B2C AND C2C E-MARKETPLACES

A Multi-layer/Multi-agent Architecture to Support them

R. Miguel, J. J. Castro-Schez, D. Vallejo, C. Glez-Morcillo and V. Herrera

Department of Information Technologies and Systems

University of Castilla-La Mancha, Ciudad Real, 13071, Spain

Keywords:

e-Commerce, B2C portals, C2C portals, e-Marketplaces, Multi-agent systems, Multi-layer architecture, Soft-

ware agents, Middleware, Auctions, Shopping.

Abstract:

Due to the growing of the Internet and the users’ trend to acquire products via Internet, the e-commerce has

become in a new very important business opportunity. In this paper we present a multi-agent and multi-layer

architecture to support general purpose e-marketplaces. The presented system is compound of software agents

that can be able to act in the users’ beneﬁt reducing their search time and facilitating the product acquisition.

These agents are grouped in different layers giving support to B2C and C2C e-marketplaces. The interaction

between the different agents is explained through the paper although we remarks the agents used to search,

acquire and recommend products. We have also used the fuzzy logic in these agents because we believe that

is very useful in order to facilitate its use and reduce the search and acquisition time.

1 INTRODUCTION

Nowadays the importance of B2C and C2C e-

commerce is undeniable. Thousands of acquisitions

take place over the Internet every day (Turban et al.,

2008). This fact leads ﬁrms to a new market that is

growing at this moment. An e-marketplace can be

deﬁned as a site of electronic exchange where ﬁrms

duct business over the Internet. E-marketplaces are

reaching a bit summit as medium that allows to small

companies to commercialize their goods (Wise and

Morrison, 2000). Some examples of this trend are

E-Bay (eBay Inc., 2010), Todocoleccion (Zoconet,

2010), etc., which generate lots of transactions ev-

ery day. This work is focused on B2C and C2C e-

Marketplaces that include on-line transactions of con-

sumer goods or services based on ﬂexible price sys-

tems, such as auctions or ﬁxed price systems, such as

direct sales (Buy it Now).

It is needed to exploit the technology establish-

ing architectures which address the requirements of

this kind of systems in order to achieve the full de-

velopment of these e-marketplaces and their success.

In previous works, a centralized architecture called

e-ZOCO (Miguel and Castro-Schez, 2009) was de-

veloped to give support to an e-marketplace (Lopez-

Lopez et al., 2009a). This architecture presents per-

formance problems when the data layer contains lots

of data and when lots of users are concurrently con-

nected. This kind of centralized architectures are not

therefore scalable when such parameters grow up, de-

manding for solutions that rely on distributed control

and data distribution. Furthermore, e-commerce ap-

plications are integrating more and more techniques

based on Artiﬁcial Intelligence to provide the users

with new capabilities and opportunities for business.

Within this context, Multi-Agent Systems (MAS)

(Weiss, 1999) can contribute to overcome these limi-

tations and improve the quality of e-marketplaces.

In order to design the architecture of the e-

Marketplace MAS, the relevant services that deter-

mine the characteristics of this kind of portals and

their functionality have been analyzed, as well as the

different actors that interact within the market.

Furthermore, in the devised design, there exists

an agent based on fuzzy logic whose goal is to as-

sist users in the buying process (Lopez-Lopez et al.,

2009a).

In order to develop this proposal, the set of spec-

iﬁcations of the Foundation for Intelligent Physical

Agents (FIPA) committee has been adopted, which

represents the de facto standard in the ﬁeld of MAS

(http://www.ﬁpa.org/).

An e-marketplace implementation is not trivial.

The e-marketplaces evolution has shown that there

100

Miguel R., J. Castro-Schez J., Vallejo D., Glez-Morcillo C. and Herrera V. (2010).

B2C AND C2C E-MARKETPLACES - A Multi-layer/Multi-agent Architecture to Support them.

