Deploying Mobile and Intelligent Agents in ... - Semantic Scholar

4 downloads 101762 Views 369KB Size Report
efficiencies and benefits especially in business-to-business (B2B) e-commerce. ... automation of decision-making as well as enablement of new communication ...
2003 Society for Design and Process Science Printed in the United States of America

DEPLOYING MOBILE AND INTELLIGENT AGENTS IN INTERCONNECTED E-MARKETPLACES Ryszard Kowalczyk Swinburne University of Technology, P.O. Box 218, Hawthorn 3122, Australia

Peter Braun Ingo Mueller Wilhelm Rossak Friedrich Schiller University, Ernst-Abbe-Platz 1-4, D-07743 Jena, Germany

Bogdan Franczyk University of Leipzig, Marschnerstr. 31, D-04109 Leipzig, Germany

Andreas Speck Intershop Research, Intershop Tower, D-07740 Jena, Germany

The business trends to trade more complex goods and services in inter-connected e-marketplaces and to use mobile communication channels drive the requirements for the next generation of emarketplaces including more advanced trading, mobile access and user support capabilities. This paper presents an attempt to address the above requirements by proposing an intelligent mobile agent-based e-marketplace solution called InterMarket that integrates mobile agents and intelligent decision-making agents in e-marketplaces. InterMarket aims at enabling mobile access and automated trading in inter-connected e-marketplaces with an add-on component to an existing commercial e-marketplace platform. The paper presents a general concept and solution of InterMarket, and discusses the proposed approach in the context of related research in mobile ecommerce agents and intelligent agent-mediated e-commerce. Keywords: E-marketplace, InterMarket, mobile agent, intelligent agent

1. Introduction E-commerce offers new channels and business models for buyers and sellers to effectively and efficiently trade goods and services over the Internet. In particular e-marketplaces that are centralized trading hubs for conducting on-line trade between several sellers and buyers provide improved efficiencies and benefits especially in business-to-business (B2B) e-commerce. E-marketplaces can operate according to different business models including catalog-based e-sales and e-procurement, auctions and reverse auctions, and exchanges addressing the trading needs and business requirements of different vertical and horizontal markets. The number of e-marketplaces is now estimated to be about 1500 worldwide (Gartner, 2003) with revenue projections up to $1.5 trillion in 2004 (Arthur Andersen, 2003, Forrester, 2003). Despite a recent economic slowdown, prospects for e-marketplaces remain strong. It is expected that over the next four years there will be 5000 – 10000 e-marketplaces

Transactions of the SDPS

SEPTEMBER 2003, Vol. 7, No. 3, pp. 109-123

with a possibility of merges and acquisitions resulting in a smaller number of large inter-connected emarketplaces with more advanced trading, mobile access and user support capabilities. Most existing e-marketplaces focus on facilitation of information exchange between the buyers and sellers who can access e-marketplaces to communicate and trade with other participants through their Web browsers. Although these e-marketplaces often provide an advanced data management support involving catalog management, search for merchants and products, ordering, payment and reporting, they typically require the participants to frequently check trading opportunities in the marketplace, engage in repetitive and often lengthy information exchange and decision-making tasks, and make most trading decisions manually. This can impose high participation costs due to mostly manual efforts and the requirement for more experienced staff to make frequent and often complex trading decisions. In addition it creates the risk of missing trade opportunities if the active participation time is constrained by the limited availability of staff or physical access to the e-marketplace. Moreover, involvement of inexperienced staff in making trading decisions may result in inefficient transactions due to sub-optimal trade arrangements. These problems become even more apparent with the business trends to trade in several inter-connected e-marketplaces and use new emerging communication channels such as personal digital assistants (PDAs) and mobile phones. Therefore, in order to ease the access and participation, reduce participation costs and improve trading efficiencies, support for automation of decision-making as well as enablement of new communication channels and mobile devices for trading will be mandatory in the next generation of e-marketplaces. There is an increasing recognition of software agents as a promising technology for developing more advanced e-commerce applications. This is reflected by a growing number of research efforts in the areas of mobile e-commerce information agents (Kotz and Gray, 1999, Papaioannou, 2000) and agent-mediated e-commerce (Bailey and Bakos, 1997, Beam and Segev, 1997, Chavez et al., 1997, Faratin et al., 1999, Guttman and Maes, 1998, Guttman et al., 1998, Kowalczyk and Bui, 2000, Lomuscio et al., 2000, Maes et al., 1999, Sandholm and Lesser, 1995), respectively. More recently, some research has also been directed towards the use of mobile agents in agent-mediated e-commerce (Tu et al., 1999, Kotz and Gray, 1999) including location-aware wireless comparison shopping (Mihailescu and Binder, 2001, MIT, 2003), networked comparison shopping (Dasgupta, et al., 1999), contract negotiation (Griffel et al., 1997) and auctions (Sandholm and Huai, 2000, Fukuta et al., 2001). This paper presents work-in-progress towards the development of an intelligent mobile agent-based e-marketplace system called InterMarket that aims at enabling mobile access and automated trading in inter-connected e-marketplaces. InterMarket integrates mobile agents and intelligent decision-making agents offered within an add-on component to an existing commercial e-marketplace platform. Sections 2 overviews a general concept and the proposed solution of InterMarket. InterMarket software solution aspects including the agent-based functionality, selected use-case scenarios and system architecture are presented in section 3. Section 4 discusses the InterMarket approach in the context of related research on mobile e-commerce agents and intelligent agent-mediated e-commerce. Finally the concluding remarks are presented in Section 5. 2. InterMarket Concept InterMarket addresses the business needs presented in the previous section by proposing a new solution based on integration of mobile agents and intelligent decision-making agents within an intelligent mobile agent-based e-marketplace that can enable mobile access and automated trading in emarketplaces. It constitutes a general concept of the next generation of e-marketplaces where trading activities can automatically be conducted by software agents sent to the e-marketplaces on behalf of buyers and/or sellers as presented in Figure 1. The InterMarket agents have the autonomous decision-making and mobility capabilities. They can proactively monitor trading opportunities, search for trading partners and products, and make trading Journal of Integrated Design and Process Science

