A MODEL OF AGENT ONTOLOGIES FOR B2C E-COMMERCE

Domenico Rosaci

DIMET Department - University of Reggio Calabria

Via Graziella - Loc. Feo di Vito 89060 Reggio Calabria (Italy)

Keywords:

Intelligent Agents, B2C E-Commerce, Ontologies

Abstract:

This paper proposes a formal model of agent ontologies, suitable to represent the realities of both customers

and sellers in a B2C electronic commerce scenario. This model is capable of describing the entities involved

in the above realities (products, product features, product categories) as well as the behaviour of customers

and sellers in performing their activities. A system architecture, based on the presented ontology model, is

also brieﬂy described.

1 INTRODUCTION

In the last few years, many agent-based systems for

supporting business-to-customer (B2C) e-commerce

activities have been proposed. In this context, agents

can be seen as mediators between the actual involved

subjects, i.e. customers and businesses. Traditional

marketing research has developed many descriptive

theories and models that attempt to capture the Con-

sumer Buyer Behaviour (CBB) and that differ in var-

ious aspects; however, we can highlight the following

ﬁve relevant steps:

1. Need Identiﬁcation. In this stage the consumer is

stimulated to become aware of some unmet need.

For instance, consider the case of a customer inter-

ested in a certain category of books. Agents can

continuously monitor the Web and advert the cus-

tomer when a new book of that category is avail-

able.

2. Product Brokering. Once a consumer has identi-

ﬁed a need to satisfy, he has to ﬁnd what to buy

through a careful evaluation of the various prod-

ucts possibly satisfying that need. This requires a

comparison of product alternatives based on some

consumer-provided criteria. At the end of this step,

a set of products, usually called the consideration

set, capable of satisfying the consumer desires, has

been identiﬁed.

3. Buyer Formation Coalition. The customers, after

having chosen the product to buy in the product

brokering stage, before choosing the most suitable

merchant in the merchant brokering stage, may in-

teract with other (similar) buyers to form a buyer

coalition, in order to approach the merchant with a

large order and thus obtain a leverage.

4. Merchant Brokering. In this step, the consideration

set is combined with merchant-speciﬁc alternatives

based on consumer-selected criteria (e.g, availabil-

ity, price, delivery time, warranty, reputation, etc.)

for helping the consumer to determine whom to buy

from.

5. Negotiation. During this step, the various terms of

the transaction as, for instance, the price, are de-

termined. The beneﬁt of negotiating the price of

a product, instead of ﬁxing it, is that negotiation

relieves the merchant from needing to determine

the value of the good a priori (Maes et al., 1999).

Rather, the price is dynamically determined by the

marketplace.

Several agent-based systems have been proposed

for mediating the above activities. For a detailed sur-

vey of these systems, see (He et al., 2003). How-

ever, none of the existing approaches supports in a

uniﬁed manner all the described stages, allowing cus-

tomers and sellers to interact in a virtual marketplace

by using a unique, integrated, framework. This paper

aims to propose such a uniﬁed approach, by deﬁn-

ing a formal ontology model which both the customer

and seller realities are represented in, and that can be

fruitfully exploited in all the CBB stages. Such an

Rosaci D. (2004).

A MODEL OF AGENT ONTOLOGIES FOR B2C E-COMMERCE.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 3-9

DOI: 10.5220/0002619000030009

 SciTePress

ontology provides a ﬂexible way to access customer

and seller information, since it contains both an in-

tensional representation of the above realities (i.e., a

meta-schema), as well as an extensional representa-

tion that allows, when it is necessary, to handle the

actual data. Furthermore, the ontology represents cus-

tomer and seller behaviour (e.g., negotiation strategy)

by using logic propositions able of dynamically acti-

vating some pre-deﬁned agent’s actions.

The paper is structured as follows: in the Section

2, we deal with some related work; in the Section 3,

we give a formal description of our ontology model.

