2364
A Survey on Neural Networks in Automated Negotiations
tal learning of a feedforward neural network in
RUGHUWRLQFUHDVHWKHHI¿FLHQF\RIELODWHUDOQHJR-
tiations and to improve the applicability towards
multilateral negotiations. The network is triggered
with values that are extracted after a utility evalu-
ation procedure and at each round the output is
forming the next counter-offer of the party. With
regards to the generalization to the multilateral
case, the proposed approach is based on match-
ing all sellers and all buyers in pairs among all
possible ones, following practical criteria as the
common negotiation range term used, indicates.
The experimental results show that the proposed
system achieves up to 2% more agreements and
carries out the negotiations at least twice as fast
as others with similar settings.
In (Wang, Chai, & Huang, 2005), the authors
attempt to solve the problem of selecting a selling
agent that meets buyer user’s requirements as well
as his utility constraints as those represented by
the corresponding intelligent agent. The problem
is solved by choosing the seller before the negotia-
tion and thus, the accuracy of the negotiation and
the buyer’s utility are improved. In order to fully
utilize negotiation history, this paper transforms
the problem of choosing seller into a K-armed
bandit problem. The utility function is a joint
summation of the utilities of both the buyers and
the sellers, while the buyer uses a properly learned

neural network in order to learn its opponents’
SUHIHUHQFHVDQG¿QDOO\FKRRVHWKHRQHWKDWZLOO
lead to the best agreement. The advantage of this
framework is that the buyer’s neural network
learns off-line and only uses the results for the
online procedure. Thus, there is not substantial
impact on the real procedure.
Finally, in (Liu, & You, 2003), a fuzzy neural
network is proposed to deal with the uncertain-
ties in real world shopping activities, such as
FRQVXPHU SUHIHUHQFHV SURGXFW VSHFL¿FDWLRQ
product selection, price negotiation, purchase,
delivery, after-sales service and evaluation. The
fuzzy neural network manages to achieve an
DXWRPDWLFDQGDXWRQRPRXVSURGXFWFODVVL¿FDWLRQ
and selection scheme to support fuzzy decision-
making by integrating fuzzy logic technology and
the back-propagation feedforward neural network.
In addition, a visual data model is introduced
to overcome the limitations of the current web
EURZVHUV WKDW ODFN ÀH[LELOLW\ IRU FXVWRPHUV WR
view products from different perspectives. The
experimental results demonstrate the feasibility
of the proposed approach for web-based business
transactions.
CONCLUSION AND DISCUSSION
In this paper, a brief survey of the most popular
UHVHDUFKHIIRUWVLQWKH¿HOGRI11DVVLVWHGDXWR-
mated negotiations is presented. An important
observation that can easily be made is that that

there is a substantial diversity on the purposes
that the NNs are used for in this domain. For
instance, in some cases they aim to estimate the
opponent’s future offers, whereas in other cases
they assist the negotiating agent on selecting the
best tactic that should be used in order to increase
its potential utility. Even though the usage of NNs
in automated negotiations may enhance various
aspects of their performance and results, there
are some cases where they are not suitable. For
example, they perform far better when they are
trained off-line, thus being less suitable when
no a-priori knowledge is available. In general,
it is preferable that relatively small NNs that are
trained off-line are used, but if this is not possible,
it is better to use NNs of minimal size that are
trained on-line, risking however that they will
eventually not be suitable enough. Furthermore,
if the negotiation strategy of the opponent is not
consistent, thus frequently demonstrating sharp
FKDQJHVLQWKHW\SHRUFRQ¿JXUDWLRQRIWKHWDFWLF
used, the NNs often fail to adjust. In case the op-
ponent employs imitative negotiation strategies,
the usability of NNs in estimating the opponent’s
behaviour is questionable. Finally, if the agent
has low storage and processing resources avail-
2365
A Survey on Neural Networks in Automated Negotiations
able, the NNs that can be employed need to be so
OLJKWZHLJKWWKDWWKH\FRQVLGHUDEO\ODFNÀH[LELOLW\

