SERVICE-CENTRIC BROKERING IN DYNAMIC
E-BUSINESS AGENT COMMUNITIES
Sumi Helal and Mei Wang
Computer and Information Science and Engineering Department
University of Florida, Gainesville, FL 32611
Tel: 352-392-6845, Fax: 352-392-3238
I I ,. ,. Itl/,.. ,l
The paper focuses on the development of protocols for brokering-based agent
communities in e-Business. The global market is driving e-Business to become
service-centric, offering highly modular e-Services that can be flexibly and
dynamically composed into rapidly deployable e-Businesses. This trend is
giving rise to a new set of requirements of negotiation-based, autonomous, and
intelligent computing. It is, therefore, expected that in the near future, e-Services
will be designed and implemented as software agents (also known as agent-
based systems). This paper prepares for the proliferation of agent-based systems
in e-Business by contributing a suite of protocols for self-organizing agent
communities. The protocols are based on a three-tier architecture of agents,
brokers, and superbrokers. We present the architecture and the protocols
(Broker-Based Agent Community Protocols, or BBACP). An implementation
using JKQML is also presented along with a case study drawn from the
electronic auto-trading domain.
Conducting business electronically is not a new idea anymore. The development of e-Business has
been observed in several ways: flexibility requirement in the business market, emerging virtual enterprise
or net enterprise, and business globalization . More and more small or medium size enterprises or
private households are connected with each other through the Internet. However, such web-based e-
Business applications and systems developed by individual companies are neither compatible nor
interoperable with each other. New mechanisms and technologies need to be invented in order to make
individual standalone e-Business interoperable and cooperative. In our view, agent technology is one of the
core technologies that will accommodate the growing proliferation of e-Business.
Agent technology has been widely adopted in the artificial intelligence and computer science
communities. An agent is a computational system that operates autonomously, communicates
asynchronously, and runs dynamically on different processes in different machines, which support the
anonymous interoperation of agents. These qualities make agents useful for solving issues in information
intensive e-Business, including speaking ontologies, advertising, service exchange, and knowledge
discovery, etc. In a fast changing e-Business environment, the interoperation and coordination across
distributed services is very important. The desire for more cost efficiency and less sub-optimal business
processes also drives the employment of agent technology in e-Business.
With the support of agent technology, more e-Business agents will appear on the Internet providing e-
Services as well as exchanging information and goods with other agents. The interoperation of e-Business
agents leads to the formation of the e-Business Mall, which is an interaction space of agent communities
under various business domains.
The significant problems in the e-Business are the information deficiency and asymmetry between the
business participants. It is also difficult for each participant to exchange information products and services
in an efficient manner, and to partner in a virtual enterprise. We contribute the concept of agent
community, in the context of e-Business as an approach to those problems.
An e-Business agent community is a self-organized virtual space consisting of a large number of
agents and their dynamic environment. Within a community, highly relevant agents group together offering
special e-Services for a more effective, mutually beneficial, and more opportune e-Business. Each agent
community consists of agents specializing in a single domain/sub-domain, or highly intersecting domains.
The ideal e-Business agent communities aim to
make it easy to develop virtual enterprises, on which companies are joining and sharing resources
and business processes for the production, marketing, and other services, provide protocols for the
exchange of information, products and services among business entities
represent certain potential relationships among potential members, and support the interoperation
help in the establishment of relationships among agents with common interests
provide discipline in the open network in anticipating of e-Business proliferation
accommodate solutions to issues such as trust, efficiency and credentials
In section II, we present the layered architecture of agent communities and describe the specification,
analysis, and design of agent organizational structure. Section III addresses the role of ontologies in agent
communities. Section IV presents the components of the Broker-Based Agent Community Protocols
(BBACP) in detail. The protocols or rules in the service-centric Brokering Layer include processes of
joining e-Business in the community, knowledge discovery and exchange, and advertisement of services or
capabilities. In section V, we illustrate how the community and the supporting protocols are designed and
implemented using the JKQML  agent shell. We present the functionality of the community components
and our implementation of the layered community architecture and protocols as Java APIs supporting
interoperation within an agent community. Finally, an e-Business case study based on the BBACP protocol
is demonstrated, showing the benefits of service-centric brokering in self-organized agent communities.