As an example of application of this model, in the

Section 4, an architecture of a system supporting a

virtual marketplace is also brieﬂy described. Finally,

in the Section 5, we draw our conclusions.

2 RELATED WORK

The necessity of representing, in the proﬁle of a

customer, not only concepts of his interest but also

his behaviour in accessing those concepts, has been

considered in several works in the Information Sys-

tems ﬁeld as, for instance, in (Bergamaschi et al.,

2001; Terracina and Ursino, 2000; Buccafurri et al.,

2002a). As far as the e-commerce research is con-

cerned, MOMIS (Mediator envirOnment for Multiple

Information Sources) (Bergamaschi et al., 2001) han-

dles both integration and querying of multiple, het-

erogeneous information sources, storing both struc-

tured and semi-structured data. Data source inte-

gration is carried out by following a semantic ap-

proach based on Description Logics, clustering and

a common data model capable of representing all in-

volved data sources. In (Buccafurri et al., 2002b;

Rosaci et al., 2002), a multi-agent system for repre-

senting and handling e-commerce activities is pro-

posed. In such an approach, an agent is present

in each e-commerce site, handling the information

stored therein. Furthermore, an agent is associated

with each customer, handling his proﬁle. The infor-

mation associated with both sites and customer pro-

ﬁles is represented and handled using a particular con-

ceptual model called B-SDR network. This latter al-

lows to uniformly manage heterogeneous data sources

and to construct and maintain a proﬁle storing infor-

mation about the visits the customer carries out to the

various e-commerce sites. MORPHEUS (Yang et al.,

2001) is a comparison-shopping agent that automat-

ically collects product descriptions from a group of

on-line stores on user’s behalf. Since the Web stores

are heterogeneous, a wrapper must be built and main-

tained for each store. Ontologies, i.e. intensional

descriptions of product characteristics and customer

and seller behaviour, have been already exploited in

B2C e-commerce context (Omelayenko, 2001) and

also many ontology-based approaches have been pro-

posed in multi-agent systems ﬁeld (OAS 2002, 2002).

All the previously described approaches try to

solve the heterogeneity by adopting techniques based

on the integration of the various sources or on the use

of wrappers. They are very interesting from the view-

point of a user that want to consider the whole Web for

searching goods of his interest. However, the price to

pay for obtaining the integration of a potentially over-

whelming amount of Web sources is often high. In-

deed, integration techniques are onerous in terms of

time to spend for constructing the integrated global

representation of the various sources. Moreover, such

a global representation is often very difﬁcult to han-

dle, since it has large dimensions and needs a con-

tinuous pruning of the less important concepts. On

the contrary, the approach we propose in this paper

is based on the construction of virtual marketplaces

whose actors (sellers and customers) are represented

in a uniform manner, due to the use of agents that op-

erate as a personal assistant and translate the user in-

terests and preferences in a pre-deﬁned standard, rep-

resented by the agent ontology model that we here

introduce.

3 AGENT ONTOLOGIES

The ontology of an agent is a representation of the re-

ality of the agent’s owner. In a B2C e-commerce con-

text, such a owner can be either a customer or a seller.

In the ﬁrst case, the agent ontology has to store the

customer interests and preferences w.r.t. the virtual

marketplace and the customer behaviour in purchas-

ing goods (e.g., the way of visiting e-commerce sites,

the strategies in negotiating, etc.). In the second case,

the agent ontology has to represent the product cate-

gories of the seller, the product characteristics (price,

availability, etc.) and the service features (warranties,

time delivery, etc.).

In this section, we propose a formal deﬁnition of

an agent ontology. In order to better explain the vari-

ous concepts we introduce below, we propose a sim-

ple situation of a customer and a seller that interact in

a virtual marketplace. This situation will serve as a

leading example along this paper.