Despite these shortcomings, it is expected that
NNs will gain a considerable share in the learn-
ing-enabled negotiating agents in the electronic
marketplace.
REFERENCES
Abreu, M., Canuto, A., & Santana, L. (2005). A
Comparative Analysis of Negotiation Methods
for a Multi-neural Agent System. 5
th
International
Conference on Hybrid Intelligent Systems (HIS
2005), Rio de Janeiro, Brazil.
Carbonneau, R., Kersten, G., & Vahidov, R.
(2006). Predicting Opponent’s Moves in Elec-
tronic Negotiations Using Neural Networks.
International Conference of Group Decision and
Negotiation (GDN 2006), Karlsruhe, Germany.
Faratin, P., Sierra, C., & Jennings, N. (1998).
Negotiation Decision Functions for Autonomous
Agents. International Journal of Robotics and
Autonomous Systems. (24)3-4, 159-182.
Haykin, S. (1999). Neural Networks: A Compre-
hensive Foundation (2
nd
edition). London UK:
Prentice Hall.
Jennings, N., Faratin, P., Lomuscio, A., Parsons, S.,
Sierra, C., & Wooldridge, M. (2001). Automated
Negotiation: Prospects, Methods, and Challenges.
International Journal of Group Decision and

Negotiation. (10)2, 199-215.
Liu, J., & You, J. (2003). Smart Shopper: An
Agent-Based Web-Mining Approach to Internet
Shopping. IEEE Transactions on Fuzzy Systems.
(11)2, 226-237.
Oprea, M. (2001). Adaptability and Embodi-
ment in Agent-Based Ecommerce Negotiation.
Workshop Adaptability and Embodiment Using
Multi-Agent Systems (AEMAS 2001), Prague,
Czech Republic.
Oprea, M. (2003). The Use of Adaptive Negotia-
tion in Agent-Mediated Electronic Commerce.
/HFWXUH1RWHVRQ$UWL¿FLDO,QWHOOLJHQFH (LNAI).
Springer-Verlag, Berlin Heidelberg New York.
2691, 594-605.
Papaioannou, I., Roussaki, I., & Anagnostou, M.
(2006). Comparing the Performance of MLP and
RBF Neural Networks Employed by Negotiating
Intelligent Agents. IEEE/WIC/ACM International
Conference on Intelligent Agent Technology (IAT
2006), Hong Kong, China.
Papaioannou, I., Roussaki, I., & Anagnostou, M.
(2007). Comparing Polynomial Approximators to
Neural Networks for Agent Behaviour Prediction
in e-Negotiations, submitted for publication to the
ACM Transactions of Autonomous and Adaptive
Systems.
Park, S., & Yang, S. (2006). An Automated System
based on Incremental Learning with Applicability
Toward Multilateral Negotiations. International

Joint Conference SICE-ICASE, Busan, Korea.
Rau, H., Tsai, M., Chen, C., & Shiang, W. (2006).
Learning-based automated negotiation between
shipper and forwarder. Journal of Computers and
Industrial Engineering, (51)3, 464-481.
Roussaki, I., Papaioannou, I., & Anagnostou, M.
(2006). Employing Neural Networks to Assist
Negotiating Intelligent Agents. 2
nd
IEE Interna-
tional Conference on Intelligent Environments
2006 (IE 2006), Athens, Greece.
Roussaki, I., Papaioannou, I., & Anagnostou,
M. (2007). Building Automated Negotiation
Strategies Enhanced by MLP and GR Neural
Networks for Opponent Agent Behaviour Progno-
sis. Lecture Notes of Computer Science (LNCS).
Springer-Verlag, Berlin Heidelberg New York.
4507, 152-161.
Shibata, K., & Ito, K. (1999). Emergence of
Communication for Negotiation By a Recurrent
Neural Network. 4
th
International Symposium
2366
A Survey on Neural Networks in Automated Negotiations
on Autonomous Decentralized Systems, Tokyo,
Japan.
Veit, D., & Czernohous, C. (2003). Automated Bid-
ding Strategy Adaptation using Learning Agents