SEPTEMBER 2002, Vol. 7, No. 3, pp. 110

decisions to satisfy users’ objectives, preferences and constraints following the users’ business and trading rules. The agents are able to move to the e-marketplaces through the Internet and can be initiated from different computer platforms and mobile devices such as mobile phones and PDAs. The agents can trade directly with a number of other participants in an e-marketplace including humans and other software agents. They can also move between several e-marketplaces to complete trading tasks (see Figure 2). After making trade arrangements, the agents can move back to the users’ locations carrying the notification of a prospective trade. Typically the final decision on the trade transaction can be made by the users who can accept the deal or engage in further negotiations.

Fig. 1 A general concept of InterMarket. There are a number of possible modi operandi for the intelligent mobile agent-based emarketplaces, depending on the business model scenarios and decision-making processes for automated trading, and the chosen mobility solutions for trading agents in e-marketplaces as follows: (1) Business model scenarios • Catalog-based trading: e-selling - the sellers make their product catalogs available in emarketplaces for the buyers accessing the e-marketplaces; e-procurement – the buyers create own procurement catalogs and allow the sellers to post their offerings/catalogs in the emarketplaces. In both cases the trading agents can perform the buyers’ and sellers’ tasks in the e-marketplace including catalog search, matching, transaction negotiation and ordering. • Auctions: the sellers create forward auctions for selling products to the bidding buyers in the emarketplace; the buyers create reverse auctions for buying products from the bidding sellers in the e-marketplace. The trading agents can manage the auctions and participate in the bidding processes on behalf of the bidders. • Dynamic exchanges: the buyers and sellers engage in buying and selling in the e-marketplace at the same time (e.g. they can post their selling and procurement catalogs, conduct double auctions, etc). As above the agents can automate catalog-based trading or engage in double auctions on behalf of the buyers and sellers.

Transactions of the SDPS

SEPTEMBER 2003, Vol. 7, No. 3, pp. 111

(2)

(3)

(4)

Decision-making processes • Human trader emulation: trading agents follow human decision-making rules according to business models in the existing e-marketplaces. • Enhanced trader: trading agents operate according to the optimized decision-making strategies designed for new trading models such as dynamic pricing and more advanced business negotiation models. Automated trading • Agent-to-Human: trading agents engage in trading activities with human participants in emarketplaces. • Agent-to-Agent: trading agents conduct a fully automated trade with other agents typically in dedicated trading rooms in e-marketplaces. Mobility • Stationary Trading/Mobile Communication Agents: trading agents that are stationary in emarketplace conduct automated trade according to the trading parameters and rules delivered by specialized mobile communication agents from user locations (these agents also deliver the results of trade to the users). • Mobile Trading Agents: trading agents are initiated at the users’ locations and move to/from emarketplaces in order to engage in automated trade in the e-marketplaces on behalf of the users.

Fig. 2 A general concept of mobile agents in inter-connected e-marketplaces. The initial version of InterMarket focuses on the human trader emulation in a catalog-based eprocurement business model scenario, primarily for agent-to-human trading with the use of stationary trading agents and mobile communication agents. It forms a basis for further and perhaps more generic solutions involving other business model scenarios, decision-making processes, automated trading modes and mobility solutions for the intelligent mobile agent-based e-marketplaces in the future. The InterMarket concept is based on add-on integration with two existing software products, i.e. Tracy Mobile Agent System (Braun, et al., 2001) developed at FSU and the commercial e-Marketplace system based on Enfinity (Intershop, 2002) from Intershop, as schematically presented in Figure 3. Tracy provides a general-purpose mobile agent capabilities, such as the agent platform, agent mobility

Journal of Integrated Design and Process Science

SEPTEMBER 2002, Vol. 7, No. 3, pp. 112

and communication mechanisms, which are tailored to the InterMarket specific requirements. It also provides the basis for stationary agents enhanced with intelligent decision-making capabilities for automated trading in InterMarket. The InterMarket solution is designed as an add-on business component to Intershop Enfinity that provides advanced catalog management, search and ordering mechanisms, transaction engines, and administration and reporting in the e-marketplace. More details on the InterMarket software solution including its agent-based functionality with the selected use-case scenarios and a system architecture involving mobile and intelligent agents in InterMarket are presented in the following section.