Example 1 Let John be a customer interested in

purchasing books and CDs. His interest in books

mainly concerns with narrative and poetry. In the

past, he purchased on the Internet the books Anna

Karenina, The Buddenbrok, Les Fleurs du mal and

La Divina Commedia. He also purchased the CD The

Ghost of Tom Joad. In the case of a non negotiable

price, John usually behaves as follows: (i) when he

purchases a book, he also purchases a CD; (ii) more-

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

over, he considers a book (resp., a CD) as an inter-

esting book (resp., an interesting CD) if the price is

smaller than 20 US$ and the delivery time is smaller

than 3 days. In the case of a negotiable price, suppose

John negotiates with a seller for a good that has a

base proposed price equal to p(0) at the step 0 of the

negotiation. The John’s behaviour is as follows. At

the step 0, John makes an offer o(0) = 0.8∗p(0). At

the step i of the negotiation (i = 1, 2...), in response

to a new proposed price p(i) of the seller, John offers

a value o(i) = (p(i) + o(i − 1))/2.

Let W ord&Music a seller of books and CDs. It

deals with three categories of books, namely narra-

tive,essay and poetry. This seller, in a bilateral ne-

gotiation with a customer, behaves as follows. Let

p(0) be the base proposed price of a book, that it pro-

poses at the step 0 of a negotiation with a customer.

Suppose it receives an offer o(i), at the step i of the

negotiation (i = 0, 1, ....), by the customer. If o(i)

is smaller than 0.7 ∗ p(i), the seller aborts the nego-

tiation. Otherwise, it proposes a new price equal to

p(i + 1) = (o(i) + p(i))/2.

From this simple situation, we can observe that dif-

ferent kinds of knowledge need to be represented in

the ontologies of the customer and the seller. Namely:

• Entities. In both the cases of the customer and the

seller in the example above, the agent ontology has

to represent some products relating to the agent’s

owner. Each product must have an ID that iden-

tiﬁes it, and a set of associated features that gives

some information about it. For instance, a book

may have a feature title, a feature author, a feature

delivery time and a feature price. Such a represen-

tation of books is an intensional entity (a metadata)

and each actual book can be viewed as an instance

of this entity (a data). For example, the book Anna

Karenina may be represented by an entity instance

with ID=1, title =“Anna Karenina”, author =“Tol-

stoj”, delivery time =2 and price =17. More for-

mally:

Deﬁnition 1 Let T be a set {String,Integer,Real...}

of data types. We deﬁne the feature set, denoted by

F , as a set of variable names, each variable hav-

ing a type that belongs to T . A feature is an ele-

ment f ∈ F . Let D be a function from F to T ,

that associates to each feature f ∈ F its data type

D(f) ∈ T . An instance of the feature f is a value

v ∈ D(f).

Deﬁnition 2 Let f, g ∈ F be two features with

D(f) = D(g) and let k ∈ D(f) be a constant.

With the notation f = g we mean that the value of

f is set equal to the value of g, and with the nota-

tion f = k we mean that the value f is set equal to

the value k.

This deﬁnition can be applied, besides to prod-

uct characteristics, also to each object belonging

to the customer’s (resp., the seller’s) world as, for

instance, proposed price and offer that may be

represented by the features proposed price and

off er, respectively. In sum, features can be ex-

ploited both in the deﬁnition of products than in the

deﬁnition of more general entities. For instance, an

entity negotiation may represent the behaviour of

both John and W ord&Music in negotiating, and

proposed price and offer may be the features of

this entity. We give below a formal deﬁnition of

entity.

Deﬁnition 3 The entity domain E is the set of all

the tuples hID, f

, f

, ..., f

i, where ID is an In-

teger variable and f

, f

, ..., f

∈ F . An entity is

an element of E.