II. ARCHITECTURE OF AGENT COMMUNITY
In order to meet the challenges in the developments of future e-Business Communities IAiil a
loosely coupled architecture with heterogeneous components is required for different e-Businesses. We
introduce a hierarchical brokering architecture for the e-Business agent community to support the
appropriate level of flexibility, scalability and interoperability in e-Business as well.
Figure 1. Hierarchical Architecture of an Agent Community
As indicated in Figure 1, an agent community is composed of an Agent Layer, a Brokering Layer and a
Super Brokering Layer. The Agent Layer is the bottom layer in the community architecture hierarchy,
where each agent or agent based system (ABS) provides specific e-Service(s) and inter-operate with each
other. In the e-Business environment, service-centric agents self-organize together to exchange information
or product under a certain knowledge domain. The business processes among agents are completed under
the facilitating of a special agent, a broker. A broker represents and facilitates a set of agents with similar e-
Services and common interest in the e-Business community, which forms a brokered agent system.
The autonomous e-Business agents organize together based on the specialization of an individual
broker that represents the best interest of a set of e-Services. With the broker's facilitating, information and
services exchange among heterogeneous agents becomes efficient. Agents communicate their needs and
capabilities toward brokers. Brokers use the acquired agent knowledge to reply or rout messages to the
appropriate agents. A brokered agent system provides a dynamic structure as agents can be introduced and
removed at runtime. Moreover, a brokered agent system can be built from a set of heterogeneous agents,
which assists in creating a very flexible coordination in a loosely coupled system. We assume that a useful
domain ontology exists and is being shared by agents in the brokered agent systems.
In the Brokering Layer, the interoperation of various brokers forms a multi-brokering system, which
consists of multiple brokers and a superbroker. A superbroker is a well-known broker in the multi-
brokering system, supporting the knowledge sharing efforts among registered brokers. A broker advertises
or unadvertises e-Services to the superbroker it registers with. Usually only the service requests that can't
be satisfied within a brokered agent system are sent to other brokers within the community. The capable
brokers in the community provide the requester with the e-Service, which is analogous to the expansion of
the knowledge of requesting brokers. A superbroker maintains the service subscriptions from its members
and notifies the subscribers as long as the right capabilities are available in the community. In the mean
while, individual brokers maintain their own autonomy, facilitating interaction of self-organized member
agents. In a multi-brokering system, brokers operate under a common protocol, the Brokering Protocol,
that governs how they advertise capabilities and services of agents they represent, how they reason and
select which broker is appropriate for the requested services, and how they communicate and coordinate
over community common ontologies.
On the top of the hierarchy is the Super Brokering Layer. In this layer, superbrokers interact with each
other directly or through superbroker consortia, which eventually leads to the concept of virtual enterprises
and e-Business Malls. In this layer, superbrokers operate under a common protocol, the Super Brokering
Protocol, that governs how they communicate and coordinate over common ontologies, how they reason
and solve the community issues, such as trading brokers or other community restructuring.
The community is easily scalable by accepting new brokers that represent a set of agents. When the
new set of agents represented by a broker joins the community, the superbroker becomes the entry point for
interaction to the new broker. The minor communication overhead also ensures the necessary scalability for
the agent community. Upon a new broker joining or leaving the community, the change of the knowledge
and services will be updated in an information repository maintained by the superbroker and known to each
of the existing member brokers. Superbrokers have tight control over load balancing, support great service
reliability, and provide potential wide ranges of services in the e-Business community.
Researchers have designed different organizational structures for brokering systems, such as the peer-
to-peer multi-brokering architecture in InfoSleuth , the partitioned repository design in blackboard
systems , centralized message routing design in JA T Lite , and the multi-facilitator mediation in OAA
. Comparatively, the hierarchical agent community is designed to meet the requirements of the
proliferation of e-Business agents, to accept new brokers without the direct necessity of adjustment on the
Agent Layer. In the Brokering Layer, brokers communicate with each other directly or indirectly under the
superbroker supervision. The communication and interaction protocols specified in Broker-Based Agent
Community Protocol are capable of avoiding a single point failure, not like those in JAT Lite and
blackboard systems. What's more, it is obvious that the knowledge of all other brokers is not necessary.
