeable system developers. AI contribution is now crucial in non-deterministic systems such as workflow, data mining, production scheduling, supply chain logistics, and most recently, e-commerce. Its new form is not the monolithic AI systems of the past, but distributed artificial intelligence, popularly known as intelligent agent technology. Intelligent agent technology is the next logical step in overcoming some shortcomings in e-commerce. Namely, successful computer systems underlying e-commerce require judgment and the knowledge of experts such as buyers, contract negotiators and marketing specialists [7].

It is useful to explore the roles of agents as mediators in electronic

commerce in the context of a common framework. The presented model stems from consumer buying behaviour research and comprises the actions and decisions involved in buying and using goods and services. The model covers many areas, but focuses primarily on retail markets (although most concepts pertain to business-to-business and business-to-consumers markets as well). Also, electronic commerce covers a broad range of issues, some of which are beyond the scope of this consumer buying behaviour model. There are a variety of descriptive theories and models that attempt to capture buying behaviour [20]. These models all share a similar list of six fundamental stages of the buying process, which also elucidate where agent technologies apply to the shopping experience:

??Identification: This stage characterizes the buyer becoming aware of some unmet need by stimulating through product information. Agents can play an

important role for those purchases that are repetitive (supplies) or predictable (habits). One of the oldest and simplest examples of software agents are so called “monitors”: continuously running programs which monitor a set of sensors or data streams and take action when a certain pre-specified condition apply [20]. There are many examples in abundant use, one very familiar is a “notification agent” called “Eyes” by , which monitors the catalog of books for sale and notifies the customer when certain events occur that may be of interest to the customer (e.g., when a new book in category X becomes available). ??Brokering:

a) Product Brokering: once a buyer has identified a need to make a purchase (possibly with the assistance of a monitor agent), the buyer has to determine what to buy through a critical evaluation of retrieved product information. There are several agents systems that lower consumers’ search cost when deciding which products best meet their needs: PersonaLogic, Firefly, and Tete-a-Tete (discussed in section 4.2). The result of this stage is a consideration set of goods. b) Merchant Brokering: this stage combines the consideration set from the previous stage with merchant-specific alternatives to help determine who to buy from. The problem that was exposed here was that most of the merchants do not want to compete on price only, and want the value-added services (e.g., warranty, availability, delivery time, reputation) to be included in consumers’ buying decision.

??Negotiation: in this stage, price and other terms of the transaction are settled on. Real-world negotiation increases transaction costs that may be too high for either consumers or merchants. There are also impediments in the real world to using negotiation such as time-constraints, frustrations, all parties to be geographically co-located etc., which mostly disappear in the digital world. The majority of business-to-business transactions involve negotiation. In retail, we are mostly familiar with fixed prices. The benefit of dynamically negotiating the price for a product instead of fixing it is that it relieves the merchant from needing to determine the value of the good a priori [20]. Rather, this burden is pushed to the marketplace.

??Payment and Delivery: this stage can either signal the termination of the negotiation stage or occur sometimes afterwards (in either order). In some cases, the available payment or delivery options can influence product and merchant brokering.

??Product Service and Evaluation: this post-purchase stage involves product service, customer service, and an evaluation of the satisfaction of the overall buying experience and decision.

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

Given the above set of stages, we can identify the roles of agents as mediators in electronic commerce. The nature of agents make them well-suited for mediating those consumer behaviors involving information filtering and retrieval, personalized evaluations, complex coordination, and time-based interactions. Those roles correspond most notably to the need identification, product and merchant brokering, and negotiation stages of the buying behavior model. Table 1 lists the six buying behavior stages and shows where several representative agent systems fall within the space [20].

Table 1. The online shopping framework with representative examples of agent

mediation

4.2 Examples

Here is a detailed description of agents [20] mentioned in Table 1:

PersonaLogic () - It is a tool that enables consumers to narrow down the products that best meet their needs by guiding them through a large feature space. The system filters out unwanted products within a given domain after a consumer specifies constraints on product features. An agent returns a list of products that satisfy all of the consumer’s hard and soft constraints, by which they are ordered.

Firefly () - This system recommends products via an automated “word of mouth” recommendation mechanism called “collaborative filtering”. The system first compares a consumer’s product ratings with those of other consumers. After identifying the consumer’s nearest neighbors, the system recommends products that neighbours had rated highly but which the

shopper may not yet have rated. Essentially, Firefly uses the opinion of like-minded people to offer recommendations. Such recommendations include commodity products such as music and books, as well as harder to characterize products such as restaurants and web pages.