Intershop Enfinity

InterMarket

New solution

Intelligent Decision Making for Automated Trading

Tracy Agent Mobility, XML Communication & Mobile Devices

Intershop e-Marketplace

Existing solution

Interface

Fig. 3 InterMarket add-on integration with Tracy and Enfinity e-Marketplace.

3. InterMarket Solution The main aim of the InterMarket approach is the integration of mobile and intelligent agent technologies into a commercial e-marketplace. This section considers the agent-based functionality of InterMarket by presenting the main scenario applicable to the average InterMarket user. More details of the full InterMarket functionality can be found in (Müller et al., 2002). Based on our research and evaluations we strongly believe that an agent-based approach for developing truly distributed systems offers in general better abilities to design flexible, adaptable, and mobile software solutions than a standard client-server architecture. The main reason is that an agentdriven methodology offers a more problem-centric perception, thus reflecting directly the application domain and its possible solution spaces, instead of forcing a very tool-driven view onto the system developer. This capability is based on the (mobile) agent core concept, i.e. to provide communicative intelligent and really autonomous software entities which can migrate to and/or be executed without major planning on any of the network’s nodes. As a consequence, a set of software agents can directly model and implement the needs of a specific problem domain, offering well structured communication, cooperation, migration, and coordination of activities. To leverage these advantages and structure them accordingly, InterMarket distinguishes two major types of software agents: • mobile and Transactions of the SDPS

SEPTEMBER 2003, Vol. 7, No. 3, pp. 113

• stationary agents. In the InterMarket context, mobile agents are called mobile communication agents, because they are limited to realize user interaction and remote access to InterMarket platforms from client devices, based on their ability to migrate efficiently between platforms. Mobile agents are utilized for these tasks also because they offer in addition to their migration capabilities the following advantages, especially for mobile devices: • they have the capability to provide flexible and personalized user interaction, e.g. via wizards or avatars • they have low demands on network connection, quality and online time because after their migration to a device no online connection must be maintained while the mobile agent is working locally • mobile agents are pro-active, which enables them to autonomously find and process data resources in the Web • they reduce maintenance efforts because of their capability to utilize thin clients and additionally by enabling an automated deployment of new agent functionality • finally, based on their truly distributed character, they offer good abilities for interconnecting several e-marketplaces on a business logic level, thus providing natural opportunities for crossmarketplace trade Stationary trading agents do not have the ability to migrate to other platforms. They implement possibly complex, local business logic and data manipulation tasks, thus complementing with their intelligence the fast but pretty “dumb” mobile agents (refer to Section 2 for more details). In addition, the separation of stationary trading agents and mobile communication agents has been chosen to support non-functional issues, such as security (protection against access violation), performance (the small size of mobile communication agents has a positive impact on migration quality and time), and design constraints (a clean role-based design offers better maintenance and extensibility). As an example for agent-based InterMarket functionality we will shortly discuss the scenario presented in Figure 4, which describes remote access to InterMarket from mobile devices. Starting the scenario, a user configures and starts a mobile agent on its local device (e.g. PDA, mobile phone). The user then locally interacts with the agent and instructs it with trading task-specific data, authentication information, and optionally with a migration route (a plan for visiting several different e-marketplaces, maybe in a special order). After the mobile agent was started, it migrates to the first target emarketplace specified by the user. If the mobile agent has not received a migration route from its user, it moves first to a special point in the InterMarket mobile agent network for retrieving data on available e-marketplaces. After arriving at a target marketplace, the mobile communication agent authenticates within the marketplace using the given user information and contacts an idle stationary agent via a special registry, which returns a reference name for a stationary agent. The mobile agent then interacts with the stationary agent via messages using the reference name to configure it with task data. The stationary agent confirms configuration, and starts the execution of single tasks or even a whole task workflow, depending on the given problem, while the mobile agent is waiting. Therefore, the stationary agent first has to analyze the given task data for creating a solution plan. This plan can comprise catalog-based activities (search, match, order), participation in auctions or dynamic exchanges (negotiations, RFP, RFQ), or even a combination of these activities. For solving its tasks the stationary agent can interact with so called system agents, which implement e.g. access to the InterMarket catalogs, auctions, or negotiation messaging system. InterMarket uses an XML-based exchange format for the data exchange. For every task a special decision-making algorithm will be executed, which has to be supported with problem-specific data. The stationary agent receives for all activities in addition to product data (describing product qualities) also quality criteria Journal of Integrated Design and Process Science

SEPTEMBER 2002, Vol. 7, No. 3, pp. 114

(describing relevance of product qualities). Using both pieces of information, the agent is able to perform a qualified matching of the product data received from the user compared to product data received from the InterMarket catalog (catalog-based activities), received from auction information, or received from negotiation responses.