Each broker only needs to understand the superbroker it registered with, and communicates with some
other brokers upon necessity. Thus the amount of messages flowing among the brokers within the
community is minimized, providing the solution to the network bandwidth restriction imposed on various
III. THE ROLE OF ONTOLOGY
In order to self-organize into communities, each broker should contain a generic broker instance,
which implements at least the basic infrastructure for an agent community. Some assumptions are
necessary to the joining protocol:
Figure 2. Universal Community Ontology, Domain Ontology and Community Specific Ontology
Each generic broker instance has a Universal Community Ontology and can access various publicly
available Domain Ontologies (Figure 2). The Universal Community Ontology is the fundamental ontology
applicable to all e-Business communities, which provides brokers the minimum amount of vocabularies
and knowledge toward self-organizing features. The Universal Community Ontology enables brokers to
interact with a superbroker, conveying the spoken ontologies, and knowledge capabilities among other
information. A Domain Ontology provides vocabularies and knowledge within an e-Business sub-domain.
In each e-Business community, the Community Specific Ontology should be made available and open
to both community members and applicants. Thus, prospective community members are able to browse and
learn about the available Community Specific Ontology maintained by the superbroker before initiating any
application. Based on Community Specific Ontology, the applicant broker can determine the relevancy
between its own knowledge and that of a community, as well as other self-organizing factors. Only if
knowledge relevancy and value added are expected, would further procedures need to proceed.
The Universal Community Ontology and Domain Ontologies possessed by the applicant broker enable
it to initiate an application procedure. Applicant brokers have the goals of learning more about the
Community Specific Ontology and other community facts. The superbroker maintains the Community
Specific Ontology and acknowledges them to the applicants during the joining procedure. The joining
procedure occurs when an applicant broker sends a membership application to the superbroker. In a
service-centric agent community, membership application presents its knowledge as a constrained sub-
domain of the e-Business domain. By using the Universal Community Ontology and Domain Ontology, the
broker provides the general information about itself, including basic information (e.g. name, location,
status) and capabilities (e.g. knowledge of a sub-domain, e-Business transaction types, service credential
characteristics). Currently, the Universal Community Ontology is implemented and described using KQML
 as low-level language.
IV. PROTOCOLS FOR AGENT COMMUNITIES
We introduce the Broker-Based Agent Community Protocols (BBACP) to enable inter-organizational
business procedures. The BBACP consists of an Agent Protocol, a Brokering Protocol, and a Super
Brokering Protocol, which provide mappings from states to action in each agent community layer.
The Agent Protocol involves interaction with other agents, which varies so as to suit to the business
application and usage scenarios. Each agent-based system follows a local protocol, the Agent Protocol. A
local protocol is a function from local states to local actions, such as applying rules or sending messages
. Therefore the Agent Protocol is the base-level commitment and responsibility to the community.
Because it is an open protocol, much space is left for negotiation when agents are built. The autonomy of
the agents themselves also requires self-government and self-organization. By accepting the Agent
Protocol, agents commit themselves to play special roles and to guarantee properties such as sincerity and
producing appropriate responses .
The Brokering Protocol extends the brokering function to meet the needs of the interoperability among
the brokers and with the superbroker. The Brokering Protocol describes a cooperative multi-brokering
system, which provides the solution to interoperation in a dynamic heterogeneous agent community. Each
broker performs basic brokering functionality (such as service discovery and knowledge sharing) within a
set of e-Business agents, as well as representing e-Services in the community. A superbroker is on the top
of the multi-brokering hierarchy, plays the monitoring role in the agent community and requires the
registration of member brokers. Individual brokers representing a set of agents are allowed to advertise
their business service and send capability queries to other brokers, as well as receive advertisements from
other brokers. In addition to regulation of joining in and leaving from the community, other issues covered
in the Brokering Protocol include business knowledge discovery and sharing, information load control, and
The Super Brokering Protocol governs how a superbroker communicates and coordinates in the e-
Business Mall, how they reason and solve the community issues, and how to form and maintain the
superbroker consortium. Our research focuses on the Brokering Protocol layer, which includes the
IV-A. The Joining Protocol
The protocol of joining a community implements a logic conversation between brokers and
superbrokers. The conversation is a sequence of rule-based communications. Conversation policy (or rule)
specifies the social behavior of brokers and superbrokers at each state. The conversation policy is mapped
into a Finite State Machine (FSM) as illustrated in Figure 3.