in Many-to-Many e-Markets. 2
nd
International
Joint Conference on Autonomous Agents and
Multi-Agent Systems (A A M A S 2 0 03) , M e l b o u r n e ,
Australia.
Wang, L.M., Chai, Y.M., & Huang, H.K. (2005).
Choosing optimal seller based on off-line learning
negotiation history and k-armed bandit problem.
International Conference on Machine Learning
and Cybernetics (ICMLC 2005), Guangzhou,
China.
Zeng, Z.M., Meng, B., & Zeng, Y.Y. (2005).
An Adaptive Learning Method in Automated
Negotiation. International Conference on Ma-
chine Learning and Cybernetics (ICMLC 2005),
Guangzhou, China.
Zhang, S., Ye, S., Makedon, F., & Ford, J. (2004).
A Hybrid Negotiation Strategy Mechanism in an
Automated Negotiation System. 5
th
ACM Confer-
ence on Electronic Commerce (EC 2004), New
York, USA.
KEY TERMS
Automated Negotiation: It is the process by
which group of actors communicate with one
another aiming to reach to a mutually acceptable
agreement on some matter, where at least one of
the actors is an autonomous software agent.

Bilateral Negotiation: A negotiation proce-
dure, where exactly two parties are involved, i.e.
a client and a provider.
Multilateral Negotiation: A negotiation pro-
cedure, where more than two parties are involved,
i.e. multiple clients and/or providers negotiate
simultaneously.
Multi-Layer Perceptron (MLP): A fully
connected feedforward NN with at least one hid-
den layer that is trained using back-propagation
algorithmic techniques.
Neural Network (NN): A network modelled
after the neurons in a biological nervous system
with multiple synapses and layers. It is designed
as an interconnected system of processing ele-
ments organized in a layered parallel architecture.
These elements are called neurons and have a
limited number of inputs and outputs. NNs can
EHWUDLQHGWR¿QGQRQOLQHDUUHODWLRQVKLSVLQGDWD
HQDEOLQJVSHFL¿FLQSXWVHWVWROHDGWRJLYHQWDUJHW
outputs.
Radial Basis Function (RBF): Function that
involves a distance criterion with respect to a
centre, such as a circle, ellipse or Gaussian.
RBF NN:,WLVDQDUWL¿FLDO11WKHDFWLYDWLRQ
functions of which are radial basis functions.
,WKDVWZROD\HUVRISURFHVVLQJZKHUHWKH¿UVW
maps the input onto each RBF neuron in the other
(hidden) layer.
7KLVZRUNZDVSUHYLRXVO\SXEOLVKHGLQWKH(QF\FORSHGLDRI$UWL¿FLDO,QWHOOLJHQFHHGLWHGE\-'RSLFR-GHOD&DOOHDQG$

Sierra, pp. 1524-1529, copyright 2009 by Information Science Reference (an imprint of IGI Global).
2367
Copyright © 2009, IGI Global, distributing in print or electronic forms without written permission of IGI Global is prohibited.
Chapter 8.3
Patterns for Designing
Agent-Based E-Business
Systems
Michael Weiss
Carleton University, Canada
ABSTRACT
Agents are rapidly emerging as a new paradigm
for developing software applications. They are
being used in an increasing variety of applica-
tions, ranging from relatively small systems
such as assistants to large, open, mission-criti-
cal systems like electronic marketplaces. One
of the most promising areas of applications for
agent technology is e-business. In this chapter,
we describe a group of architectural patterns for
agent-based e-business systems. These patterns
r e l a t e t o f r o n t - e n d e - b u s i n e s s a c t i v i t i e s t h a t i n vo l v e
interaction with the user, and delegation of user
tasks to agents. Patterns capture well-proven,
common solutions, and guide developers through
the process of designing systems. This chapter
should be of interest to designers of e-business
systems using agent technology. The description
of the patterns is followed by the case study of
an online auction system to which the patterns
have been applied.