Fig. 4 Mobile agent-based execution. The stationary agent is then able to search for suitable auctions or possible trading partners via the InterMarket catalog using the same approach as mentioned before for product search and comparison. With this information, the stationary agent is able to register at an auction or to contact another marketplace participant via the negotiation system. In an auction the stationary agent gets automatically informed of every change. Thus, it is able to perform its decision-making algorithm to determine how to react to those changes. If the auction is a long time activity, a message will be handed via all involved agents back to the user. Finally, the stationary agent places negotiation requests via a system agent in the proper registry of InterMarket. The trading partners, who have received a request message, can respond to it by filling out a special negotiation form. The decision-making algorithm of the stationary agent then decides on a response, i.e. how and in what way to react to an offer (accept it, make a counter-offer or withdraw from the negotiation). After having performed all tasks, the stationary task agent creates a response and returns the results of all performed transactions. The mobile agent stores these (intermediate) results. If it was configured to visit several e-marketplaces, the mobile agent can now migrate to additional e-marketplaces following its route. After visiting all designated e-marketplaces, the mobile agent migrates back home for presenting its results to the user. Finally, the user analyzes the result data, can make final decisions (e.g. accept or reject the agent’s recommendations, continue negotiations) and may, of course, initiate other tasks possibly involving agents.

Transactions of the SDPS

SEPTEMBER 2003, Vol. 7, No. 3, pp. 115

InterMarket offers two other main scenarios involving mobile and stationary agents. However, these two scenarios work similar to the one described here and are not detailed any further due to space limitations. The integration of mobile agent technology into the overall InterMarket architecture is presented in Figure 5. InterMarket is not a monolithic block. It directly supports the integration of both underlying technologies, i.e. Tracy and Enfinity, distinguishing agent-based and standard marketplace functionality and supporting two different ways of accessing a marketplace. The Tracy sub-system (lower part in Figure 5) implements the agent-based functionality and the mobile access of InterMarket. The user can connect to the marketplace by migrating mobile agents from its client Agent Server to the InterMarket Gateway Agent Server. The Gateway Agent Server is responsible for authentication checks, load balancing, and distributes incoming mobile agents (migration) as well as requests from the Enfinity sub-system (Tracy Remote API) to the Agent Server sub-net, where mobile and stationary agents perform their tasks. Stationary agents can use the Enfinity functionality for their task execution by simulating HTTP requests to the Enfinity sub-system’s Web Server (e.g. catalog search). The Agent Servers in the sub-net share a common database for ensuring persistence and failover capabilities of agent-based execution.

Fig. 5 InterMarket architecture. The upper part of Figure 5 presents the Enfinity sub-system including its Procurement solution functionality (Intershop, 2002) that offers standard marketplace functions and standard client access to InterMarket. The customer can use a client Browser to access via a HTTP connecting to the Web Server of the Enfinity sub-system. The Web Server is responsible for load balancing and routes HTTP requests to the Application Server cluster. The InterMarket business logic resides on the Application Journal of Integrated Design and Process Science

SEPTEMBER 2002, Vol. 7, No. 3, pp. 116

Servers, executing standard marketplace functionality and exchanging information with the Tracy subsystem using the Tracy Remote API (e.g. configuring and starting agents). The Application Servers also share a database for ensuring persistent execution of the business logic, which is accessed using a special product-specific communication. In summary, InterMarket aims at providing a framework for building e-marketplaces with widely differing technical requirements, such as scalability, performance, and availability. It also deals with the execution effort and required resources for agent-based tasks, as well as with the balancing of the agent server workload in relation to the whole InterMarket workload. It implements a special version of a 3-Tier architecture to be able to accommodate different e-marketplace scenarios, thus offering flexibility to integrate and support varying user needs. As experience shows, this is in particular achieved through separation of the Gateway Agent Server from “other working” Agent Servers in the Tracy sub-system (see Figure 5). Further details of the InterMarket software solution can be found in (Müller et al., 2002). 4. InterMarket Approach and Related Research Mobile agents and intelligent agents have independently been recognized as very promising technologies for e-commerce applications. This is reflected by a growing number of research efforts in the areas of mobile e-commerce information agents (Kotz and Gray, 1999, Papaioannou, 2000) and agent-mediated e-commerce (Bailey and Bakos, 1997, Beam and Segev, 1997, Chavez et al., 1997, Faratin et al., 1999, Guttman and Maes, 1998, Guttman et al., 1998, Kowalczyk and Bui, 2000, Lomuscio et al., 2000, Maes et al., 1999, Sandholm and Lesser, 1995), respectively. More recently, in recognition of the potential benefits of integrating mobile agents and intelligent agents in more advanced dynamic e-commerce applications, some research has also been directed towards the use of mobile agents in agent-mediated e-commerce (Dasgupta, et al., 1999, Griffel et al., 1997, Sandholm and Huai, 2000, Tu et al., 1999, Fukuta et al., 2001, Mihailescu and Binder, 2001). InterMarket aims at exploring further the synergistic nature of those technologies with the focus on integration of their complementary capabilities and development of a next generation e-commerce application, i.e. the intelligent mobile agent-based e-marketplace. In particular the InterMarket approach integrates intelligent decision-making agents for automated trading and mobile agents for mobile access and communication that differs from the existing related research efforts in a number of ways summarized in Table 1 and described below. InterMarket proposes intelligent trading agents to automate the users’ decision-making tasks in emarketplaces and mobile agents to support deployment of the trading agents and provide mobile access to e-marketplaces from mobile devices such as Personal Digital Assistants (PDA) or mobile phones. It clearly differs from most research related to the agent-mediated e-commerce and mobile e-commerce information agents that tend to separately focus on the trading decision automation support or mobile communication in e-commerce, respectively. InterMarket’s approach also differs from the current research efforts on the use of mobile agents in agent-mediated e-commerce including location-aware wireless comparison-shopping (e.g. (Mihailescu and Binder, 2001, MIT, 2003), networked comparison-shopping (Dasgupta, et al., 1999), contract negotiation (Griffel et al., 1997) and auctions (Sandholm and Huai, 2000, Fukuta et al., 2001). A typical scenario of location aware shopping involves comparison shopping (Mihailescu and Binder, 2001, MIT, 2003) where mobile agents from the buyer’ mobile device (e.g. PDA) can move and communicate with sellers’ agents at the point of purchase in order to locate and compare products, and possibly negotiate the terms of purchase while the buyer is in a mall, on a way to a store or within the store. Similar principles have also been proposed by (Dasgupta, et al., 1999) for a mobile agent system that allow the shopping agents to compare quotes from different seller sites they can visit through the wired Internet. Although similar approach can be envisaged for InterMarket, it should be