B->SB SB->B B->SB SB->B
SB->B B->SB B->SB
Reject Abort Pendmg
Figure 3. FSM for Community Joining Protocol
Each broker declares to the community what kind of ontology concepts it can deal with, enlisted in its
application to the superbroker. After receiving a registration attempt (State 1), the superbroker conducts
multi-faceted membership evaluation of the applicant, such as knowledge and capability relevancy,
creditability, possibility of value-added, potential expansion, among others. The conclusion of the
evaluation is either reject (State 1 -> State 6) or acknowledge (State 1 -> State 3). The acknowledgement
message contains three components: Domain Ontology, Community Specific Ontology and Bylaws.
At State 3, the broker received the acknowledgement and automates its behavior related to ontology
adjustment or other knowledge changes.
If a positive conclusion can be made immediately, a message notifying the acceptance of Domain
Ontology, Community Specific Ontology and Bylaws will be sent to the superbroker (State 3 -> State 4).
The finite state (State 5) is reached when the superbroker confirms the grant of membership upon receiving
the acceptance message.
If a negative conclusion is made, an abort message will be sent to the superbroker (State 3 -> State 6),
which indicates the end of the conversation resulting in the finite state (State 6) being reached.
If a decision can not be made, a pending message is sent to the superbroker. In order to allow the
applicant to continue the member application in the future, the superbroker must enforce the knowledge of
access history. It is clear that the state information should be kept and used in guiding conversation policy.
The question is where the state information should be kept. Our approach is to embed state information in
the conversation policy in the superbroker side. During the application processing in the future, the
superbroker will check the applicant accessing history, thus following the conversation policy based on
If no replies are received by the superbroker within a specified period of time, the synchronous
conversation will move to the finite state of application failure (State 6).
Upon approval of a successful registration, the superbroker updates the new broker in its repository.
Failure of registration could encourage broker to engage in additional learning activities. Thus the
superbroker manages the relationship between ontologies and brokers, and maintains ontology consistency
IV-B. The Departure Protocol
The unregisteration is used to inform the superbroker that the broker won't be able to provide the
advertised service anymore. The leaving procedure is initiated by the leaving broker sending an
"UNADVERTISE" message to the superbroker. The updated knowledge and services available to the
community are "TELL" to all members. Alternatively, the leaving broker can broadcast the
"UNADVERTISE" message to member brokers directly. The previous approach is implemented in the
thesis. Upon receiving the unadvertsing message, superbroker and related brokers remove the services from
their repositories. The obtained knowledge in the leaving brokers is removed as well.
Then the leaving broker sends an "UNREGISTER" message to the superbroker. Upon receiving the
message, the superbroker removes the broker's entry from the member repository and marks it in the
member credit system. The member credit system is the repository maintaining the history and service
quality of previous and existing members. It acts as a quality monitor to ensure the knowledge sharing
efforts in the community and prevent the malfunctions of community members.
IV-C. The Knowledge Discovery and Exchange Protocol
The knowledge discovery is initiated by the inquiring broker, who multicasts the service requests) to
the community. Upon receiving the requestss, the capable brokers reply with appropriate service
information that enables the requester to communicate with the service providers and obtain services from
them directly. In our implementation, the inquiring broker sends a request to a superbroker. The
superbroker searches its capability repository for services) that satisfy the request. If there exits at least one
match, it sends the result as a "REPLY" KQML message to the requester. The received service information
contains the required parameters to communicate with the service provider through KQML messages. Thus
the requester is able to directly negotiate with the service providing brokers) using specified language and
ontology. The amount of knowledge discovery is based on the brokers' objectives: ASK-ONE (one capable
service is recommended), ASK-ALL (a list of services are returned).
Subscription means that a broker requests to be informed if some specific service becomes available in
the community, which is another approach of knowledge discovery. The subscription is presented in the
performative "SUBSCRIBE". A broker may broadcast a subscription to the community so that each
member broker stores it. If a broker is capable of serving the subscribed service, it will reply to the
subscription and start a conversation with the message sender. Otherwise, the receiver does not response
except for storing the subscribed subjects in its subscription repository. Once a broker, or an agent it
represents, develops new capabilities that meet the specification of the subscription, the broker will notify
the subscriber and initiate conversation. In our implementation, a superbroker acts as a special broker that
maintains all the capability information in the community. Thus upon receiving a subscription, the
superbroker searches the community capabilities repository to see if any services available in the
community match the requests, and checks if the capable broker is up and running. Then, the query result is
sent to the subscriber by using the "TELL" performative. If no knowledge is able to serve the request at the
time being, the subscription is stored in the superbroker's subscription repository. If some new services
satisfying the subscription are advertised in the community, the superbroker then notifies the subscribers of
the availability of the new capabilities. With the received information, the subscriber can then obtain the
service by initiating the conversation with the service providerss. On that occasion, the knowledge sharing
efforts are directly conducted among the subscribers and the service providers.