INTRODUCTION
Agents are rapidly emerging as a new paradigm
for developing software applications. They are be-
ing used in an increasing variety of applications,
ranging from relatively small systems such as
assistants to large, open, mission-critical systems
like electronic marketplaces. One of the most
promising areas of applications for agent tech-
nology is e-business (Papazoglou, 2001). In this
chapter, we describe a group of architectural pat-
terns for agent-based e-business systems. These
patterns relate to front-end e-business activities
that involve interaction with the user, and delega-
tion of user tasks to agents.
The chapter is structured as follows. First,
we provide a background on patterns and their
application to the design of agent systems. Then,
we discuss the forces or design constraints that
need to be considered during the design of agents
for e-business systems. This is followed by a de-
scription of the agent patterns for e-business. A
2368
Patterns for Designing Agent-Based E-Business Systems
number of examples illustrate the application of
these patterns. Finally, we discuss current trends
and opportunities for future research and offer
concluding remarks.
BACKGROUND
Patterns are reusable solutions to recurring design
problems and provide a vocabulary for com-

municating these solutions to others. The docu-
mentation of a pattern goes beyond documenting
a problem and its solution. It also describes the
forces or design constraints that give rise to the
proposed solution (Alexander, 1979). These are
the undocumented and generally misunderstood
features of a design. Forces can be thought of as
pushing or pulling the problem towards different
solutions. A good pattern balances these forces. A
set of patterns, where one pattern leads to other
SDWWHUQVWKDWUH¿QHRUDUHXVHGE\LWLVNQRZQDV
a pattern language. A pattern language can be
likened to a process: it guides designers who wants
to use those patterns through their application in
an organic manner. As each pattern of the pattern
language is applied, some of the forces affecting
the design will be resolved, while new unresolved
forces will arise as a consequence. The process of
using a pattern language in a design is complete
when all forces have been resolved.
There is by now a growing literature on using
patterns to capture common design practices for
agent systems. Aridor and Lange (1998) describe
domain-independent patterns for the design of
mobile agent systems. They classify mobile agent
patterns into traveling, task, and interaction pat-
terns. Kendall, Murali Krishna, Pathak, et al.
(1998) use patterns to capture common build-
ing blocks for the architecture of agents. They
integrate these patterns into the layered agent

pattern, which serves as a starting point for a
pattern language for agent systems based on the
strong notion of agency. Schelfthout, Coninx,
et al. (2002), on the other hand, document agent
implementation patterns suitable for developing
weak agents.
Deugo, Weiss, and Kendall (2001) identify
a set of patterns for agent coordination, which
are, again, domain-independent. They classify
agent patterns into architectural, communication,
traveling, and coordination patterns. They also
describe an initial set of global forces that push
and pull solutions for coordination. Kolp, Gior-
gini, and Mylopoulos (2001) document domain-
independent organizational styles for multi-agent
systems using the Tropos methodology. Weiss
(2004) motivates the use of agents through a set
of patterns that document the forces involved in
agent-based design and key agent concepts.
On the other hand, Kendall (1999) reports on
ZRUNRQDGRPDLQVSHFL¿FSDWWHUQFDWDORJGHYHO-
oped at BT Exact. Several of these patterns are
documented using role models in a description
of the ZEUS agent building kit (Collis & Ndumu,
1999). Shu and Norrie (1999) and the author in a
precursor to this chapter have also documented
GRPDLQVSHFL¿FSDWWHUQVUHVSHFWLYHO\IRUDJHQW
based manufacturing and electronic commerce.
However, unlike most other authors, they present
the patterns in the form of a pattern language.

This means that the relationships between the
patterns are made explicit in such a way that
they guide a developer through the process of
designing a system.
Lind (2002) and Mouratidis, Weiss, and
*LRUJLQLVXJJHVWWKDWZHFDQEHQH¿WIURP
integrating patterns with a development process,
while Tahara, Oshuga, and Hiniden (1999) and
Weiss (2003) propose pattern-driven development
processes. Lind (2002) suggests a view-based
categorization scheme for patterns based on the
MASSIVE methodology. Mouratidis et al. (2006)
document a pattern language for secure agent
systems that uses the modeling concepts of the
Tropos methodology. Tahara et al. (1999) propose
a development method based on agent patterns and
distinguish between macro and micro architecture
patterns. Weiss (2003) documents a process for
mining and applying agent patterns.
2369
Patterns for Designing Agent-Based E-Business Systems
FORCES
The design of agent-based systems in the e-
business domain is driven by a number forces,
including autonomy, the need to interact, infor-
mation overload, multiple interface, ensuring
quality, adaptability, privacy concerns, search
costs, and the need to track identity. Not all of
WKHVHIRUFHVFDQEHHTXDOO\VDWLV¿HGE\DJLYHQ
design, and trade-offs need to be made. The pat-