Transactions of the SDPS

SEPTEMBER 2003, Vol. 7, No. 3, pp. 117

noted that InterMarket targets e-marketplaces (which are predominantly B2B e-commerce as opposed to B2C location aware comparison-shopping) involving more complex and frequent transactions thus more advanced decision-making capabilities. Table 1. InterMarket in relation to other systems

Automation of trading Market Mechanism

Decision Apparatus

AuctionBot

eAuction House

MAgNET

M&B

Negotiation stage

Negotiation stage

Product Merchant Brokering

Auction

Auction

Comparison shopping

Game Theoretic

Game Theoretic

&

Matching & comparison algorithms

(+API for other agents)

(English and combinatorial auctions)

Agents

Stationary

Mobile

Mobile

Mobility

NA

within marketplace

Web

Nomad (Concordia)

IBM Aglets

Deployment ?

Product Merchant Brokering

InterMarket & All stages

Location aware comparison shopping

Catalogbased eprocurement

Matching & comparison algorithms

Knowledge based reasoning

Mobile & stationary Mobile devices & Web IBM Aglets KVM SDK

Mobile & stationary Mobile devices & Web

(initially)

Tracy

Mobile e-marketplace agents have been proposed in eAuctionHouse (Sandholm and Huai, 2000) that deploys mobile agents created within an auction-based e-marketplace by the users through the Web browser for participation in the bidding processes on the users’ behalf. Mobile agents called BiddingBots that can visit and bid in different e-marketplaces have been investigated in (Fukuta et al., 2001). While InterMarket deploys intelligent trading agents within the e-marketplace with a mobile agent system, it also enables mobile access with the use of mobile communication agents that can reside on the users’ personal computers and mobile devices, and move to and between e-marketplaces to deliver the users’ instructions to the trading agents, and return to user locations to notify the users on a prospective trade. In (Griffel et al., 1997) mobile agents are proposed to circulate contract documents between different parties engaged in the contract negotiation process over the Internet who can manually review and alter the contract through their Web browsers. In InterMarket each contract document is stored on the dedicated whiteboard within the e-marketplace and it is delivered by the mobile agents to the corresponding users only once when the trading agents complete their tasks involving preparation of a prospective contract on the users’ behalf using the secure communication channels within the emarketplace. Mobile agents with more advanced plug-in decision-making capabilities for automated e-commerce negotiation are considered in research proposed in (Tu et al., 1999, Tu et al., 2001). Because of the limited computational resources of mobile devices and network bandwidth, InterMarket adopts “thin” mobile communication agents to deliver instructions to “thicker” trading agents having the intelligent decision-making capabilities for automated trading within the e-marketplace. This is similar to an