The unsubscription is used to inform the community that a broker no longer needs to be informed for
the subscribed service. A broker can unsubscribe one service at a time by "UNSUBSCRIBE" performative.
The "UNSUBSCRIBE" message is either multicast to all community members directly or sent to a
superbroker and consequently sent to member brokers. In either approach, the receiving brokers remove the
subscription from their subscription repositories. We implement the latter approach. When a superbroker
receives an unsubscription message, it removes the corresponding broker's entry from the subscription
repository so that no more notifications are sent to that broker.
Knowledge discovery follows knowledge discovery rules (a finite state machine). If a satisfied
discovery is received, e.g. "TELL", the broker updates its acquired capabilities repository; if unsatisfied
discovery, e.g. "SORRY", it could refine the request content and re-send it to the superbroker.
IV-D. The Capability Advertisement Protocol
The advertisement is sent first to the superbroker by a broker who wants to advertise its capabilities to
the community. The advertisement specifies the required KQML fields, which the interested broker has to
use in order to be able to get the desired service. The performative is "ADVERTISE". If a broker wants to
advertise multiple services, it does so by sending multiple advertisements. The advertisement should also
be broadcast to the community by sending an "ADVERTISE" message to each member broker.
When it receives an advertisement, a broker updates its repository by adding a new entry for the new
advertised services, inserting the name of the broker, the credential information, and domain dependent
information. If the advertised service is new, or the advertisement exists and adjustment is made, then a
"TELL" message is sent back to acknowledge the advertiser. Otherwise, a "SORRY" message is sent to
inform the advertising broker about the failure.
The unadvertisement process is triggered whenever changes are made to any services provided by an
individual agent. Upon receiving a service change message from its member agent, a broker updates its
self-capability repository, which means either removing the entry permanently or updating the
characteristics of this particular service. The broker then sends an "UNADVERTISE" message to the
community to indicate the necessary changes needed to keep member's information repository up-to-date.
Upon receiving the unadvertising message, all member brokers update their service information
In our implementation, a superbroker has much broader knowledge and maintains all the
advertisements in the community. Services are advertised and unadvertised to the superbroker and
interested members and then notified by the superbroker. Since the superbroker maintains the subscriptions
from its members, the flow of advertising messages is reduced by only sending to subscribers and
requesters instead of broadcasting to all community members. Upon receiving an advertisement, the
superbroker first updates the entries in the capability repository. Then it checks the subscription repository
to see if the advertised service satisfies any subscriptions. If one or more matches are found, the
superbroker notifies the corresponding brokers) by sending a "REPLY" message in which the service
information is specified.
Upon receiving a service-changing message from its member agent, a broker sends an
"UNADVERTISE" message to the superbroker to indicate the necessary changes needed to keep the
information repository up-to-date. The superbroker updates the community capability repository by either
removing the entry permanently or updating the characteristics of this particular service. Subsequently it
sends notifications to member brokers who have acquired or subscribed to that particular service.
Agents operate and interact in the infrastructure provided by the agent shell-JKQML. Figure 4 shows
the design and implementation of our agent community architecture.
User Community-Ready Agent Operator Interface
SI Agent Community Model = JKQML
Figure 4. Agent Community Architecture Design and Implementation
The lowest layer is the JDK 1.2 that compiles the Java source code into bytecode that can be executed
in any open network through Java Virtual Machine. Above the Java infrastructure rests the Java-based
KQML (JKQML) , the agent shell (or agent development toolkit). Developed by IBM, JKQML is a
framework to construct KQML-speaking software agents, to build loosely coupled distributed systems by
using an agent communication language-KQML , to provide flexible interoperability to exchange
information and services between software systems. JKQML is written entirely in the Java language to
construct software agents communicating over the Internet. It is designed to support various transport
protocols, and to provide flexibility for the extension of the framework. As show in the figure, it provides
interfaces to process KQML messages for agent application through the KQML Manager. The agent
applications can access other major components such as conversation pool and protocol manager through
these KQML Manager's interfaces. Currently JKQML supports three transport protocols, one is the KQML
Transfer Protocol (KTP) for the ordinary TCP/IP environment implemented in Java.net.Socket, the Agent
Super Brokering Protocol
Join/Leave Knowledge Discovery Service Advertisement
Protocol & Exchange Protocol Protocol
SIni 1 Ir '' u r It u I 21
Inl r.i.ll ti r IIK I "I I
Transfer Protocol (ATP) for the Aglets environment , and the Object Transfer Protocol (OTP) for
transferring Java Object.