In Proceedings of the 12th International Conference on Enterprise Information Systems - Software Agents and Internet Computing, pages 100-105

DOI: 10.5220/0002907801000105

 SciTePress

are a set of features common to the most successful

e-marketplaces today (Bruckner and Kiss, 2004; Li

et al., 2001; Bar, 2001; Hu, 2009; Brunn et al., 2002).

The rest of the paper is organized as follows: The

features which are offered in the most successful e-

marketplaces are presented in Section 2. Section 3

presents a general description of the multi-layer and

multi-agent architecture. In Section 4, an example of

interactions between agents, users and environment is

described. Finally, Section 5 is devoted to the conclu-

sions and future work.

2 GENERAL DESCRIPTION

The main aim of the proposed architecture is to sup-

port a virtual market where both the sellers and buy-

ers depicted as persons or virtual agents can interact

and come together to carry out electronic transactions.

These agents provide a range of common services to

this kind of portals but with the added value of man-

aging from an approach based on intelligent agents.

These basic services encompass aspects like the

management of shops and users within the por-

tal, product catalog, product buying/selling, product

search and recommendation within the portal, mes-

sages between the users, users’ trust, dispute between

users, or reporting useful information. Furthermore,

they are used as a design guide to determine the nec-

essary intelligent agents for the proposed system.

The proposed architecture facilitates the use of

these e-commerce services in a free and open way,

allowing the portability and ensuring the interoper-

ability. These features that draw a framework based

on Service-Oriented Computing (Huhns and Singh,

2005) ﬁts with the MAS notion. The agents com-

municate and cooperate to improve the quality of ser-

vices that compose the e-marketplace.

It is necessary to highlight that a service can be

represented by an agent and that an agent can repre-

sent several services (see Figure 1).

The adoption of FIPA pursues two main objec-

tives: i) to assure the interoperability with other MAS

that follow the guides deﬁned by FIPA, and ii) to pro-

vide a set of basic management services common to

any application developed from this general purpose

system.

The FIPA standard deﬁnes three common basic

services (conceived like agents in the architecture) to

any MAS (FIPA, 2000): i) Agent Management Sys-

tem, ii) Message Transport Service and iii) Directory

Facilitator. This allows to adopt an homogeneous

approach in which all the platform components are

agents providing relevant services for the proposed

Figure 1: Overview of the Multi-Agent Architecture.

system.

Figure 1 shows the different layers that compose

the e-commerce architecture and how the communi-

cations between them is carried out. Next, each one

of these layers is described.

2.1 User Layer

The user layer represents the interface between the e-

marketplace users and the set of provided services.

Such interface is accessible from a web browser.

There are three different user roles: i) the Admin-

istrator, responsible for managing the e-marketplace;

ii) the Buyer, representing the users interested in buy-

ing products (via direct shopping or auctions); and iii)

the Seller, representing the users interested in selling

products to the buyers of the e-marketplace (via direct

marketing or negotiation). A user can assume multi-

ple roles simultaneously.

When the user registers to the system as a buyer or

a seller, he/she can delegate on software agents cer-

tain operations, allowing them to act in their bene-

ﬁt. The inclusion of this kind of intelligent agents in

e-marketplace is within the semantic web (Hendler,

2004). The intelligent agents which are described be-

low have the ability to interpret the market informa-

tion, satisfy the user requests and generate a beneﬁt

for the user who represents. This allows to reduce the

human interaction and automate several tasks related

to the portal management.

2.2 Multi-agent Layer

In this section the multi-agent layer that supports the

e-marketplace services is described. We will focus on

B2C AND C2C E-MARKETPLACES - A Multi-layer/Multi-agent Architecture to Support them

101

the functionality provided by each agent and the in-

teractions between them. To facilitate the understand-

ing of the proposal, the subsystems that compose the

multi-agent layer are discussed (see Figure 2).

2.2.1 Basic Functionality

The Basic Functionality Agent carries out several

simple services and does not require intelligence be-

cause it implies reactive behavior.