Journal of Integrated Design and Process Science

SEPTEMBER 2002, Vol. 7, No. 3, pp. 118

approach adopted in location-aware comparison-shopping agents by (Mihailescu and Binder, 2001) where mobile wireless communication agents are “light-weighted”, and wired and stationary comparison agents are ”heavy-weighted”. In general the InterMarket approach enables access to the emarketplace from mobile devices and automated trading when the users move or are even disconnected from the network. In particular the mobile communication agents enable mobile devices and reduce network traffic and communication latency, and the intelligent trading agents automate trading tasks within InterMarket to ensure a quicker response to changes in the e-marketplace and making trading decisions faster than remote human participants could. A general-purpose mobile agent system Tracy developed by the authors at FSU (Braun, et al., 2001) has been chosen for implementation of InterMarket. Most existing research prototypes in mobile agentmediated e-commerce have also used general-purpose mobile agent systems (e.g. Concordia in eAuctioHouse/Nomad (Sandholm and Huai, 2000), IBM Aglets in MAgNET (Dasgupta, et al., 1999) and Aglets/KVM SDK in (Mihailescu and Binder, 2001). However the used systems are typically third-party “closed” systems that provide specific mobility and communication capabilities and permit implementation of some high-level application specific functions. To fulfill all the requirements of InterMarket, Tracy allows enhancements at both the application and system levels (Kowalczyk et al., 2002). An additional function of Tracy is to provide an interface to a commercial Intershop eMarketplace to enable InterMarket to be integrated and demonstrated as an add-on business component in a real-world e-commerce trading scenario. InterMarket’s intelligent agents aim at making trading decisions on behalf of the users that emulate human traders’ behavior in the e-marketplace. Most related research in agent-mediated e-commerce focuses on supporting and automating different stages of the e-commerce transaction process including product brokering, merchant brokering and negotiations (Bailey and Bakos, 1997, Guttman and Maes, 1998, Guttman et al., 1998, Jennings et al., 2001, Maes et al., 1999, Sandholm and Lesser, 1995). Research on the product and merchant brokering agents typically involves development of searching, filtering and comparison algorithms for finding products and suppliers (MIT, 2003). Research on agent-mediated negotiation is often directed towards developing optimization algorithms (Rosenschein and Zlotkin, 1994, University of Michigan, 2003, Washington University, 2003) or trading heuristics for different market mechanisms (Chavez et al., 1997, Collins and Gini, 2000, Faratin et al., 1999, Guttman and Maes, 1998). Rather than using specialized algorithms for different stages of the transaction process, InterMarket’s trading agents simulate human traders’ behavior provided by an automated reasoning mechanism using human traders’ knowledge/rules and instructions that allow the trading agents to get involved in different trading tasks in the e-marketplace in a similar fashion as the human participants could. Intelligent trading agents in InterMarket use a decision-making model based on automated fuzzy reasoning with knowledge and meta-knowledge representing the users’ preferences, constraints and objectives, and trading protocols, business rules and trading strategies, respectively. Most related research in this area usually focuses on game-theoretical and decision-theoretical methods (Rosenschein and Zlotkin, 1994), or developing specialized heuristics for making trading decisions in specific trading situations (Lomuscio et al., 2000). The game-theoretical approach is a very important and active area of research in agent-based trading that aims at designing negotiation mechanisms with the objective to optimize trading outcomes in different market types such as auctions. For example research presented in (Sandholm and Huai, 2000) and (Wurman, 1998) explore the design of rational computational agents involving negotiation protocols and strategies for specific auctions. Although the game-theoretical approach provides a sound theoretical foundation for studying and designing optimal agent interactions it has been acknowledged that it is often based on a number of assumptions limiting applicability of its solutions in many real-world problems. It may also result in negotiation strategies that are computationally expensive (or even intractable) or difficult to understand by the human traders, and hence very difficult to use in practical applications (Lomuscio et al., 2000). There are also a Transactions of the SDPS

SEPTEMBER 2003, Vol. 7, No. 3, pp. 119

number of examples of research based on decision theory (Keeney and Raiffa, 1976) and multiattribute utility theory for supporting agent-based negotiations (Sycara, 1992). For example a multiattribute utility theoretic architecture for electronic commerce has been proposed in (Barbuceanu and Lo, 2000), and the principles of multi-attribute utility theory have been used in designing negotiation strategies for trading agents in (Faratin et al., 1999) and (Kowalczyk and Bui, 2000). In recognition of the practical limitations of many current results based on the theoretical approaches including their assumptions and the cost associated with computation and decision making, the heuristic methods aim to produce “good” rather than “optimal” solutions. They may be either computational approximations of the game and decision-theoretical techniques or computational realizations of more informal negotiation models (Jennings et al., 2001). For example Kasbah’s agents (Chavez et al., 1997) demonstrate the use of simple price functions to model trading strategies of the buying and selling agents for changing their bidding and asking prices, respectively, over time. Faratin et al. (Faratin et al., 1998, Faratin et al., 1999) propose an approach for modeling agents’ negotiation strategies with the predefined decision functions that comprise a set of weighted tactic functions addressing specific aspects of agents’ negotiation behavior. Constraint-based approaches have been proposed in (Guttman and Maes, 1998) and (Kowalczyk and Bui, 2000) to model and control the negotiation processes in trading agents. To avoid simplistic and sometimes “ad hoc” solutions the heuristic approaches often involve the requirement for extensive experimentation with different trading scenarios and the use of machine learning techniques. Applicability of automated learning for evolving negotiation strategies for intelligent agents with the use of genetic algorithms has been explored in (Tu et al., 1999). Oliveira and Rocha (Oliveira and Rocha, 2001) present an e-commerce architecture in which a designated agent coordinates the market using a decision-making apparatus built on a reinforcement learning algorithm. Both the heuristic and machine learning methods may result in the specialized functions and heuristics tailored to specific trading scenarios under consideration and may be difficult to use in the extended context. InterMarket’s approach can be described as knowledge-based where the intelligent trading agents use human traders’ knowledge rather than either simple heuristics or complex “black box” optimization algorithms. It allows trading agents to handle more trading situations depending on the knowledge and instructions provided to the agents and to “simulate” human traders, thus increasing flexibility and acceptability of agents’ decisions. It should be noted however that because the quality of trading outcomes depends on the agents’ knowledge and instructions provided by the users it may not ensure the theoretically optimal solutions but it can provide “human-like-made” practical solutions. It is expected that these solutions should be better than solutions obtained with simple ad-hoc heuristics. Moreover such a decision-making approach can handle a wider range of trading models than ones supported by the specialized game-theoretical strategies valid for the selected specific market mechanisms such as some auction types. It also ensures that agents’ trading decisions and results can be understandable and “owned” by the human trader so increasing the acceptability of the automated trading in e-marketplaces. 5. Concluding Remarks The work on InterMarket is driven by the industry needs for new solutions providing advanced trading, mobile access and user support capabilities in inter-connected e-marketplaces. InterMarket, an intelligent mobile agent-based e-marketplace system, aims at enabling mobile access and automated trading in e-marketplaces based on integration of mobile agents and intelligent decision-making agents offered as an add-on component to a commercial e-marketplace platform. It is envisaged that InterMarket’s novel approach can provide several benefits to industry including the reduced participation costs, easy access from different mobile devices and increased efficiency of trading in Journal of Integrated Design and Process Science