Deﬁnition 4 Let e = hID, f

, f

, ..., f

i be an

entity. An instance of e is a tuple i =

hidv, f v

, f v

, ..., fv

i, where idv ∈ Integer and

∈ D(f

), f v

∈ D(f

), ..., fv

∈ D(f

• Categories. A category is a collection of enti-

ties. For example, in the ontology of the seller

W ord&Music, the category BOOKS may group

all the books available to be purchased. More-

over, a category can be organized in some sub-

categories. For instance, in our leading example,

the category BOOKS of John contains the sub-

categories P OETRY and NARRAT IV E. Thus,

generally, if we consider a set of entities as a limit-

case of category (i.e., a category that does not con-

tains any sub-categories), we can inductively say

that a category is either a set of entities or a set of

sub-categories. More formally:

Deﬁnition 5 The category domain C is the set of

all the tuples

hID, e

, e

, ..., e

i, where ID is an Integer vari-

able and e

, e

, ..., e

∈ E, and of all the tuples

hID, c

, c

, ..., c

i, where ID is an integer and

c1, c2, ..., c

∈ C. A category is an element of C.

In the Figure 1, all the categories, instances and fea-

tures involved in the John

s reality are described.

Note that we have inserted the entity N egotiation

in a particular category BEHAV IOUR.

Categories are intensional information. An in-

stance of a category is a set of instances of all the

entities belonging to it. We can inductively deﬁne

a category instance as follows:

Deﬁnition 6 Let c = hID, e

, e

, ..., e

i be a

category, where ID is an Integer variable and

, e

, ..., e

∈ E. An instance of c is a tuple

i = hidv, ei

, ei

, ..., ei

i, where idv ∈ Integer

and ei

, ei

, ..., ei

∈ Integer are identiﬁers of

entity instances.

A MODEL OF AGENT ONTOLOGIES FOR B2C E-COMMERCE

JOHN=hID, BOOKS, CDS, BEHAVIOURi;

BOOKS=hID, NARRATIVE, POETRYi;

NARRATIVE=hID, Book, Booki;

POETRY=hID, Book, Booki;

CDS=hID, Cdi;

BEHAVIOUR=hID, Negotiationi;

Negotiation=hID, proposed

price, oﬀeri;

Book=hID, title, author, delivery time, pricei;

Cd=hID, title, author, delivery

time, pricei;

title, author=String;

delivery time, price, proposed price,offer=Integer;

Figure 1: The categories, entities and features of the customer J ohn

Let c = hID, c

, c

, ..., c

i, where ID is an integer

and c1, c2, ..., c

∈ C. An instance of c is a tuple

i = hidv, ci

, ci

, ..., ci

i, where idv ∈ Integer

and ci

, ci

, ..., ci

∈ Integer are identiﬁers of

category instances.

In the Figure 2, all the category instances involved

in the John

s reality are described.

This ﬁgure represents the following information:

The category instance JOHN1 (representing all the

interests of the customer John) has ID=5 and

contains the category instance with ID=4 (i.e.,

BOOKS1), the category instance with ID=1 (i.e.,

CDS1) and the category instance with ID=6 (i.e.,

BEHAVIOUR1). In its turn, the category instance

BOOKS1 has ID=4 and contains the category in-

stance with ID=3 (i.e., NARRATIVE1) and the cat-

egory instance with ID=2 (i.e., POETRY1). The

category instance CDS1 has ID=1 and does not

contain any category instances since, as speciﬁed

by the schema of its category CDS, it is com-

posed by an entity instance, namely that one hav-

ing ID=5 (i.e., Cd1). Similarly, the category in-

stance BEHAVIOUR1 has ID=6 and is composed

by the entity instance with ID=6 (i.e., Negotia-

tion1). The category instance NARRATIVE1 has

ID=3 and is composed by the entity instance with

ID=1 (i.e, Book1) and by the entity instance with

ID=2 (i.e., Book2). The category instance PO-

ETRY1 has ID=2 and is composed by the entity

instance with ID=3 (i.e, Book3) and by the entity

instance with ID=4 (i.e., Book4).

The entity instances with IDs=1..4 represent books.

