Practice
Artificial Intelligence and Machine Learning

Agent-Oriented Technology in Support of E-Business

Enabling the development of "intelligent" business agents for adaptive, reusable software.
Posted
  1. Introduction
  2. Agents for Business-to-Business E-Commerce
  3. Basic Characteristics of E-Business Agents
  4. Types of Agents
  5. Conclusion
  6. References
  7. Author
  8. Figures

The rapid growth of the Internet, networking systems such as electronic data interchange systems, and the penetration of ISDN-based applications are stimulating an ever-increasing number of businesses to participate in e-commerce worldwide. For example, businesses use the Web to improve internal communication, help manage supply chains, conduct technical and market research, and locate potential partners. Moreover, innovative enterprises with good partner relationships are beginning to capitalize on the enormous potential of new global networking possibilities and are beginning to share sales data, customer buying patterns, and future plans with their suppliers and customers.

One of the key characteristics of the e-business world is that companies will inevitably move more and more into a customer-centric paradigm in order to increase competitiveness. Customer behavior cannot be accurately predicted using traditional analytic methods like forecasting or budgeting. Instead, companies seeking a competitive edge will investigate other kinds of analytical methods based on, for example, heuristics and AI techniques. Intelligent business agents are the next higher level of abstraction in model-based solutions to business-to-business e-commerce applications. By building on the distributed object foundation, agent technology can help bridge the remaining gap between flexible design and usable applications. Agents support a natural merging of object orientation and knowledge-based technologies. They can facilitate the incorporation of reasoning capabilities within the business application logic (for example, encapsulation of business rules within agents or modeled organizations). They permit the inclusion of learning and self-improvement capabilities at both infrastructure (adaptive routing) and application (adaptive user interfaces) levels. Unlike objects, business agents can participate in high-level (task-oriented) dialogues through the use of interaction protocols in conjunction with built-in organizational knowledge. In many cases, the need for communication is greatly reduced, as within these high-level dialogues, complex packets of procedural and declarative knowledge as well as state information may be exchanged in the form of mobile objects. In addition, agent technology can help address serious technological challenges such as concerns about effective searching, security and privacy, and effective use of interoperability between diverse business processes and diverse information required to achieve tele-cooperation and global e-commerce.

The opportunities for using intelligent agents in an e-business application are enormous. For example, they can be used for real-time pricing and auctioning, involving different parties in a supply-chain network. Suppliers can present their products on the Web and collect real-time price bids from multiple customers. Intelligent software agents can carry out work on behalf of human knowledge workers both on the supplier’s and customer’s behalf. A vision of how agent-enabled, business-to-business e-commerce could provide an unprecedented level of functionality to people and enterprises is described here.

Back to Top

Agents for Business-to-Business E-Commerce

The agent metaphor, due to its suitability for open environments, has recently become popular with distributed, large-scale, and dynamic nature applications such as e-commerce and virtual enterprises. The key aspects of agents are their autonomy, their abilities to perceive, reason, and act in their surrounding environments, as well as the capability to cooperate with other agents to solve complex problems [5]. An essential quality of a business agent is its dimension of “intelligence.” Intelligence is the amount of learned behavior and possible reasoning capacity that an agent may possess. At the most basic level, the agent may follow a set of rules that are predefined by the user. The agent can then apply these rules, for instance, to the Internet at large. The most intelligent agents will be able to learn, and will be able to adapt to their environment, in terms of user requests and the resources available to the agent.

Agent-enabled e-business computing is concerned with combining appropriately intelligent agents (both humans and computers) working cooperatively over space and time to solve a variety of complex problems in medicine, engineering, finance, banking, commercial enterprises, and so on. Figure 1 illustrates how specialized agents collaborate to implement a complex business application that may span diverse enterprises. Agents provide the semantic support infrastructure for high-level elements such as business objects, business processes, and workflows. To accomplish a complex task, an intelligent business agent uses an incremental processing style, which might include recruiting other agents in the process of task execution in a dynamic and opportunistic way [11]. This agent capability assumes that an information agent can invoke the functionality of other such agents (see Figure 1). In each vertical domain tackled by a network-centric business ecosystem, there is:

  • Distribution of data: purchase information, product order information, and invoices.
  • Distribution of control: each individual business process is responsible for performing a set of tasks, for example, order processing or goods shipping.
  • Distribution of expertise: a business process’s knowledge (for example, order handling), is quite different from that of another (for example, request for tender).
  • Spatial distribution of resources: a legacy information system may be responsible for providing information and functionality about purchase requisitions while a remote ERP package may provide financial or accounting activity services.