terns described in this chapter help with making
informed trade-offs.
AUTONOMY
The currently dominant metaphor for interacting
with computers is direct manipulation. Direct
manipulation requires the user to initiate all tasks
explicitly and to monitor all events. For example,
a user searches the Web for an auction that of-
fers the desired item for sale, and subsequently
monitors the state of the auction. The obvious
drawback of this approach is that most of the time
the user is occupied in tasks that are peripheral to
WKHSULPDU\REMHFWLYHV7KHXVHU¶VDELOLW\WR¿QG
the best deal available at any of the many online
auctions in operation is also greatly limited.
Agents can be used to implement a comple-
mentary interaction style, in which users delegate
some of their tasks to software agents which then
perform them autonomously on their behalf. This
indirect manipulation style engages the user in
a cooperative process in which human and soft-
ware agents both initiate communication, moni-
tor events, and perform tasks. Autonomy is the
capability of an agent to follow its goals without
i n t e r a c t i o n s o r c o m m a n d s f r o m t h e u s e r o r a n o t h e r
agent. An autonomous agent does not require the
user’s approval at every step of executing its task,
but is able to act on its own.
With agents performing autononmous actions,
users are now facing issues of trust and control over

their agents. The issue of trust is that by engag-
ing an agent to perform tasks (such as selecting
a seller), the user must be able to trust the agent
to do so in an informed and unbiased manner.
The agent should not, for example, have entered
contracts with sellers to favor them in return for a
cut on their proceeds to the developer of the agent
or the server that hosts and executes the agent.
The user would also like to specify the degree of
autonomy of the agent. For example, the user may
not want to delegate decisions to the agent that have
OHJDORU¿QDQFLDOFRQVHTXHQFHVDOWKRXJKDEX\HU
DJHQWLVFDSDEOHRIQRWRQO\¿QGLQJWKHFKHDSHVW
seller, but also placing a purchase order.
NEED TO INTERACT
Agents typically only have a partial representa-
tion of their environment, and are thus limited in
their ability—in terms of their expertise, access
to resources, location, and so forth—to interact
with it. Thus, they rely on other agents to achieve
goals that are outside their scope or reach. They
also need to coordinate their activities with those
of other agents to ensure that their goals can be
met, avoiding interference with one another. The
behavior of an individual agent is thus often not
comprehensible outside its social structure—its
relationships with other agents. For example, the
behavior of a buyer agent in an auction cannot be
fully explained outside the context of the auction
itself, and of the conventions that govern it (for

example, in which order—ascending or descend-
ing—bids must be made, and how many rounds
of bidding there are in the auction).
An important issue in designing systems
of interacting agents is dealing with openness.
The Internet and e-business applications over
the Internet are both examples of open systems.
Open systems pose unique challenges in that their
components are not known in advance; they can
change unexpectedly, and they are composed of
heterogeneous agents implemented by different
developers at different times with different tools
2370
Patterns for Designing Agent-Based E-Business Systems
a n d m e t h o d o l og ie s . S i m i l a r l y, a s w e d o n o t c o nt r o l
all the agents, one can also not assume that the
agents are cooperative. Some agents may be be-
nevolent and agree on some protocol of interaction,
but others will be self-interested and follow their
own best interests. For example, in an electronic
marketplace, buyer and seller agents are pursuing
WKHLURZQEHVWLQWHUHVWVPDNLQJSUR¿WDQGQHHG
to be constrained by conventions.
INFORMATION OVERLOAD
3HRSOHDQGRUJDQL]DWLRQVZLVKWR¿QGUHOHYDQW
information and offerings to make good deals and
JHQHUDWHSUR¿W+RZHYHUWKHODUJHVHWRIVHOOHUV
in conjunction with the multiple interfaces they
XVHPDNHVLWGLI¿FXOWIRUDKXPDQWRRYHUYLHZ
the market. One solution has been to provide