The agent community model defines how agents interact with each other in a cooperative manner. It
utilizes and interacts with the KQML Manager and extends the functionality to support the BBACP, which
is composed of three sub layers: Agent Protocol, Brokering Protocol, and Super Brokering Protocol. The
Agent Protocol is the base-level commitment and responsibility to the community, which differs among
different business providers. Because it is an open protocol, much space is left for adjustment when agents
are built. The middle layer of the community model is the Brokering Protocol that provides the inter-
brokering among community members. The upper level of the community model is the Super Brokering
Protocol, which extends the concept of agent community of virtual enterprise to the e-Business Mall.
We implement the Java classes such as CRAgent, Broker, SuperBroker, and SocialViewer, each of
which captures one or multiple functions in the community. There are two inheritance relationships in
Figure 5. The Broker class inherits the Agent class and the SuperBroker inherits all the functions of the
Broker class. Inter-agent and inter-broker interaction is also illustrated.
Figure 5. Community Components Diagram
Each broker, implemented as Broker class, maintains three information repositories: self-capabilities
repository, acquired-capabilities repository, and subscription repository.
The superbroker, implemented as SuperBroker class, maintains the service information repository,
guarantees relevance and affinity, enforces interoperations and knowledge sharing efforts, and tracks
member brokers' credential characteristics. The superbroker can monitor the knowledge sharing efforts
among members, provide efficient load balancing, and keep track of other credential information of each
broker. The credential information repository contains the broker name, location, ontologies, status, cost of
services, reliability, and quality. By maintaining credential information of member brokers, the superbroker
can initialize a membership evaluation procedure. All members vote on the quality of every membership
and submit voting results to the superbroker. Based on the voting result, the superbroker expunges the
membership of unqualified brokerss.
The social viewer is a special agent that provides a domain-independent social view service. It is not an
essential part of our architecture even through it views and analyzes social roles of agents, brokers, and
superbrokers, shows their organizational inter-relationship, as well as indicates the message exchange and
collective behaviors between brokers and superbrokers during their interoperations.
Our research focuses on the middle layer of community protocols, the Brokering Protocol layer. Our
implementation provides community-ready interface, and supports the self-organizing feature in open e-
Business communities. The community-ready interfaces are sets of APIs, allowing software programmers
to easily write the agent applications that inherit the functionality of the organizational roles needed in the
agent communities. The complexities and details of the protocol implementation are hidden from
programmers. The APIs allow programmers to extend the classes and to overwrite the methods so as to
implement more functions for specific business domains while maintaining the essence of brokered
Appendix A summarizes the community-ready API for the BBACP protocols. It shows the API
components of each layer and describes component interactions through KQML performatives. More
detailed explanations of the API implementation can be found .
VI. CASE STUDY
We demonstrate an application of the agent community and the community protocols to the on-line
auto-trading domain. The objective is to enable the business participants to electronically buy and sell cars
over an open network via e-Business agents.
To present the characteristics of the business process of buying and selling cars, we assume there are
several brokers, one superbroker and a variable number of agents. The size and type of brokers are not
limited to the description. As an open network, the auto trading community allows new brokers to join
based on their knowledge relevancy and value-added to the community. Currently we designed a simple
community consisting of participants such as: Individual Buying Broker, Dealing Broker, Individual
Selling Broker, Manufacturer Broker, Warranty Broker, Auto Transport Broker, Vehicle History Broker,
Auto SuperBroker, Whitepages and Yellowpages.
As described in the domain specification, all brokers automate the trading process among themselves.
However, the direct business protocols are facilitated and maintained by a specialized broker, the
AutoSuperBroker. The multi-brokering trading community contains several roles. It is possible for each
individual roles to be played by an individual broker. Alternatively, a single broker can play several roles.