The entity instance with ID=5 represents a CD and

the entity instance with ID=6 represents a step of a

negotiation.

• Knowledge patterns. An ontology has to store,

besides information about the involved entities,

also other information relative to the behaviour of

the customer (resp.,the seller) in purchasing (resp.,

in selling) the products. In our leading example, the

negotiation behaviour of the customer John and

the seller W ord&Music are some examples of this

kind of knowledge that have to be represented.

Such a knowledge can be stored into an ontol-

ogy by considering a set of events, that rep-

resents actions belonging to the activity of the

customer (resp., the seller), that may happen or

not. For examples, in the leading example’s sit-

uations, the concepts of interesting book, in-

terestingCD, buy a book, buy a CD, make

an offer, propose a new price can be modeled

as events and represented by boolean variables

as interestingBook, interestingCD, buyBook,

buyCD, makeO, proposeP , respectively. More-

over, also a relational expression involving features

is an event as, for instance, (deliveryT ime < t)

or (price < p). An event can be thus represented

by a boolean expression.

Deﬁnition 7 An event is either: (i ) a boolean vari-

able or (ii ) an expression of the form aθb, where

a, b ∈ F , D(a) = D(b) and θ ∈ {=, <, ≤, >, ≥}

is a relational operator or (iii ) an expression of the

form aθc, where a ∈ F and c ∈ D(a) and θ is a

relational operator. The events of the type (ii ) and

(iii ), that involve only features and do not involve

any boolean variable, are called feature-events.

The existing relationships between events can be

represented by propositional rules as, for instance,

relative to the situations described in our leading

example for the John customer:

: buyBook ⇒ buyCD

: (price < 20), (time < 3) ⇒ interestingBook

: (price < 20), (time < 3) ⇒ interestingCD

: interestingBook, proposeP ⇒ makeO

or, for the W ord&Music seller:

: makeO ⇒ proposeP

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

JOHN1=h5, 4, 1, 6i;

BOOKS1=h4, 3, 2i;

NARRATIVE1=h3, 1, 2i;

POETRY1=h2, 3, 4i;

CDS1=h1, 5i;

BEHAVIOUR1=h6, 6i;

Negotiation1=h6, 5, 4i;

Book1=h1, “AnnaKarenina

, “T olstoj

, 17, 2i;

Book2=h2, “T heBuddenbrook

, “M ann

, 18, 2i;

Book3=h3, “Lesf leursdumal

, “Beaudelaire

, 15, 2i;

Book4=h4, “LaDivinaCommedia

, “Alighieri

, 25, 2i;

Cd1=h5, “T heGhostof T omJoad

, “Springsteen

, 18, 2i;

Figure 2: The category instances and the entity instances of the customer J ohn

These rules, that we call knowledge patterns, afﬁrm

that some events happen when other related events

happen. For instance, the rule k

afﬁrms that if the

event buyBook has the value true, also the event

BuyCD has the value true.

More formally:

Deﬁnition 8 A knowledge pattern k

is a propositional rule of the form

, a

, ..., a

, ¯a

n+1

, ¯a

n+2

, ..., ¯a

⇒ b , where a

, a

n+1

, a

n+2

, ..., a

, b are events. This nota-

tion means that, if both a

, a

, ..., a

assume at the

same time the value true, and ¯a

n+1

, ¯a

n+2

, ..., ¯a

assume at the same time the value false, then b

assume the value true. Let fs = {f

, f

, ..., f

}

be a set of features. We say that k is a knowledge

pattern on fs, if each feature of f s appears in

almost one of the events a

, a

, ..., a

contained in

k, and all the events a

, a

, ...a

are feature-events.