BargainFinder () - The first virtual shopping agent to query pricing and availability of user specified music CDs. It uses the parallel search architecture used in meta search engines such as MetaCrawler. It submits consumers’ query in parallel to a group of on-line vendors by filling out the form at each site. It parses the query results after filtering out the header, trailer, and advertisements to find each vendor’s price for the required product and prepares a summary to the consumers. BargainFinder offers valuable insights into the issues involved in price comparison in the online world.

Jango () - Jango partially solved the problem of involving value-added (returns a limited set of product features) services to consumer’s requests. In this way, requests to merchants from a Jango-augmented Web browser appeared as requests from “real” customers. This kind of aggressive interoperability makes it convenient for consumers to compare prices from multiple merchants’ online catalogs whether they like it or not.

Kasbah () - It is an on-line, multi-agent consumer-to-consumer transaction system. A user wanting to sell or buy a good creates an agent, gives it some strategic direction, and sends it off to a centralized agent marketplace. Kasbah agents pro-actively seek out potential buyers or sellers and negotiate with them on their creator's behalf. Each agent's goal is to complete an acceptable deal on behalf of its user subject to a set of user-specified constraints, such as an initial asking (or bidding) price, a highest (or lowest) acceptable price, and a date by which to complete the transaction and restriction on what parties to negotiate with and how to change price over time.

Negotiation in Kasbah is straightforward. After buying and selling

agents are matched, the only valid action in negotiation protocol is for buying agent to offer a bid to selling agent with no restriction on time or price. Given this protocol, Kasbah provides buyers with one of three negotiation strategies: anxious, cool-headed, and greedy (see Figure 3), corresponding to a linear, quadratic, or exponential function respectively for increasing its bid for a product over time. The simplicity of this negotiation heuristic makes it intuitive for users to understand what their agents are doing in the marketplace.

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

Figure 3. Price adjustment curves for buying agent [3]

(left to right: anxious, cool-headed, and greedy

Currently, these price adjustments are deterministic (they follow some

function of time), but in the future, there are plans on building agents which can adjust their price according to real-time marketplace factors (the number of other agents selling the same good, the number of interested buyers etc.).

AuctionBot () - It is a multi-purpose Internet auction server developed at the University of Michigan. Its users create new auctions to sell goods by choosing from a selection of auction types and by specifying the parameters (e.g., clearing times, method for resolving bidding ties, the number of sellers permitted, etc.). Customers can then bid according to the multilateral distributive negotiation protocol of the created auction. AuctionBot differs from the other auction sites in providing an application programmable interface (API) for users to create their own software agents to autonomously compete in the AuctionBot marketplace.

Tete-a-Tete (T@T) () - T@T operates in the three core stages, namely the product brokering, merchant brokering, and negotiation. Unlike other shopping and sales agent systems, it caters to the needs of both consumers and merchants by balancing the contention between lowering consumer search costs and differentiating merchants in an otherwise homogenous-looking market-place. It’s unique approach engages consumer-owned shopping agents and merchant-owned sales agents in integrative negotiations over the full value of each product offering to maximize their owners' individual needs as shown in Figure 4. It also helps merchants differentiate themselves along dimensions other than just price. For example, consumers are able to consider delivery time, return policies, support and repair services, loan options, brand and reputation, privacy policies, gift services, extended warranties, and other value-added services in their buying decisions. For example, a shopping agent may receive proposals from multiple sales agents. Each proposal defines a complete product offering including a product configuration, price, and the merchant’s value-added services. The

shopping agent evaluates and orders these proposals based on how well they satisfy its owner’s preferences (expressed as multi-attribute utilities). If the shopper is unsatisfied with the present proposals, he can critique them along one or more dimensions. User’s shopping agent broadcasts this preference changes to the sales agents which, in turn, use them to counter-propose better product offerings.

Figure 4. Consumer-owned shopping agents integratively negotiate with

multiple merchant-owned sales agents

AuctionWeb () and OnSale () These are two popular web sites that sell refurbished and second-hand products using a choice of auction protocols. Likely reasons for their popularity include their novelty and entertainment value in negotiation the price of everyday goods, as well as a potential of getting a great deal on a wanted product [16].