Each role played by a broker contains some responsibilities, such as sending registration in order to
join the trading community, advertising its services or requesting one, and interacting with other type of
brokers. This section introduces the role responsibilities for each broker including Member Applicant,
Membership Inquirer, Services Advertiser, Service Inquirer and Subscriber and Broker. Tables 1 to 5
describe each responsibility in detail.
Table 1: Auto Trading Community Role Descriptions-Membership Applicant
ROLE MEMBERSHIP APPLICANT
Role Model Auto Trading Community
Description Register with AutoSuperBroker and obtain community ontology and bylaws
To send membership application message Destination Role: AutoSuperbroker
To send leaving community message
Explanations By default, each broker is responsible for registering as a member in order to participant the any trading
Table 2: Auto Trading Community Role Descriptions- Membership Inquirer
ROLE MEMBERSHIP INQUIRER
Role Model Auto Trading Community
Description Query with AutoSuperBroker
To query and receive member information of a named broker or Destination Role: AutoSuperbroker
a list of brokers
Explanations By default, each broker is capable of querying and notifying
Table 3: Auto Trading Community Role Descriptions- Services Advertiser
ROLE SERVICES ADVERTISER
Role Model Auto Trading Community
Description Advertise services the broker represents to AutoSuperbroker or direct to Brokers
To send advertisements or unadvertisement message of agent Destination Role: AutoSuperbroker
capabilities and services it represents Broker
Explanations Each broker advertises the services it represents through AutoSuperBroker or direct to multiple brokers.
The business relationship usually groups a set of brokers as service partner. Brokering protocol specifies
how the service advertisement are represented.
Table 4: Auto Trading Community Role Descriptions-Services Inquirer and Subscriber
ROLE SERVICES INQUIRER AND SUBSCRIBER
Role Model Auto Trading Community
Description Request the AutoSuperbroker or other brokers for services that could not be handled by broker itself.
To query and obtain brokers' information with certain service Destination Role : AutoSuperbroker
from the AutoSuperBroker, or
To query and negotiate contract specifications with particular Broker
To subscribe to and obtain certain types of services
Explanations Each broker requests services information from AutoSuperBroker and conducts business contracting with
one or multiple brokers.
Each broker provides the services and respond any inquirer and subscriptions.
Table 5: Auto Trading Community Role Descriptions- Broker
Role Model Auto Trading Community
Description Provide service brokering in particular brokered agent system, include process member applications and
unregistrations, maintain service advertises of its members and provide capabilities matching and routing.
The major goal is to hide all the multi-brokering interactions from member agents.
To maintain directory of member and their abilities, Source Role: Agent
To receive and process service advertisements and queries,
To serve the request of agents without their direct involving the
contracting. The interaction details among multiple brokers are
hidden from the answers to agents
To negotiate service contracts with other brokers. Source Role: Broker
The basic objective of the auto trading community is to model the process chain by which businesses
add value to all participants and support pair-wise contracts among multiple brokers to fulfill transactions.
Figure 6. Interaction, Electronic Contract and Transaction in Auto Trading Community
Figure 6 illustrates the relationships between organizational units, processes in the trading community.
It also illustrates the information flow between these entities.
Being members of brokered systems, individual agents can take advantage of all the capabilities and
services in the whole community. However, individual agents do not have to understand all the business
process and contract specifications. All business deals can be made through the representative brokers. A
buying agent registers with the Individual Buying Broker without any further knowledge about dealers,
manufactories, moving companies, etc. The Individual Buying Broker can buy from the Dealing Broker,
the Manufacturer Broker, or the Individual Selling Broker, each of whom represents several specialized
agents. In the best interest of their member agents, Brokers interact and negotiate with each other based on
the service knowledge they share. A service contract usually involves specifications of multiparty. For
instance, the Individual Buying Broker buys a new Honda Civic from the dealer in Miami, while it makes a
deal with the auto moving companies to transport the car from Miami to Gainesville at particular date with
insurance. The car is actually routed directly from the factory. Thus the additional shipping expenses are
handled by the dealer. The car comes with a manufacturer warranty and 6 year extended warranty with
discount price from a warranty provider, the business partner of the car dealer. This particular buying
process involves various brokers and the service contracts among them.