SEPTEMBER 2002, Vol. 7, No. 3, pp. 120

inter-connected e-marketplaces. The work on InterMarket is in progress and attempts to demonstrate these benefits with the commercial e-marketplace, focussing initially on the human trader emulation in a catalog-based e-procurement business model scenario, primarily for agent-to-human trading with the use of stationary trading agents and mobile communication agents. It is intended as a basis for further and perhaps more generic InterMarket solutions involving other business model scenarios, decisionmaking processes, automated trading modes and mobility solutions for the intelligent mobile agentbased e-marketplaces in the future. 6. References Authur Andersen, 2003, http://www.andersen.com/. Bailey, J., and Bakos, Y., 1997, “An Exploratory Study of the Emerging Role of Electronic Intermediaries”. International Journal of Electronic Commerce, vol. 1, no. 3. Barbuceanu, M. and Lo, W., 2000, “A multi-attribute utility theoretic architecture for electronic commerce”. Proceedings of 4th Int. Conf. on Autonomous Agents, Barcelona, Spain, pp. 239-247. Beam, C. and Segev, A., 1997, “Automated Negotiations: A Survey of the State of the Art”. Wirtschaftsinformatik, 39(3): 263-268. Braun, P., Eismann, J., Erfurth, C., and Rossak W., 2001, “Tracy - A Prototype of an Architected Middleware to Support Mobile Agents”. Proceedings of the 8th Annual IEEE Conference and Workshop on the Engineering of Computer Based Systems (ECBS), Washington D.C. (USA), April 2001, pp. 255-260. Chavez, A., Dreilinger, D., Guttman, R. and Maes, P., 1997, “A real-life experiment in creating an agent marketplace”. Proc. of the Second Int. Conference on the Practical Application of Intelligent Agents and MultiAgent Technology PAAM’97. Practical Application Company, London. Collins, J. and Gini, M., 2001, “Exploring decision processes in multi-agent automated contracting”, Technical Report, TR 00-53, University of Minnesota, 2000. An edited version appeared in IEEE Internet Computing, March/April 2001, pp 61-72. Dasgupta, P., Narasimhan, N., Moser, L., Smith, P.M.M., 1999, “MAgNET: Mobile Agents for Networked Electronic Trading”, IEEE Transactions on Knowledge and Data Engineering, Special Issue on Web Technologies, vol. 24, no. 6, July/August 1999, pp 509-525. Faratin, P., Sierra, C., and Jennings, N., 1998, “Negotiation decision functions for autonomous agents”. International Journal of Robotics and Autonomous Systems 24 (3-4), pp. 159-182. Faratin, P., Sierra, C., Jennings, N., and Buckle, P., 1999, “Designing Flexible Automated Negotiators: Concessions, Trade-Offs and Issue Changes”, Institut d'Investigacio en Intelligencia Artificial Technical Report, RR-99-031, 1999. Forester, 2003, http://www.forrester.com/. Fukuta, N., Ito, T., Ozono, T., and Shintani, T., 2001, “A Framework for Cooperative Mobile Agents and Its Case-Study on BiddingBot”. Proceedings of the JSAI 2001 International Workshop on Agent-based Approaches in Economic and Social Complex Systems, pp. 91-98. Gartner, 2003, http://www.gartner.com/. Griffel, F., Tuan, M., Munke, M., da Silva, M., 1997, “Electronic contract negotiation as an application niche for mobile agents”. Proceedings of Int. IEEE Workshop on Enterprise Distributed Object Computing (EDOC), Australia. Guttman, R. and Maes, P., 1998, “Agent-mediated integrative negotiation for retail electronic commerce”. Proc. of the Workshop on Agent-Mediated Electronic Trading AMET’98, Minneapolis.