4.3 Lists of e-commerce agents

We have prepared a list of sites where one can find real-life solutions of shopping agents (commentaries on comprehensiveness and accurateness included):

?? (academic)

?? (non-academic)

?? (non-academic)

??academic) (non-

A list of shopping agents:

??Bargain Dogservice

() - online merchant sales alert ??BarPoint specific information including manufacturer contact, comparative price () - enter a bar code number to get product-information, and product reviews and reports

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

??BigCompare shopping technology that constantly searches the web for products () - providing mega-search ??buyer at price specified by user

() - service that locates products for ??Click The Button () - software that allows users to compare prices of books, consumer electronics, movies, music, and more

??Dash () - personal shopping assistant offers rebates from selected merchants online

??DealTime () - searches retailers, auctions, and classifieds for the best price

??eBoodle () - free, downloadable, personal shopping assistant

??GoTo Comparison Shopping () - helps you find and compare products you want to buy ??iChoose () - searches for better deals while you shop ??LYCOShop Product Comparison Servicebased comparison shopping engine that searches for a product based on () - value-personal input

?? ()- helps user find products and compare prices

?? () - deals and specials alerts from a variety of merchants

??PriceScan () - takes the hassle out of finding the best price on thousands of computer hardware and software products

??R U Sure () - comparative online shopping search engine. Features reviews, classifieds, message boards, directories, auctions, and consumer comments

??Sales Channelsends email updates when department stores are having a sale

() - free news service that ?? () - compares prices and features for products in local retail stores ??SALEseeker () - free online service that helps you find sales and specials at local, department, specialty stores and catalogs

??Shopping Sleuth () - tracks news and information on a variety of products and sends e-mail updates to users

?? () - offering in-store sales, specials, and personal sale alerts

??Siliconrex software, electronics, movies, music, and more

() - links to computer hardware and ??ValueFind and classified ad sites for the best values on millions of products

() - search online merchants, auctions, ??ValueSpeed () - compares prices for books and selected other products by adding a button to the Web browser

??Valulinksshopping tips for brand nam (e merchandise and wholesale products

) - membership club providing discount 5 Intelligent employment agent 5.1 Description

E-commerce domains indirectly include employment tasks. Internet employment systems offer information about desired objects, which differ in properties, benefits and costs. Users browse through a large number of potential candidates and compare several similar possibilities to finally choose specific objects. Users typically want to get a job - any job, a good job, better than current. Benefits are the major factor for evaluation of several possibilities. Jobs and available workers are among most valuable human resources. Unemployment is one of major problems in developed countries although trends differ in various states. Situation in Slovenia is similar to that in Europe. With 2.000.000 inhabitants there are around 120.000 unemployed workers and 100.000 new jobs each year. A typical task of an employment agent is to find relevant information from various information sources, provide it to users in a human-like way, and try to match workers and jobs as efficiently as possible (see Figure 5). This is also the aim of our intelligent employment EMA. It provides user-friendly information on demand or when it notices relevant information. It tries to find relevant information from various information sources autonomously and with collaboration with other actors n the Internet. EMA stores search patterns, performs search on its own or following direct commands, sends online replies or offline emails regarding vacant jobs or available workers. EMA can store and observe interesting WEB sites chosen by users, and match available jobs and workers.

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

Figure 5. The Internet-based intelligent employment agent provides

employment information as human agents do.

EMA has been in use for over seven years, all the time being further

developed. It was internationalised with the INCO-Copernicus Project: 960154, Cooperative Research in Information Infrastructure (CRII) (). The Internet site of the EMA system is available at ). The Employment Services of Slovenia (national employment office) site is at .

agents.

EMA follows general ideas of humanizing computers through human-like communication. In particular, text and speech modules are introduced as supplements to classical communication. A common input is a partially constrained Slovenian text. In bulletin boards, the language is a matter of choice, but titles and information are either in Slovenian or in English. Consequently, majority of text is either Slovenian or English, with a couple of exceptions. No censorship is performed regarding language or specific details as long as input if dedicated to the desired employment task and inside minimal decency requirements. The system is capable of translating Slovenian text into English. The translation is based on a dictionary consisting of up to four words observed before in the employment data. New combinations are in the worst case translated as direct word-by-word translation and stored for further overview by humans. Stored combinations are sorted by frequency and translated by humans if reasonable. In addition, the translation system looks into the morphology dictionary to capture different forms of the same words. Finally, a spell-checker module corrects spelling errors. The translation is currently not yet at the level performed by systems translating between larger European languages; however, it is sufficiently good to enable understanding since the syntax is quite limited.