This agent comes into operation simultaneously

with the e-marketplace by performing two manage-

ment operations within the multi-agent architecture

(shared with the others agents of the system): i) regis-

ter to the FIPA agent platform through the Agent Man-

agement System to obtain a valid and unique identi-

ﬁer, and ii) Register to the Directory Facilitator in

order to facilitate its location from a functional de-

scription. To do that, a description according to the

ontology deﬁned by FIPA must be used (FIPA, 2000).

This agent provides the following services:

a) User Management

The user management implies two tasks mainly: i) to

assure the correct subscription, modiﬁcation and dele-

tion of users within the portal; and ii) to manage the

delegation of the users’ tasks to the intelligent agents

that act in their beneﬁt.

b) Catalog Management

The catalog management involves the proper admin-

istration of such catalog by the portal administrator.

This component is considered like a set of classes or

categories C = {object, c

, c

, . . . , c

} hierarchically

organized through an is a relationship. Such relation-

ship is transitive, fact which ensures that ∀ c

, c

is an

object. The class object is the root class.

When a seller puts up a product for sale, it must

belong to some class c

. Hence, each class comprises

a set of objects c

= {e

, . . . , e

}. Each e

belong-

ing to a class c

is described by means of a set of

attributes or variables V

= {v

, . . . , v

}, which val-

ues serve to distinguish between different products

belonging to the same class. Furthermore, each vari-

able v

has a range of deﬁnition, denoted as RDV

, x

, . . . , x

}, which speciﬁes the possible values

that the variable v

may take. The range of def-

inition of class c

, denoted as RDV

, is deﬁned as

RDV

= {RDV

, . . . , RDV

}. The variables that de-

scribe a product class may represent different data

types (Castro-Schez et al., 2004).

c) Product Management

A mechanism to manage the portal products is

needed. Its main function is to allow the users to

posed by the catalog deﬁnition according to the data

types of the category variables. Thus, the consis-

tency and uniformity of e-commerce portal catalog is

guaranteed. This functionality maintains a reactive

scheme that allows to check the user data when there

are interactions with the portal.

d) Dispute Management

The dispute management is reduced to an action per-

formed by the user. This action is to describe the

dispute and register it via the communications kernel

into the data layer. Another relevant task is to monitor

each dispute during its life cycle when necessary and

as established in each case.

e) Trust Management

One of the main problems for users in e-marketplaces

is reliability since the user does not often know the

other party when carrying out transactions. To over-

come this limitation, most of the portals have a trust

system that allows to associate a “prestige” to each

user within the community or the portal. This “pres-

tige” depends on aspects such as efﬁciency in the buy-

ing process or the product description ﬁdelity given

by the user interested in selling.

The Basic Functionality Agent performs the fol-

lowing tasks: to process the score of each transac-

tion carried out in the portal to determine the user’s

prestige, to provide the user’s prestige to other users

or agents when they request this information, and to

control the concurrentaccess of differentusers to such

information.

2.2.2 Product Acquisition Subsystem

This subsystem provides all the services related to the

product purchase/sale within the portal. A user (or

an agent) executes it in order to acquire the product.

The acquisition may be through direct purchase, auc-

tion, negotiation, etc. It is composed of the follow-

ing agents: sellers, buyers, auctioneers, and payment

agent.

Since the auction processes are common in e-

marketplaces, it is interesting to equip the system with

an auction agent. Within this context, such an agent

registers request for auction, usually on demand by

seller agents, registers the bids carried out by buyer

agents and controls that the process is normally de-

veloped. Moreover, there is another agent, the Buy-

It-Now Agent, which controls the process of direct

sales.

The seller agent automates some tasks related to

the product sale process: i) to contact with other users

ICEIS 2010 - 12th International Conference on Enterprise Information Systems

102

Figure 2: Overview of the multi-agent layer that supports the e-marketplace.

or agents interested in acquiringproducts accordingto

the sale constraints imposed by the represented user,

ii) to ask the buyer users or agents for direct shop-

ping processes in the appropriate time and in the best

conditions for its owner, and iii) to initiate auction

processes to obtain beneﬁts from the competition be-

tween several buyer users or agents interested in some

product managed by the seller agent.