Transactions of the SDPS

SEPTEMBER 2003, Vol. 7, No. 3, pp. 121

Guttman, R.H., Moukas, A.G., and Maes, P., 1998, “Agent-mediated Electronic Commerce: A Survey”. Knowledge Engineering Review, Vol. 13, No. 2, p 147-159. Intershop, 2002, , Intershop Business Components: Procurement Solution for Intershop Enfinity 2, (User Guide - internal paper), http://www.intershop.com. Jennings, N., Faratin, P., Lomuscio, A., Parson, S., Sierra, C., and Wooldridge, M., 2001, “Automated Negotiation: Prospects, Methods and Challenges”. International Journal of Group Decision and Negotiation, Vol. 10, No. 2, pp. 199-215. Keeney, R. and Raiffa, H., 1976, Decisions with Multiple Objectives: Preferences and Value Trade-offs. John Willey and Sons. Kotz, D. and Gray, R., 1999, “Mobile Agents and the Future of the Internet”. ACM Operating Systems Review, pp. 7-13. Kowalczyk, R. and Bui V., 2000, “On Constraint-based Reasoning in eNegotiation Agents”. In F. Dignum and U. Cortés (Eds.). Agent Mediated Electronic Commerce III, LNAI, Springer-Verlag, pp. 31 - 46. Kowalczyk, R., Franczyk, B., Speck, A., Braun, P. Eismann, J., and Rossak, W., 2002, “InterMarket Towards Intelligent Mobile Agent e-Marketplaces”, IEEE International Conference ECBS’2002 (in press). Lomuscio, A. Wooldridge, M. and Jennings, N., 2000, “A classification scheme for negotiation in electronic commerce”. In F. Dignum and C. Sierra (Eds.). Agent-Mediated Electronic Commerce: A European Perspective, Springer Verlag, pp. 19-33. Maes, P., Guttman, R., and Moukas, A., 1999, Agents That Buy and Sell. Communications of the ACM, Vol. 42, No. 3, pp. 81-91. Mihailescu, P. and Binder, W., 2001, “A Mobile Agent Framework for M-Commerce,” GI Jahrestagung (2), pp. 959-967. MIT, Agents, 2003, http://agents.www.media.mit.edu/groups/agents/projects/impulse. MIT, 2003, Tete-a-Tete, http://ecommerce.media.mit.edu/Tete-a-Tete/ Müller, I., Braun, P., Rossak, W.R., 2002, Integrating Mobile Agent Technology into an e-Marketplace Solution, the InterMarket Solution. Technical Report, Computer Science Department, Friedrich Schiller University Jena (publ. as Jenaer Schriften zur Mathematik und Informatik). Oliveira, E. and Rocha, A.-P., 2001, “Agents advanced features for negotiation in electronic commerce and virtual organisation formation process”. In F. Dignum and C. Sierra (Eds.) Agent Mediated Electronic Commerce – A European AgentLink Perspective, vol. 1991 of LNAI, Springer-Verlag, pp. 78-97. Papaioannou, T., 2000, “Mobile Information Agents for Cyberspace – State of the Art and Visions”. Proceedings of Cooperating Information Agents (CIA-2000), vol. 1860 LNCS, Springer-Verlag. Rosenschein, J. and Zlotkin, G., 1994, Rules of Encounter: Designing Conventions for Automated Negotiation among Computers. MIT Press. Sandholm, T. and Huai, Q., 2000, “Nomad: Mobile Agent System for an Internet-Based Auction House”. IEEE Internet Computing, pp. 80-86. Sandholm, T. and Lesser, V., 1995, “Issues of Automated Negotiation and Electronic Commerce: Extending the Contract Net Framework”. Proceedings of 1st Int. Conf. on Multiagent Systems (ICMAS’95), San Francisco. Sycara, K., 1992, “The PERSUADER”. In D. Shapiro (ed.). The Encyclopedia of Artificial Intelligence. John Wiley Sons. Tu, M., Griffel, F., and Lamersdorf, W., 1999, “Integration of Intelligent and Mobile Agents for Ecommerce – A Research Agenda”. In: St. Kirn, M. Petsch (Eds.). Workshop Intelligente Softwareagenten und betriebswirtschaftliche Anwendungsszenarien, TU Ilmenau, FG Wirtschaftsinformatik 2, Arbeitsbericht.

Journal of Integrated Design and Process Science

SEPTEMBER 2002, Vol. 7, No. 3, pp. 122

Tu, M., Seebode, C. Griffel, F., and Lamersdorf, W., 2001, “DynamiCS: An Actor-based Framework for Negotiating Mobile Agents”. Electronic Commerce Research Journal, Vol. 1, No. 1/2. University of Michigan, 2003, http://auction.eecs.umich.edu/. University of Washington in St. Louis, 2003, http://ecommerce.cs.wustl.edu/. Wurman, P. R, Walsh, W. E., and Wellman, M. P., 1998, “Flexible double auctions for electronic commerce: Theory and implementation”. Decision Support Systems, Vol. 24, pp. 17-27.

Transactions of the SDPS

SEPTEMBER 2003, Vol. 7, No. 3, pp. 123