There are two speech modules, one for Slovenian and one for English

language. The English speech system is based on the Microsoft agent (see Figure 6). Our Slovenian speech module is freely available on the Internet (). The system was implemented for Slovenian language based on the concatenation of basic speech units, diphones, using TD-PSOLA technique improved with a variable length linear interpolation process [24]. It is slightly less understandable then the English system speaking English, however, is specialized for Slovenian employment information and is more understandable than adapted general speech systems. The ability to represent information in several languages and both through typing and speech is an important user-friendly ability. Agents are humanizing computers and their important property is communication with humans in a human-understandable way. The English and Slovene speech modules showed that this methodology is ripe for generating audible speech from text in English or Slovenian (the word intelligibility rate was around 90%). The quality of the system is inferior to human speech, but improvements are noticeable compared to older systems. It seems that speech and translation technologies are ripe for implementation in any intelligent system. On the other hand, most of the queries through the Internet are performed without speech options. Hopefully, the speech system will be intensively used when the application becomes operational through mobile phones.

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

During seven years a substantial amount of energy was related to

software engineering [21]. Our employment agent system is an integrated heterogeneous software system consisting of several modules and submodules, written in different programming languages and running on different platforms and operating systems. The system is a 30.000 lines program written mainly in C, partly in other languages and by Internet tools. Together with text and data it occupies 30 MB on a disk. There are over 200.000 visits monthly which represents one tenth of the whole population in Slovenia.

EMA is among most successful applications of intelligent agents in

this part of Europe. In the first year of its implementation, our country was the forth in Europe to offer national employment information through the Internet. At that time, we were the first country in the world to provide over 90% of all nationally available jobs on the Internet. On the other hand, in absolute terms there are employment systems in big countries or employment systems connected to major Internet information providers such as Yahoo or AltaVista that provide orders of magnitude bigger amounts of employment information (). In addition, now we provide substantionally less than 90% of all available jobs due to policy that any amployment subject can on its own decite so send information about available jobs to the Internet or not. Currently, the system runs in its professional application version at national employment office, while at the Jozef Stefan Institute we maintain a prototype mirror with several modules in development.

There are several modules in progress. One is an agent that performs

multiple parallel searches on various search engines, and then visits each Web site in the results and downloads that page. The search agent then analyzes the downloaded results and thus gathers new valuable information on specific domains and how to represent it to users. Another system learns ontologies by learning which job definition corresponds text description of a job application of description of a worker. In this way the system is able to act more sensibly in a particular domain. A couple of other modules are described in the following sections.

5.2 Information integration

Inside the integrated agent [1] there are several single agents, e.g. the speech agent. Agents can be grouped into functional units (see Figure 7), e.g., into NLP agents: speech agents, translating agents; communication agents: database agents, Internet agents, local agents, etc. An important group of agents deals with computer-computer communication, e.g. with HTML sites, and Internet-accessible databases. The Internet-connections group consists of agents, programs and lists of HTML sites. Users can search interesting sites using their

own programs, look at interesting employment links stored by EMA, or use EMA's search programs. A user is typically interested in getting information from HTML sites relevant to him/her. Users can store addresses of interesting sites, and order EMA to inform them when a particular site changes. In a previous implementation of the module, EMA also supplied the changed text by e-mail, but users in polls preferred just the information that a change had occurred.

Figure 7. The integrated agent incorporates several communication agents.

5.3 Uniform Internet database access – improved database wrappers

Humans usually access several databases nationally and internationally. Internet-based employment databases are regardless of their geographical location usually accessible through particular HTML forms. We tried to copy human-type of communication by developing an agent capable of first observing and later exploiting communication with the database. For example, if an HTML form demands "Profession", it means that the profession of a worker or an available job should be typed in according to the type of activity. By storing these patterns, EMA should be able to cope with common employment input fields. The small differences in patterns should be overcome by soft matching [11,12]. During the process, EMA should learn new patterns. While all functions remain a near-future research aim, we have implemented an experimental prototype based on these ideas that already connects to a couple of databases

through the Internet.