portals or c om m o n e n t r y p o i nt s t o t h e We b . T h e s e
portals periodically collect information from a
multitude of information sources and condense
WKHPWRDIRUPDWWKDWXVHUV¿QGHDVLHUWRSURFHVV
typically taking the form of a hierarchical index.
One disadvantage of this solution is that the cat-
egories of the index will be the same for every
user. Individual preferences are not taken into
account when compiling the information, and
niche interests may not be represented at all.
MULTIPLE INTERFACES
2 Q H RIW KHG L I ¿F X OW LH VL Q¿ Q G L QJL Q IRU P DW LRQ HJ 
when comparing the offerings of different sellers)
is the large number of different interfaces used
to present the information. Not only are store
fronts organized differently, sellers do not fol-
low the same conventions when describing their
products and terms of sale. For instance, some
sellers include the shipping costs in the posted
price; others will advertise one price, but add a
handling charge to each order. A solution is to
agree on common vocabularies, but these must
also be widely adopted. With the introduction
of the extensible markup language (XML) for
associating metacontent with data and current
developments in the Semantic Web such as on-
tology representation languages (OWL), this is
slowly becoming a reality. For example, a price
in a catalog can be marked up with its currency
whether it already includes the shipping cost.

+RZHYHUWKHGLI¿FXOW\ZLWKDQ\VWDQGDUGIRUPDW
is that it takes a considerable amount of time to
¿QG DJUHHPHQW DPRQJ WKH LQWHUHVWHG SDUWLHV
One also needs to allay the fear of sellers in los-
ing business to competitors once their product
information becomes easily accessible.
ENSURING QUALITY
Shopping online lacks the immediate mechanisms
for establishing trustworthiness. How can you
trust a seller, with whom you have had no previous
encounter, whether the order you placed will be
IXO¿OOHGVDWLVIDFWRULO\")RUH[DPSOHDQ\VHOOHULQ
an online auction could claim that the item offered
for sale is in superior condition, when the buyer
cannot physically verify that claim. One solution
is to solicit feedback about the performance of a
seller (respectively, buyer) from buyers (respec-
W L YHO \ V H O OH U V D I W H U R U G H U I X O ¿ OO P H Q W  )R U H[D P SOH  
the online auction site eBay keeps records of how
a seller was rated by other buyers. Potential new
buyers will take the ratings from previous buy-
ers into account before considering buying from
a seller. However, eBay’s solution falls short in
two ways. Old low ratings are not discarded or
discounted when more recent ratings are higher.
Also, if a seller gets a low overall rating, it is easy
for the seller to assume a new identity and start
afresh with a new rating. A mechanism for ensur-
ing quality must avoid this, as discussed in the
context of reputation system design by Zacharia,

Moukas, and Maes (1999).
2371
Patterns for Designing Agent-Based E-Business Systems
ADAPTABILITY
Users differ in their status, level of expertise,
needs, and preferences. The issue of adaptability is
that of tailoring information to the features of the
user, for example, by selecting the products most
suitable for the user from a catalog, or adapting
the presentation style during the interaction with
the user. Any approach to tailoring information
involves creating and maintaining a user model.
When creating a user model, two cases need to be
distinguished (Ardissono, Barbero, et al., 1999):
IRU¿UVWWLPHYLVLWRUVQRLQIRUPDWLRQDERXWWKHP
is available, and the user characteristics must be
recognized during the interaction; on subsequent
visits, a detailed user model about a visitor is
already available and can be used to tailor the
information.
Several design considerations for user model-
ing are detailed in Ardissono et al. (1999). Users
need to register permanently with the system to
have their data stored; otherwise user models
will only be maintained during a single interac-
tion. In the context of online shopping, a system
must also deal with direct and indirect users. A
customer of a Web store may browse for products
for himself, as well as for other people (indirect
users), who have different needs and preferences.