Back to Top

Basic Characteristics of E-Business Agents

An agent must have a model of its own domain of expertise—the agent expertise model—and a model of the other agents that can provide relevant information—the agent awareness model [11]. The awareness model of an information agent does not need to contain a complete description of the other agents’ capabilities, but rather only those portions that may be directly relevant when handling a request that cannot be serviced locally. In general, we require that intelligent business agents possess distinguishing characteristics described in the following paragraphs.

Delegation abilities. The central idea underlying agents is that of delegation. The owner or user of an agent delegates a task to the agent and the agent autonomously performs the task of behalf of the user. Alternatively, a business agent may decompose the task and delegate parts of it to other agents, which perform the subtasks and report back to the business agent. The agent must be able to communicate with the user or other agents to receive its instructions and to provide results of its activities.

Agent communication languages and protocols. A business agent is an autonomous entity, hence it must negotiate with other agents to gain access to other sources and capabilities. To enable the expressive communication and negotiation required and organize communications between agents, a language that contains brokering performatives can be particularly useful. For example, the Knowledge Query and Manipulation Language (KQML) can be used to allow information agents to assert interests in information services, advertise their own services, and explicitly delegate tasks or requests for assistance from other agents. KQML’s brokering performatives provide the basic message types that can be combined to implement a variety of agent communication protocols. This type of language can be used for both communication and negotiation purposes and provides the basis for developing a variety of interagent communications protocols that enable information agents to collectively cooperate in sharing information.

Some general examples of agent development environments include the AgentBuilder (www.agentbuilder.com) and the Intelligent Agent Library (www.bitpix.com/business/main/bitpix.htm). AgentBuilder is an integrated tool suite for constructing intelligent agents. Agents constructed using AgentBuilder communicate using KQML and support the performatives defined for KQML. AgentBuilder allows developers to define new interagent communications commands and customize them to their application’s needs. The Intelligent Agent Library provides an intelligent agent framework that includes extensive facilities for agent communications and for building larger agent assemblies. It offers a KQML-based agent framework and many examples illustrating agents that perform activities for Web-enabled applications.

Self-representation abilities. One of the most challenging problems is for agents to express naturally and directly business and system aspects and then combine these into a final meaningful application or implementation. This results in self-describing, dynamic, and reconfigurable agents that facilitate the composition (specification and implementation) of large-scale distributed applications, by drawing upon (and possibly specializing) business processes and the functionality of existing information sources. Such ideas can benefit tremendously from techniques found in reflection and metaobject protocols.

Back to Top

Types of Agents

In a multiagent e-business environment it is necessary to organize agents into different categories depending on their functionality and competencies. The four basic forms of agents can be distinguished as described here.

Application agents. A business-to-business e-commerce application is a networked system that comprises a large number of application agents. Each agent is specialized to single area of expertise and provides access to the available information and knowledge sources in that domain and works cooperatively with other agents to solve a complex problem in that vertical domain. This results in the formation of clusters (or groups) of information sources around domains of expertise handled by their respective agents.

An interesting consequence of e-business is that companies are turning to alliance-building strategies to strengthen and extend their value/supply chains. The business landscape is rapidly changing into a dynamic field where companies with complementary capabilities and common interests can coalesce to win competitive advantage. This leads to a phenomenon known as integrated value chains. Agents can help transform closed trading partner networks into open markets and extend such applications as production, distribution, and inventory management functions across entire value chains spanning diverse organizations. Figure 1 shows an example of such an integrated value chain employing various types of application agents such as inventory, production, and distribution agents that need to collaborate to implement a business ecosystem. Each application agent has a unique area of specialization. Figure 1 shows how such specialized application agents collaborate as part of a distributed workflow application that implements a typical range of cross-organizational activities. This figure also shows how application agents, such as the procurement agent, rely on general commerce activity agents to accomplish their functions.