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

Database wrappers are programs that enable access to Internet

databases. They are programs written by a human programmer without intelligent additions. The communication module in EMA can be represented as an improved intelligent database wrapper. It connects to a potential database in three steps. First it observes a user communicating with an interesting employment base. At this stage of application, EMA's superuser starts particular programs and thus also provides information that the database is relevant for the system. In the second phase, EMA remembers communication patterns. Finally, EMA uses stored patterns for future queries. There are several simple solutions regarding the desired task. One is to store the whole filled form without any knowledge about any particular field, and resubmit it when appropriate. But then only one query can be performed, and in addition, any change in the input form should cause intervention by a superuser. In another simple solution, EMA could only repeat the last specific query posted by the user. E.g., when looking for a "Postman" in specific geographic area, EMA could remember the query and reapply it to a particular database when the same or very similar query appears. In this case, users should better provide a very large number of possible queries in order to enable EMA to perform them. Still, a certain amount of all queries would be new and the system would not be able to react reasonably in such situations.

EMA possesses additional ontology knowledge about low-level fields

in forms such as "date" or "submit". With this knowledge, EMA can hopefully cope with reasonable changes in the form. In reality, since EMA has to provide serious information 24 hours a day, EMA checks the input form before sending a query on its own. If a previous successful form or query exists, i.e. if a simple solution exists, EMA performs it. If no remembered form or query exists, i.e. because no such thing was encountered before, EMA creates the query/form on its own by filling it step by step. Therefore, EMA combines several approaches and uses the simplest one possible. But without advanced possibilities, EMA would only be capable of primitive replicating of observed commands.

There have been several attempts to combine different data structures

together from different sources [27]. Recently, major computer companies including IBM, Microsoft or Ericsson are intensively working on this subject. Technologically, the market of Internet services is moving to mobile telephones demanding integration of several information sources. Another technological push is from merging TV, a computer, a mobile telephone etc. into a smart house. A common approach is to put an agent (or an interface) on top of a specific information source. For databases, these interfaces are sometimes called "database wrappers" especially when data warehouses are imposed over single databases. Another relevant technique is to put an agent on top of arbitrary data structure, and thus transform programs into agents [4]. However, there is a price to pay in these approaches. From a practical aspect, these systems have to be adapted for each different information source; therefore they need substantial amount of human intervention. Theoretically, it has been shown by Eiter et al. [4] that by providing more sophisticated semantics of such agent systems these semantics become epistemically more desirable, but they demand higher computational efforts.

Our system uniformly transforms different information sources into

plain text. Therefore, the internal data structure of the EMA system is a list of records of plain text. There are different lists for different tasks; e.g., there is a list of vacant jobs, a list of available workers etc. When loaded or input into local databases, these text records get structured because of databases; however, the uniform internal presentation is a list of records, and EMA performs all its basic functions there. Such an approach has certain advantages and disadvantages. The major advantage is the universality of data. Whatever the input, e.g., a database or an HTML form, the system is able to perform information retrieval on its text copy. The major problem is presentation of the text on the Internet for users. In the search process there is no problem with a couple of strange commands e.g. in JavaScript. But when the agent shows application-specific commands on the Internet, these commands can cause strange outlook or even cause error. If the system shows only data it recognizes (e.g. profession), relevant information can get lost.

To perform its tasks, EMA must use low-level Internet protocols and

commands. Only when formulating the final query, EMA translates low-level commands into a normal Internet level. Databases do not notice any difference compared to normal human communication. One chunk of knowledge is one HTML command. Chunks are stored in lists as text strings with additional information about statistical success rate, number of uses; there is some additional information regarding the database, query etc. In addition, EMA has knowledge about HTML, employment, and forms coded in their stored memory patterns. EMA stores not only the visible HTML fields, but also the input text written by a user, and the source HTML text. If the input information such us name, date, profession etc. is familiar to EMA, it can later replace it with stored patterns from other users. Since the basic employment information tasks are quite frequent, techniques based on frames or memory-based learning can be applied [19]. However, EMA is currently fairly limited in its capability to adapt to previously unfamiliar patterns.