)LQDOO\ WKH XVHU PRGHOPXVW EH DEOHWR UHÀHFW
changes in interest over time. One approach to
collecting user information is to ask the user to
provide the information explicitly, for example,
E\¿OOLQJRXWDIRUP7KLVDOORZVRQHWRFUHDWHD
SUR¿OHRIWKHXVHUWKDWLVSRWHQWLDOO\YHU\DFFXUDWH
and to provide personalized service to the user
from the beginning. However, there are at least
two problems with this solution. First, by requir-
ing the user to provide this information upfront;
the threshold for the user to do so is very high.
Only very advanced users will want to tune their
RZQSUR¿OHV6HFRQGZKHQWKHXVHU¶VLQWHUHVWV
FKDQJHWKLVZLOOQRWEHUHÀHFWHGLQWKHSUR¿OH
XQOHVVWKHXVHUNHHSVXSGDWLQJKHUSUR¿OH$JDLQ
L QSUDFWLFHXVHU VGRQ RWXSGD WHWKHLU SU R¿OHVDIWH U
installation.
PRIVACY CONCERNS
Personalization requires the collection and rele-
sase of personal information to the agent providing
the personalized service. One way of personal-
izing interactions between buyers and sellers is
for the seller to collect information about a buyer
from the buyer’s behavior (e.g., their clickstream).
The buyer may not be aware of the information
collected, nor does she always have control over
what information is gathered about her. Although
effective from the seller’s perspective, this is
not a desirable situation from the perspective
of the buyer. Users are typically not willing to

allow just anyone to examine their preferences
and usage patterns, in particular without their
knowledge or consent. They want to remain in
control, and decide on an interaction-by-inter-
action basis which information is conveyed to
the seller. A solution that addresses the force of
privacy concerns must put the user in charge of
which information is collected and who it is made
available to. An additional complexity results from
the desire of some buyers to remain anonymous.
If a buyer remains anonymous, a seller cannot
provide personalized service. Thus, generally,
users are willing to share personal information
with sellers, if the expected gains outweigh the
possible threats for their privacy.
SEARCH COSTS
,WFDQEHH[SHQVLYHIRUEX\HUVDQGVHOOHUVWR¿QG
each other. In a static marketplace, each buyer can
store a contact list of sellers for each product, and
then quickly locate an appropriate seller when a
particular product is needed. However, an elec-
tronic marketplace is dynamic. Buyers and sellers
can join and leave the marketplace, and change
2372
Patterns for Designing Agent-Based E-Business Systems
their requirements and offerings qualitatively and
quantitatively at any point in time. It, therefore,
becomes impossible for a market participant to
maintain an up-do-date list of contacts. Another
problem is that of restricting the buyer’s options.

If each buyer maintains its own list of contacts,
they run the risk of not being aware of better
deals available elsewhere. One possible solution
to these problems is to use a mediator which can
match potential trading partners in the market.
With the introduction of mediators, buyers and
sellers no longer maintain their own lists of
contacts, or need to contact a large number of
DOWHUQDWLYHWUDGLQJSDUWQHUVWR¿QGWKHRSWLPDO
one. One trade-off of this solution is, however,
that individual preferences or history of interac-
tion with a particular trading partner cannot be
accounted for by a mediator. Thus, it is reasonable
to maintain individual lists of trading partners that
one has dealt with in the past, keeping track of
the quality provided and using this personalized
UDQNLQJRISDUWQHUVWR¿OWHUWKHOLVWRIFRQWDFWV
provided by a mediator.
IDENTITY
For various reasons, buyers and sellers need to
be represented by unique identities. The most
important reasons are authentication, nonre-
pudiation, and tracking. One way of assigning
a unique identity to trading partners is to use
one of the many unique labels which are readily
available on the Internet, for example, an e-mail
address, or a Yahoo! account name. A problem
with this approach is that it is also very easy to
obtain a new identity, thus making authentication,
nonrepudiation, or tracking schemes that rely on