Personal agents. Personal (or interface) agents work directly with users to help support the presentation, organization, and management of user profile, requests, and information collections. A personal agent gives its user easy and effective access to profile (preference) related specialized services and information widely distributed on the Web. The user’s agent observes and monitors the actions taken by the user in the interface and suggests better ways to perform the task. These agents can assist users in forming queries, finding the location of data, and explaining the semantics of the data, among other tasks. Examples include intelligent tutoring systems and Web browsing assistants. As for learning, a personal agent learns typically to better assist its user in different ways [9]: by observing and imitating the user; through receiving positive and negative feedback from the user; and by receiving explicit instructions from the user.

Typical examples of personal agents designed to learn user interests from browsing for recommendation purposes are Letizia [8] and WebMate [1]. Letizia is an autonomous interface agent that assists the user in browsing the Web by performing look ahead searches and making real-time recommendations for related pages. WebMate assists browsing by learning user preferences in multiple domains. These agents need supervision from the user in order to work; no truly autonomous search is possible.

General business activity agents. The activities and functions of e-business need certain basic agent technology support that is likely to become the basis for developing standard digital agents for e-business. General business agents perform a large number of general commerce support activities that can be customized to address the needs of a particular business organization. Such general functionality agents may include search agents that navigate effectively through fragmented online electronic information and services in order to find trading partners and items of interest; negotiation agents that negotiate on behalf of a buyer or a seller; billing agents; marketing agents to market product and services on the Internet; or legal agents that advise on issues surrounding privacy, taxes, export controls, and so forth.

Information brokering agents. Information brokers for e-business provide facilities such as locating information on Web sources or other agents that are required to solve a common problem, for example, sales and distribution processing, by name (white pages), by capabilities (yellow pages), content-based routing, or problem decomposition. Brokering agents have the ability to maintain, update, and access distributed directory services (listing products and business services). They also employ advanced navigation services such as maintaining hyperlinks, advanced keyword and context search agents. The term broker agent may be perceived as synonymous to what is also known in the literature as a matchmaking agent [7]. Brokering agents, in general, maintain information about other agents in a business ecosystem. Brokering agents use distributed information resource discovery and retrieval facilities to assist service providers in listing and publishing their services, and help seekers to find services and information of interest by employing search and information agents.

Search agents, for example, searchbots, access the network looking for particular kinds of information, filter it, and return it to their users. Search agents are designed to mitigate the information overload caused by the availability of large amounts of poorly cataloged business data. Search agents could be mobile, that is, they may be able to traverse the Web, gather information, and report what they have retrieved to a home location. Information agents perform the role of managing, summarizing, manipulating or collating information from many diverse distributed sources. The information is then collated and sent back to the user.

E-business applications are based on the existence of standard ontologies for a vertical domain that establish a common terminology for sharing and reuse. A domain ontology consists of terminology descriptions and other domain specific information and establishes a common vocabulary for interacting with a brokering/information agent. An ontology can be expressed informally using natural language, semi-formally (Ontolingua) and formally in first order logic languages such as the Knowledge Interchange Format (KIF). The brokering agent employs a disambiguation agent that uses the terms in the ontology to make assertions that will assist it in accomplishing its mission. For instance, the information broker uses standard terms in the ontology to describe products in a particular domain and identifies correspondences between the standard term(s) and terms used by different companies to describe the same product. In this way, consistent business semantics can be provided across different market domains and segments. Several projects that develop ontologies for use with the Web such as SHOE [4] and the Wide Knowledge Base (WebKB) [2] can be beneficial for use with agent technology. SHOE is a Web-based knowledge representation language that supports multiple versions of ontologies and ontology integration. WebKB uses ontologies and machine learning to achieve automatic classification of Web pages.