A self-organizing community requires basic ontologies and protocols to support communications and
interoperations among heterogeneous agents. The social nature of knowledge sharing carries high
complexity. The capability advertisement and knowledge discovery should be achieved by message
interaction among dynamic processes.
This paper addresses these issues. It designs a layered agent community architecture, defines protocols
guiding interoperations across the open network, and provides a set of community-ready and application
independent Java classes for e-Business application in various domains and sub-domains.
We have contributed four community protocols that guide the conversations and interoperations among
different role players in the community. The knowledge discovery protocol and capability advertising
protocol, where service-centric brokers share capability information with each other, are shown to enhance
individual brokers' capability of solving problems within a set of agents. The joining protocol and leaving
protocol, where both the superbroker and the applicant brokers evaluate the knowledge relevance, resolve
the self-organizing issues in a dynamic e-Business environment.
We have also contributed a modeling that facilitates the conceptualization of the e-Business community
through the encapsulation of natural business components into agents. The use of the auto trading business
domain in the case study verifies the system features such as domain independence, user-friendly interface,
and service and capability expandability.
Knowledge or service relevancy is one basis of self-organization of community brokers. Brokers join
or leave a community based on their relevancy and consequently, their added value to the community.
Currently, we use simple methods to determine such relevancy. More sophisticated algorithms need to be
developed to more accurately predict a broker's relevancy to a community, as well as to bring the approach
a step further towards introduction to real life business procedure and service contracting.
 Aglets, http://www.trl.ibm.co.jp/aglets/, Sept 1999
 Carver, N. & Lesser, V., "The Evolution of Blackboard Control Architectures," CMPSCI Technical
Report 92-71, Oct 1992.
 Dan, A. et. al., "The Coyote Project: Framework for Multi-party E-Commerce," Lecture Notes in
Computer Science, Vol.1513, pp873, Springer-Verlag, Heidelberg 1998.
 Finin, T., Labrou, Y. & Mayfield, J., "KQML as an agent communication language In J.M.
Bradshaw, editor, Software Agents, AAAI Press, 1997.
 Huhns, M.N. & Singh, M.P., "Readings in Agents," Morgan Kaufmann Publishers, San Francisco,
 JKQML, http://www.alphaworks.ibm.com/formula/ikqml, May 1999.
 JATLite, http://iava.stanford.edu/, May 1999.
 Kendall, E.A., "Agent Roles and Role Models,"
http://www.labs.bt.com/projects/ibsr/papers/pattems/iaipm.ps.gz, Aug 1999
 Martin, G.L., Unruh, A., & Urban, S.D., An Agent Infrastructure for Knowledge Discovery and
Event Detection," http://www.mcc.com/projects/infosleuth/publications/, Oct 1999.
 Nodine, M.H., Bohrer, W., & Ngu, A., "Semantic Brokering over Dynamic Heterogeneous Data
Source in InfoSleuth," ICDE 1999: 358-365, 1999.
 Ouzounis, V., and GMD-Fokus, "R&D for New Methods of Work and Electronic Commerce,"
State-of-the Art and Visions Workshops, Dec 1997-Apr 1998, Brussels, July 1998,
http://www.ispo.cec.be/ecommerce, Oct 1999.
 Vreeswijk, G.A.W., "Open Protocol in Multi-Agent Systems," Technical Report CS 95-, University
of Limburg, Jan 1995.
 Wang, M., "Service-Centric Brokering in Dynamic e-Business Agent Communities," Master
Thesis, CISE Department, University of Florida,
http://www.cise.ufl.edu/-helal/~student/thesis/Mei, Dec 1999.
 Zimmermann, H. D., "Business Media: A new approach to overcome current problems of
Electronic Commerce," Benbasat, Izak; Hoadley, Ellen: Proceedings of the 1998 Americas
Conference on Information Systems AIS '98. Baltimore, Maryland, Aug 1998.
IX. APPENDIX A
A-1. Summary of the Joining Protocol API
Broker Conversation SuperBroker
o^~\ REGISTER o
O `" )) TELL/SORRY ec
--Domn untoogi es
C rmmunty Ontologls
I I I I -O -- ,
Class Method FSM State Information Flow
A-2. Summary of the Knowledge Discovery Protocol API
Class Attribute Method Information Flow
A-3. Summary of the Capability Advertisement Protocol API
Attribute Method Information Flow
A-4. Summary of the Departure Protocol API