such identities impractical. Similarly, a user could
obtain multiple identities and pretend to represent
multiple different parties, where instead there is
only one. A solution that remedies this situation
m u s t m a ke i t a dv a nt a g e o u s f o r i n d iv i d u a l s t o k e e p
their identities over those users who change them
often (Zacharia et al., 1999).
PATTERNS
7KHSDWWHUQVZHLGHQWL¿HGDQGWKHLUUHODWLRQVKLSV
DUHVKRZQLQ)LJXUH7KHDUURZVLQGLFDWHUH¿QH-
ment links between the patterns. Each arrow in
WKHGLDJUDPSRLQWVLQWKHGLUHFWLRQIURPD³ODUJHU´
SDWWHUQWRD³VPDOOHU´SDWWHUQ7KHVWDUWLQJSRLQW
for the language is the
AGENT SOCIETY pattern,
which motivates the use of agents for building
WKHDSSOLFDWLRQ$WWKHQH[WOHYHORIUH¿QHPHQW
the diagram leads the designer to consider the
patterns agent as
DELEGATE, AGENT AS MEDIATOR,
AND COMMON VOCABULARY.
1
agent as delegate and the patterns it links to
deal with the design of agents that act on behalf
of a single user. The agent as mediator pattern
guides the designer through the design of agents
that facilitate between a group of agents and their
users.
COMMON VOCABULARY provides guidelines
IRUGH¿QLQJH[FKDQJHIRUPDWVEHWZHHQDJHQWV

7KH UHVW RI )LJXUH  VKRZV UH¿QHPHQWV RIWKH
AGENT AS DELEGATE pattern. For example, the USER
AGENT
pattern prescribes to use a single locus of
interaction with the user and represent the concur-
rent transactions a user participates in as buyer
and seller agents. User interaction also includes
SUR¿OLQJWKHXVHU
USER PROFILING) and subscrib-
ing to information (e.g., the status of an auction)
relevant to the user (
NOTIFICATION).
In the following, each pattern is represented by
its context, the problem it addresses, a discussion
of the forces, its solution, and a resulting context.
The context is represented by the dependencies
between the patterns. The problem is a succinct
statement on what problem the pattern addresses.
The solution takes the form of a role diagram.
7KHVHUROHVZLOOEH¿OOHGE\DJHQWV)RUH[DPSOH
consider the
USER AGENT pattern. It is applied after
AGENT AS DELEGATEDQGLQWXUQUH¿QHGE\USER
PROFILING
and NOTIFICATION. The it addresses the
2373
Patterns for Designing Agent-Based E-Business Systems
problem of how users instruct agents to act on
their behalf (as buyers and sellers) and how they
k e e p i n c o n t r o l o ve r w h a t t h e a g e n t d o e s (e . g ., d o e s

it have authority to complete a trade?). The role
diagram for the
USER AGENT pattern is shown in
Figure 2. Role diagrams and their semantics are
discussed further in
AGENT SOCIETY. The resulting
context points to related patterns in this pattern
language.
AGENT SOCIETY
Context
<RXUDSSOLFDWLRQGRPDLQVDWLV¿HVDWOHDVWRQHRI
the following criteria: your domain data, control,
knowledge, or resources are decentralized; your
application can be naturally thought of as a sys-
tem of autonomous cooperating entities, or you
have legacy components that must be made to
interoperate with new applications.
Problem
How do you model systems of autonomous co-
operating entities in software?
Forces
• Autonomy
• Need to interact
Solution
Model your application as a society of agents.
Agents are autonomous computational entities
(autonomy), which interact with their environment
(reactivity) and other agents (social ability) in order
to achieve their own goals (proactiveness). Often,
agents will be able to adapt to their environment

and have some degree of intelligence, although
these are not considered mandatory characteris-
tics. These computational entities act on behalf
of users or groups of users (Maes, 1994). Thus,
Figure 1. Patterns for e-business agents and their dependencies (arrows indicate dependencies and
dashed lines patterns that are not described here)