An example of a successful commercial product that uses agent technology and combines search and information agents is the Verity Corporate Portal (www.verity.com). This portal is a navigation aid that unifies corporate content and organizes it by using familiar business categories. Information from different systems and sources can be grouped together logically and navigated visually, reducing the cost of content management and increasing the usability of the corporate Intranet. Verity’s fuzzy search uses pattern-matching techniques based on spelling, grammar, synonym and topic matching. Wiznet (www.wiznet.com) is another example of an electric commerce portal, which creates customized virtual procurement marketplaces and provides a wide range of services such as sourcing, searching, buying, and communicating.

Negotiation and contracting agents. Agents may also elect to negotiate the terms of a business transaction as regards to exchange and payment. The process of negotiation can be stateful and may consist of a “conversation sequence,” where multiple messages are exchanged according to some prescribed protocol. Terms may, for instance, cover delivery, refund policies, arranging for credit, installment payments, copyright or license agreements, usage rights, distribution rights, and so forth. Negotiation protocols can be structured by making use of formalized conversations, such speech acts. These may, for example, be developed in KQML, which provides the means to define formalized negotiation messages describing offers, rejections, propositions, acceptance of terms and the like. Contract instructions can address liabilities, acceptance forms of payment, terms of payment, billing and payment instructions, delivery instructions, return policies, methods of dispute resolution, and so on. Contracts can be negotiated including prices, terms of payment, penalties, necessary documentation, credit checks or required insurance, or collateral or margin. Given such advanced characteristics and functionality it is only natural to consider the use of agent-oriented technology in support of negotiation and e-business contracts.

A successful example of a multiagent negotiation system, which can be adapted for use with e-business, is Tete-a-Tete [10]. This system provides a negotiation approach to retail sales, allowing consumer shopping agents and merchant-owned sales agents to cooperatively negotiate across multiple terms of a transaction including warranties, delivery times, service contracts, return policies, loan policies, and so on.

System-level support agents. System-level agents exist on top of the distributed objects infrastructure, typically implemented in CORBA by means of the Internet Inter-ORB Protocol (IIOP), which provides objects with transparent access not only to other application objects but also to such facilities as transaction processing, permanent object storage, event-services, and the like. Agent solutions are deployed as an extension of the distributed object foundation and may assist in accomplishing the following systems-related tasks. Some of the advanced functionality agents required to provide support for e-commerce and interoperation of open market business processes are described here.

Planning and scheduling agents. In a multiagent planning approach to cooperation, a multi-agent plan is formed that specifies the future actions and interactions for each agent. Typically, in e-business applications an agent may act as the group planner for a cluster of agents surrounding an application agent, for example, the procurement agent. The planning agent forms a plan, which it uses to coordinate the other agents. In case that two application agents need to interact they do so by means of their planning agents. The plan specifies how agents coordinate in a multiagent planning system and also identifies all actions and interactions of agents. However, in dynamic situations plans need to change over time. In such situations the multiagent planning approach would be less effective.

To work cooperatively in dynamic situations, planning agents require the ability to cooperate with other agents despite having inconsistent views of planned actions and interactions. To be able to reason under uncertainty agents must rely on AI technologies of fuzzy systems, neural networks, and genetic algorithms. SharedPlan is a formalization that enables developers to deploy planning agents in the dynamic e-business environment where agents are uncertain concerning their own actions and have incomplete information about the other agents and their environment [3].

Agents for interoperation. Supporting new or extended e-business processes requires “old” information systems (including legacy systems) and transactional software components interworking with new applications and systems [12]. Moreover, software components are the subjects for integration between companies in supply chain networks. The interoperability requirements mandate that the information flows that move between business processes (like the flow between any procurement processes in one organization and the order fulfillment process in another) are managed and merged with the legacy assets of an organization. For example, in the value-chain workflow application in Figure 1 information that may be acted upon by the various application agents may originate from different information (and possibly legacy) systems that provide such services as:

  • Entry to an enterprise resource planning (ERP) system checking inventory for the products described in the procurement order; and
  • Entry to a distribution/shipping company system using the customer address and delivery condition information to schedule a delivery.