The buyer agent is responsible for representing the

buyer user in the portal with the aim to obtain beneﬁt.

Its responsibilities are as follows: i) to represent the

buyer user in the product direct shopping processes

within the portal, ii) to send bids to the seller agent

in an auction process, which satisfy the seller’s con-

straints when the best circumstances occur in order to

acquire the product, and iii) to order periodic searches

in the portal, having into account the buyer’s prefer-

ences.

Nowadays, little attention has been paid to

the payment system due the fact of being an e-

marketplace prototype. A payment agent will be in-

cluded in the future. This agent will allow to make

transfers related to the cost of the purchased product.

2.2.3 Product Search Subsystem

The product search subsystem is formed by a set of

search agents. Two main functionalities have been

delegated to them: i) to provide an arranged prod-

uct list which may be of interest for a buyer because

they are similar to the provided ideal product descrip-

tion, and ii) to advise to the portal users in the product

search process by giving information about the set of

more frequently used variables by the portal users in

previous searches.

The search agent owns one or more descriptions

of products or ideal objects (o

) for the buyer (or its

buyer agent). An ideal object speciﬁcation consists in

the set of variables V

′

⊆ V

selected from the whole

set of variables describing the class c

together with

their required values, which are selected from the de-

scribed domains of deﬁnition DDV

′

. Plus, each vari-

able is provided with a weight P(v

′ j

) ∈ [0, 1], which

establishes its importance (Castro-Schez et al., 2005).

From such an ideal description o

, which may be

precise, imprecise or both, the agent provides a sim-

ilar list of products. The imprecision is mainly due

to the ignorance related to the knowledge of the mar-

ket. To deal with vagueness and precise information,

an homogeneous information representation model is

maintained. The common representation formalism

used in this work is based on the use of Fuzzy Sets

(Zadeh, 1965).

The next step is to calculate the similarity between

each product e

∈ c

of the list of similar products

and the ideal object. Basically, this search process

consists in calculating the partial distance between

each one of the variables that compound the ideal ob-

ject and the other products c

. Afterwards, when this

process ends up the global distance is calculated as

the weighted sum of the partial distances depending

on the importance of each attribute that describes the

product. In this way, the search agent can build a list

of ordered products based on the global distance with

the more relevant products for the user/agent.

Finally, this agent can be replicated on different

machines to distribute the workload associated with

its two main functions (search and counseling). The

design solution adopted for carrying out such replica-

tion is to use a master-slave scheme and a synchro-

nization mechanism between replicas. To do this, the

pattern observer (Gamma et al., 1995) has been used.

2.2.4 Product Recommendation Subsystem

The product recommendation subsystem can be un-

derstood as a natural extension to the search subsys-

tem. The goal pursued is to make suggestions or rec-

ommendations to the portal users depending on the

relationships between variables to perform realistic

searches. Furthermore, this subsystem facilitates the

analysis of the information gathered by the search

agent by taking into account the user’s preferences.

In this way, the main responsibility of this subsystem

is to offer a product classiﬁcation service depending

on these preferences or the ﬁtness for the portal users.

B2C AND C2C E-MARKETPLACES - A Multi-layer/Multi-agent Architecture to Support them

103

Figure 3: Deﬁnition domain of the variable ﬁtness.