An important part of any flexible system is learning. In EMA, learning

is performed as storing new patterns [11,12], and as changing statistics about patterns. With stored patterns, knowledge about previous events is introduced. It

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

is enhanced by domain knowledge encoded as program routines. For example, when looking for jobs one has to input specification of a particular job. It can happen that there is just one open text field in the looking-for-job forms, and the task of finding the relevant field to input job specification is completed. Many other fields have default values indicating the data type. The assumption in EMA is that there will be at most a couple of necessary modifications of the query filing the HTML form, and that they will be correctly constructed based on the domain knowledge and stored patterns.

These techniques can represent an important improvement in Internet

communications and e-commerce. Practically the same techniques can be applied in e-commerce by observing any user and freely add other databases. Currently one of the problems is related to reliability and legal questions. Our actual experimental application is severely restricted compared to the possibilities of the general approach. Future will show whether the proposed approach is generally applicable also in other problem domains.

6 Discussion

Agents are a key component in the Internet-wide information and electronic commerce systems that are currently being developed across the globe. But there is still a long way before software agents will significantly transform how businesses conduct business. The greatest change will occur once standards are adopted and evolved to unambiguously and universally define goods and services, consumer and merchant profiles, value-added services, secure payment mechanisms, complex goals, changing environment, etc [20].

Advanced third-generation of agents in electronic commerce system

will explore new types of transactions in the form of dynamic relationships among previously unknown parties [20]. That is the generation where companies will be at their most agile and marketplaces will approach perfect efficiency. The ultimate test of agent's success will be the acceptance and (mass) usage by users. The road to the success is most likely to be laid by developers, suppliers, and many commercial companies, who will join in, as there are many interesting opportunities for them. However, there are a few important points that need to be settled before this can really be done well. Solid standards need to be established such as common agent communication language, common marketplaces, etc. [17]

But the potential advantages of intelligent agents are huge. With

additional knowledge and intelligent methods intelligent agents offer substantial improvements in e-commerce.

Acknowledgement: This work was supported by an international project INCO-Copernicus 960154, Cooperative Research in Information Infrastructure, CRII, by the Ministry of science and technology in Slovenia, and by ESS. We would like to thank the CEO of the Employment Service of Slovenia, Mr. J. Glazer.

7 References

[1] Basu, A. and R. Blanning, “Model integration using metagraphs”, Information Systems Research, 5(3), (1994).

[2] Bradshaw, J.M., “Software Agents”, in: Software Agents, ed. J.M. Bradshaw, AAAI Press, Menlo Park, California (1997).

[3] Chavez, A., D.Dreilinger, R.Guttman, and P.Maes, “A Real-Life Experiment in Creating an Agent Marketplace”, Proceedings of the Second International Conference on the Practical Application of Intelligent Agents and Multi-Agent Technology (PAAM'97), London, UK, (April 1997).

[4] Eiter, T., V.S.Subrahmanian and G.Pick, "Heterogeneous Active Agents, I: Semantics", Artificial Intelligence 108, 1-2, 79-255 (1999).

[5] Eriksson, J. and N.Finne, “MarketSpace: an open agent-based market infrastructure”, Uppsala Master’s thesis in Computing Science 106, Examensarbete DV3, (1997).

[6] Etzioni, O. and D.S.Weld, “Intelligent agents on the Internet: Fact, fiction, and forecast”, IEEE Expert, Vol.10, No.4, 42-49 (1995).

[7] Fingar, P., “A CEO's Guide to eCommerce Using Object-Oriented Intelligent Agent Technology”, (June 1998), .

[8] Franklin, S. and A.Graesser, “Is It an Agent or just a Program? A Taxonomy for Autonomous Agents”, In Proceedings of the third International Workshop on Agent Theories, Architectures, and Languages. New York: Speinger-Verlag. [9] Ga, M., “Agents in E-commerce” (guest editor), Communications of the ACM, Vol. 42, No.3, 79-80, ACM Press, (March 1999).

[10] Gams, M., “Information Society and the Intelligent Systems Generation”, Informatica 23, 4, 449-454, (1999).

[11] Gams, M., A.Karalic, M.Drobnic and V.Kri man: "EMA - An Intelligent Employment Agent", Proc. of the Forth World Congress on Expert Systems, Mexico, 57-64, (1998).

[12] Gams, M. and B.Hribov ek, "Intelligent-Personal-Agent Interface for Operating Systems", Applied Artificial Intelligence, Vol. 10, 353-383 (1996).