Legacy systems can be componentized physically by developers or they can be perceived as components by means of their interfaces [12]. A translation agent can be used here to provide a common information model based on the underlying information models of each application. The translation agent provides the appropriate interface code (and translation facilities), which allows an existing information source to conform to the conventions of an organization. This agent is responsible for exposing the functionality of the underlying application in an abstract way and for translating information from the form required by the application to the common information model and vice versa. The business logic that enforces business rules and policies is then placed in components and business objects that are available for inter-application integration. Standard ontologies provide the basis for rectifying terminology mismatches and disambiguating agents provide mappings to the terms used in the interfaces of business objects.

Once diverse representations originating from heterogeneous information systems have been homogenized by means of the translation agents the following step is to selectively combine semantically interrelated legacy and other data into cohesive components. These components are then combined with business objects in the enterprise framework by means of an integration agent that provides consolidated access to distributed data and services by aggregating various types of data and functions at the systems level (see Figure 1).

The e-business interoperability challenge places particular emphasis on integration at the transaction level and not on data integration, replication and batch transfers of data. In addition, the virtual nature of the e-commerce end-to-end business processes requires that business rules and transactions be available to partners for incorporating within their own systems. Such considerations can benefit by the use of frameworks such as BizTalk (www.biztalk.org) and standards such as the Open Buying on the Internet (www.openbuy.org/obi). BizTalk is a framework for e-business integration through data interchange standards based on XML. BizTalk first defines how enterprises should use XML to integrate their processes and then augments that framework with other capabilities such as workflow and security. The OBI standard is an open, flexible design for business-to-business Internet purchasing activities. It defines a product-neutral architecture that ensures buying and selling organizations are able to interoperate.

Business transaction agents. A key activity in integrated value chains is the collection, management, analysis, and interpretation of the various commercial data to make more intelligent and effective transaction-related decisions. Examples include collecting business references, coordinating and managing marketing strategies, determining new product offerings, granting/extending credit, and managing market risk. Transactions in e-business are usually long-lived propositions involving negotiations, commitments, contracts, floating exchange rates, shipping and logistics, tracking, varied payment instruments, exception handling, and customer satisfaction. Business transactions are used to interchange everything from product information and pricing proposals to financial and legal settlements [12]. They can be dynamically constructed from data in relational databases and can include descriptive text and graphics from document management systems.

When applied to e-commerce transactions, business agents could simplify the processing, monitoring and control of transactions by automating a number of activities. Agent support for e-commerce business may, for example, include controlling the workflow governing a set of electronic transactions or monitor and enforce the terms and conditions of electronic contracts. Distributed workflow managers are a special type of system-level agents, which may track business transactions across unit, company and enterprise boundaries. As part of a multiagent system, workflow agents can capture and apply semantic constraints among processes spanning diverse departments and organizations in order to enact distributed workflows, see Figure 2.

Business transaction and workflow agents can be influenced by products such as Movex (www2.intentia.com), which provides integrated enterprise management solutions for the management and control of business transactions in a supply chain network. The Movex intelligent agent can be used to search the Internet and negotiate deals within different areas such as customer relationship management, supply chain management, and maintenance, repair, and overhaul activities.

Security agents. E-business communication need to be guarded by specially designed agents that provide the security services required for the conduct of e-business. Agents can, for example, be used to collect commercial data only from trusted and controlled sources. Agent support for secure e-business can be segmented into five distinct categories: authentication, authorization, data integrity, confidentiality, and non-repudiation, each of which requires specialized agents:

  • Authentication agents can be used to identify the source of a message sent over the Internet.
  • Authorization agents may control access to sensitive information once identity has been verified. Thus, certain transactions may need to be partly accessible to certain parties, while the remainder of the transaction is not. The transaction workflow and authorization agents can coordinate these tasks.