The agent that gives support to the product rec-

ommendation subsystem is the recommender agent,

which has been designed as an utility-based agent,

that is, it is based on an utility function that allows

to establish the more interesting products for users

(Lopez-Lopez et al., 2009b). In order to make recom-

mendations close to the user, the recommender agent

uses the linguistic variable ﬁtness, which takes three

values: very recommendable, recommendable and not

much interesting (Figure 3). In this way, the agent

is able to associate each element e

to one or two

of these values by means of a fuzzyﬁcation process,

making use of the global distance existing between

the ideal product o

and each e

∈ c

2.2.5 Maintenance Subsystem

Most e-marketplaces provide mechanisms for gener-

ating reports and statistics about the portal use in or-

der to improve and reﬁne them. Within this context,

the maintenance agent is responsible for generating

reports about the following aspects: i) Activity of the

portal and registered users/agents. This private infor-

mation can be internally used to suggest items or of-

fers in an intelligent way. Furthermore, it is possible

to reward the more active users within the portal; ii)

more desired items; iii) more purchased/sold items;

iv) transactions made within the portal (direct sales or

auctions); v) number of registered users; vi) number

of unregistered users; vii) number of inactive users.

2.3 Infrastructure Layer

The infrastructure layer represents both the software

and hardware resources used by the e-marketplace

supported by the multi-agent architecture discussed

in the previous section. When it comes to software,

the infrastructure layer hosts the following compo-

nents: i) a web server that supports the portal web

interface to make easy the interactions between the

portal users, ii) a communications middleware that

allows the communication between the agents of the

multi-agent layer and provides basic services such as

location or implicit activation. Furthermore, it gives

support to secure communication channels to guar-

antee the data integrity, and iii) a relational database

management system to deploy the database of the e-

marketplace.

3 THE COMMUNICATION FLOW

In order to understand the architecture presented here,

we show the communication steps between the agents

since a user registers a product in the e-marketplace

until this is acquired by another user.

Suppose that a seller wants to register a product

in the e-marketplace. Then, he interacts with the

e-marketplace interface that communicates with the

web server. In the web server a service communicates

with the Basic Functionality Agent that informs the

user about the required data to register the product,

checks this data and, ﬁnally, makes it persistent into

the Data Layer.

Now suppose that a buyer wants to acquire a prod-

uct. The ﬁrst event for him is to interact with the e-

marketplace interface. Next, he has three possibili-

ties: i) to buy the product (if possible), ii) to bid for

it (if possible), or iii) to delegate the acquisition of

the product to a software agent. In all the cases, the

web server has to interact with the suitable software

agents. When the acquisition process ﬁnalizes, the ap-

propriate agent contacts with the Payment Agent that

is responsible for the payment. A message is sent to

both the seller and the buyer at this point. This mes-

sage is shown to the suitable user via the interface

of the involved agent, the Buyer Agent or the Seller

Agent.

Then, the general communication mechanism is

as follows: i) to interact with the user interface, ii) to

execute the suitable service in the web server, iii) to

locate the suitable software agent via the middleware,

and iv) to execute the software agent.

4 CONCLUSIONS

The consolidation of the Internet as an everyday tech-

nology has opened a new environment of business

oportunities. Many ﬁrms have expanded their busi-

ness in the past decade to take advantage of the

new information technologies, whereas others have

emerged and are solely based on them. This fact has

caused the appearance of new business paradigms,

along with new technological infrastructures to sup-

port them. This is the case of electronic commerce

(which can be deﬁned as any business that is elec-

tronically transacted).

Third-party (independent) B2C and C2C e-

marketplaces are reaching a bit summit as a medium

ICEIS 2010 - 12th International Conference on Enterprise Information Systems

104

that allows to small companies to do business by

means of Internet. These portals are based on com-

plex technological architectures. One of the main rea-

sons why e-marketplaces fail are related to their tech-

nology that is not mature often.

To address this challenge, a multi-agent architec-

ture has been presented to overcome the limitations of

previously arhitectures developed. The new designed

architecture has the following advantages: i) scalabil-

ity to easily extend the system functionality, ii) it is

efﬁcient in computional terms, since the calculations

are distributed over several machines depending on

the workload, iii) it uses open standards to promote

the interoperability and data exchange among compa-

nies processes and system internal processes, iv) it is

general, being possible to sell or buy goods or ser-

vices of any kind, and v) it has a safety communica-

tion framework to avoid “man in the middle” attacks.