[13] Gilbert, D., M.Aparicio, B.Atkinson, S.Brady, J.Ciccarino, B.Grosof, P.O’Connor, D.Osisek, S.Pritko, R.Spagna, and L.Wilson, “IBM Intelligent Agent Strategy”, IBM Corporation, (1995).

Intelligent agents have been applied to electronic commerce, promising a revolution in the way we conduct business, whether business-to-business, business-to-customer or customer-to-customer. This article gives a brief review of agent technologies involved

[14] Gini, M., “Agents and other ‘Intelligent Software’ for e-Commerce”, CSOM, February (1999).

[15] Gutmann, R.H., A.G.Moukas, and P.Maes, “Agents as Mediators in Electronic Commerce”, Electronic Markets, Vol.8, No.1, 22-27 (1998).

[16] Guttman, R.H. and P.Maes, “Cooperative vs. Competitive Multi-Agent Negotiations in Retail Electronic Commerce”, Proceedings of the Second International Workshop on Cooperative Information Agents (CIA'98), Paris, France, (July 1998).

[17] Hermans, B., “Intelligent Agents on the Internet: an inventory of currently offered functionality in the information society & a prediction of (near-) future developments”, Tilburg University, The Netherlands, (March 1997), .

[18] Jennings, N.R. and M.J.Wooldridge, “Agent Technology, Foundations, Applications, and Markets”, Springer-Verlag Berlin, (1998).

[19] Kitano, H., "Challenges of Massive Parallelism", Proceedings of the 13th International Joint Conference on Artificial Intelligence, 813-834 (1993).

[20] Maes, P., R.H.Gutmann, and A.G.Moukas, “Agents that Buy and Sell: Transforming Commerce as we Know It”, Communications of the ACM, ACM Press, Vol. 42, No.3, 81-91 (March 1999).

[21] Mueller, J.P., “The Design of Intelligent Agents”, Springer, (1996).

[22] Nwana, H.S., “Software Agents: An Overview”, Knowledge Engineering Review, Vol.11, No.3, 1-40 (September 1996), .

[23] Pivk, A., “Electronic trading with application in Java”, Graduation thesis, University of Ljubljana, Slovenia, June 1999.

[24] Sef, T., A.Dobnikar and M.Gams, "Improvements in Slovene Text-to-Speech Synthesis", Proceedings of the 5th International Conference on Spoken Language Processing (ICSLP'98), Sydney, Australia, 2027-2030 (1998).

[25] Shoham Y., “An Overview of Agent-oriented Programming”, in: Software Agents, ed. J.M. Bradshaw, AAAI Press, Menlo Park, California (1997).

[26] Storey, V.C., D.W.Straub, K.A.Stewart, and R.J.Welke, “A Conceptual Investigation of the E-commerce Industry”, Communications of the ACM, ACM Press, Vol. 43, No.7, 117-123 (July 2000).

[27] Wobcke, W. and A.Sichanie, "A Reactive Scheduling Agent Architecture for Coordinating Autonomous Agents", Proc. of the 1st Asia-Pacific Conference on Intelligent Agent Technology, Hong Kong, 28-37 (1999).

Aleksander Pivk graduated at the Faculty of Computer and Information Science in 1999. He is a post gradute student at the same faculty and a young researcher at biggest Slovenian research institute where he reserches methods of artificial intelligence (machine learning, data mining,

decision support) in the area of electronic commerce, more specifically e-trading. As a resercher, he also actively collaborates in the European research project Sol-Eu-Net.

Prof. Dr. Matjaz Gams () is an associate professor of computer science and informatics at the Ljubljana University and senior researcher at the Jozef Stefan Institute, Ljubljana, Slovenia. He teaches several courses in computer sciences at graduate and postgraduate level. His research interests include artificial intelligence, intelligent systems, intelligent agents, machine learning and cognitive sciences. His publication list includes over 200 items, 46 of them in scientific journals and 77 in proceedings. He is an executive contact editor of the Informatica journal () and editor of several international journals. Among others, he is member of the governmental board of the JS Institute, currently president of two societies, cofounder of the Engineering Academy of Slovenia and the Artificial Society in Slovenia. He headed several major applications in Slovenia including the major national employment agent on the Internet, the expert system controlling quality of practically all national steel production, and the Slovenian text-to-speech system donated to several thousand users.

本文来源：https://www.bwwdw.com/article/qrfj.html