Secure transactions should guarantee that a message has not been modified while in transit. This is commonly known as integrity and is often accomplished through digitally signed digest codes. Transactions should also guarantee confidentiality. Confidentiality refers to the use of encryption for scrambling the information sent over the Internet and stored on servers so that eavesdroppers and interlopers cannot access the data.

Non-repudiation is of critical importance for carrying out transactions over the Internet. It consists of cryptographic receipts that are created so that the author of a message cannot falsely deny sending a message. Again, these tasks fall well within the premises of agent technology. In [6] the authors describe a security model and architecture for Aglets Workbench, a Java-based environment for building mobile agent applications based on these ideas.

Back to Top

Conclusion

Agent-oriented technology can help enable the development of intelligent business agents, which are the next higher level of abstraction in model-based solutions to e-business applications. This technology allows the development of rich and expressive models of an enterprise and lays the foundation for adaptive, reusable business software. Agent-oriented technology can be leveraged to enhance enterprise modeling as well as to offer new techniques for developing intelligent applications and infrastructure services. By building on the distributed object foundation, agent technology can help bridge the remaining gap between flexible design and usable applications.

In a multiagent e-business environment it is necessary to organize agents into different categories depending on their functionality and competencies. We distinguish four basic forms of agents: application, personal, general business activity, and system-level agents. Each of these classes of agents draws on a host of agents with specialized functionality to implement a variety of e-business application related tasks. The agent approach to e-commerce provides conceptual simplicity, enhances scalability and makes interactions in a large collection of information sources become tractable.

Back to Top

Back to Top

Back to Top

Figures

F1 Figure 1. Business agents in a vertical application.

F2 Figure 2. Business agent typology.

Back to top

    1. Chen, L. and Sycara, K. WebMate: A personal agent for browsing and searching. In Proceedings of the 2nd International Conference on Autonomous Agents, 1998.

    2. Craven, M. et al. Learning to extract symbolic knowledge from the World Wide Web. In Proceedings of the AAAI-98 Conference on Artificial Intelligence, 1998.

    3. Grosz, B.J. and Kraus, S. Collaborative plans for complex group action. AI Journal 86, 2 (1996).

    4. Heflin, J., Hendler, J., and Luke, S. Coping with changing ontologies in a distributed environment. In Proceedings of the AAAI-99 Workshop on Ontology Management, (Orlando, Florida, July, 1999).

    5. Jennings, N.R. and Wooldridge, M. Applications of intelligent agents. In N.R. Jennings and M. Wooldridge, Eds., Agent Technology Foundations, Applications and Markets, Springer-Verlag, 1998.

    6. Karjoth, G., Lange, D., and Oshima, M. A security model for Aglets. IEEE Internet Computing 1, 4 (July 1997).

    7. Klusch, M., Ed. Intelligent Information Agents. Springer-Verlag, 1999.

    8. Lieberman, H. Integrating user interface agents with conventional applications. In Proceedings of the ACM Conference on Intelligent User Interfaces (San Fransisco, Jan. 1998).

    9. Maes, P. Agents that reduce work and information overload. Commun. ACM 38, 11 (Nov. 1995).

    10. Maes, P., Guttman, R., and Moukas, A. Agents that buy and sell. Commun. ACM 42, 3 (Mar. 1999).

    11. Papazoglou, M.P., Laufmann, S. and Sellis, T.K. An organizational framework for cooperating intelligent information systems. International Journal of Cooperative Information Systems 1, 1 (1992).

    12. Yang, J. and Papazoglou, M.P. Interoperation support for e-business. Commun. ACM 43, 6 (June 2000).

Join the Discussion (0)

Become a Member or Sign In to Post a Comment

The Latest from CACM

Shape the Future of Computing

ACM encourages its members to take a direct hand in shaping the future of the association. There are more ways than ever to get involved.

Get Involved

Communications of the ACM (CACM) is now a fully Open Access publication.

By opening CACM to the world, we hope to increase engagement among the broader computer science community and encourage non-members to discover the rich resources ACM has to offer.

Learn More