ACKNOWLEDGEMENTS

This work has been funded by the Spanish Min-

istry of Science and Innovation and by the Regional

Government of Castilla-La Mancha under the Re-

search Projects Ref. TIN2009-14538-C02-02 and

Ref. PII2I09-0052-3440.

REFERENCES

Bar, F. (2001). The Construction of Marketplace Architec-

ture, volume 2001, pages 27–49. Tracking a Transfor-

mation: E-commerce and the Terms of Competition in

Industries, the brie-igcc e-conomy project task force

on the internet edition.

Bruckner, L. K. and Kiss, T. (2004). Grid solution for

e-marketplaces integrated with logistics. Distributed

and Parallel Systems.

Brunn, P., Jensen, M., and Skovgaard, J. (2002). E-

marketplaces: crafting a winning strategy. European

Management Journal, 20(3):286–298.

Castro-Schez, J., Jennings, J., Luo, X., and Shadbolt, N.

(2004). Acquiring domain knowledge for negotiating

agents: a case of study. Int. J. of Human-Computer

Studies, 1(61):3–31.

Castro-Schez, J., Jimenez, L., Moreno, J., and Rodriguez,

L. (2005). Using fuzzy repertory table-based tech-

nique for decision support. Decision Support Systems,

3(39):293–307.

eBay Inc. e-bay: http://www.ebay.com/ [accessed on

20/01/2010].

FIPA (2000). Fipa agent speciﬁcation. http://www.ﬁpa.org/

specs/ﬁpa00023/index.html [accessed on

12/01/2010].

Gamma, E., Helm, R., Johnson, R., and Vlissides, J. (1995).

Design patterns: elements of reusable object-oriented

Software. Addison-Wesley Professional, Reading,

MA.

Hendler, J. (2004). Agents and the semantic web. IEEE

Internet Systems, 16(2):30–37.

Hu, B. (2009). Research on third-party b2b e-marketplaces

service orientation from global production networks

embeddedness of smes perspective. 2nd Int. Symp.

Electronic Commerce and Security, pages 385–389.

Huhns, M. and Singh, M. (2005). Service-oriented com-

puting: Key concepts and principles. IEEE Internet

Computing, 9(1):75–81.

Li, C.-S., Chang, Y.-C., and Smith, J. R. (2001). An

e-marketplace infrastructure for model-based match-

making between consumers and providers of multi-

modal earth science data. IEEE Int. Conf. Multimedia

and Expo (ICME’01), page 312.

Lopez-Lopez, L., Miguel, R., Albusac, J., and Castro-

Schez, J. (2009a). Improving searching usability in

direct shopping portals. In 28th North American Fuzzy

Information Processing Society Annual Conference.

Lopez-Lopez, L. M., Castro-Schez, J. J., Vallejo-Fernandez,

D., and Albusac, J. (2009b). System based on machine

learning algorithm for b2c portals. In IEEE/WIC/ACM

Int. Joint Conf. on Web Intelligence and Intelligent

Agent Technology, pages 524–531.

Miguel, R. and Castro-Schez, J. (2009). e-zoco: an open

e-marketplace. http://oreto.esi.uclm.es/aplicaciones/

ezoco [accessed on 4/12/2009].

Turban, E., Lee, J. K., King, D., McKay, J., and Marshall,

P. (2008). Electronic Commerce: A Managerial Per-

spective. Prentice-Hall, 5 edition.

Weiss, G. (1999). Multiagent systems: a modern approach

to distributed artiﬁcial intelligence. The MIT Press,

Cambridge, Massachusetts London, England.

Wise, R. and Morrison, D. (2000). Beyond the ex-

change: The future of b2b. Harvard Business Review,

78(6):86–97.

Zadeh, L. (1965). Fuzzy sets. Information and Control,

8:338–353.

Zoconet. Todocoleccion: http://www.todocoleccion.net/

[accessed on 15/01/2010].

B2C AND C2C E-MARKETPLACES - A Multi-layer/Multi-agent Architecture to Support them

105