• Actions. Often, when an event happens in the

world of a customer or a seller, an action is conse-

quently produced. For instance, in our leading ex-

ample relative to the customer John, when John

decides to make an offer in a negotiation phase, the

value of the offer is equal to the mean between

his previous offer and the price proposed at the

present by the seller. We can thus say that, when

the event makeO becomes true, a program, that we

denote by of, is called that sets the value of the fea-

ture offer equal to (proposed

price + off er)/2,

where proposed price is the feature representing

the price proposed by the seller. Similarly, in the

case of the W ord&Music seller, we can say that,

when the event proposeP becomes true, a pro-

gram pf is activated that behaves as follows: if

off er < 0.7 ∗ proposed price, it sets the event

end to the value true and then terminates, other-

wise it sets the value of proposed price equal to

(proposed price + off er)/2. In this latter case,

we observe that the program modiﬁes both a fea-

ture value and an event value.

Whe call action a 5-tuple composed by an event, as

makeO, a program, as co, that is activated by the

event, a set of features, as {proposed price}, that

are the arguments of the program, and another two

sets of features and events, as {offer} and {end},

respectively, whose value is modiﬁed by the pro-

gram. More formally:

Deﬁnition 9 An action is a tuple

he, P, fs1, es, fs2i where e ∈ E, P is a pro-

gram, fs1, fs2 ∈ 2

, es ∈ 2

, such that the

program P is activated if e = true by passing it as

input arguments the features belonging to the set

fs1 and P modiﬁes both the value of the features

belonging to the set fs2 and the events belonging

to the set es.

In our leading example, we can deﬁne the actions

J ohn

= hmakeO, of, {proposed price}, {},

, {of fer}i

W ord&M usic

= hproposeP, pf, {of fer}, {end},

, {proposed price}i.

Now, we can give our deﬁnition of agent ontology.

This is a collection of four sets: a set of entities, rep-

resenting all the products which the customer is in-

terested in (resp., which are in the seller’s catalog)

and all the other entities belonging to the customer

(resp., the seller) activities; a set of categories, de-

scribing the hierarchical structure of the entities; a set

of knowledge patterns, describing the rules followed

by the customer in purchasing (resp., followed by the

seller in selling) products and a set of actions, speci-

fying what actions the customer (resp., the seller) per-

forms when the environment changes. Since all these

sets contain only intensional information, such an on-

tology is purely intensional and can be viewed as a

schema. We also deﬁne the instance of an ontology,

that contains, for each entity (resp., category) of the

ontology, also a set of instances of that entity (resp.,

category). More formally:

Deﬁnition 10 An agent ontology schema is a 4-tuple

hE, C, K, Ai, where: (i) E is a set of entities; (ii) C

is a set of categories; (iii) K is a set of knowledge

patterns; (iv) A is a set of actions.

Deﬁnition 11 An ontology instance of an ontology

O=hE, C, K, Ai is a 4-tuple hEP, CP, K, Ai, where:

A MODEL OF AGENT ONTOLOGIES FOR B2C E-COMMERCE

(i) EP is a set of of pairs (e, ei), such that e ∈ E is an

entity and ei is a set of instances of e; (ii) CP is a set

of of pairs (c, ci), where c ∈ C is a category and ci

is a set of instances of c; (iii) K is a set of knowledge

patterns; (iv) A is a set of actions.

As an example, the ontology of the customer John

is shown in the Figure 3. An instance of this ontology

is shown in the Figure 4.

4 THE OBA-B2C SYSTEM’S

ARCHITECTURE

The OBA-B2C system allows to realize a virtual mar-

ketplace. It is composed by a central unit, called

agency, and several agents connected to the agency.

There are two agent typologies in the system, namely

the customer agents and the seller agents. On the

one hand, each customer agent is associated with a

real customer, and operates on his behalf in order to

automatically carry out the various stages of the e-

commerce. On the other hand, each seller agent is

associated to a real seller and performs e-commerce

activities on the seller’s behalf.

Each agent has to be registered on the agency in

order to be enabled to participate to the virtual mar-

ketplace. The agency is a Web site that provides sev-

eral services to the actors of the virtual marketplace.

The registration of an agent consists in the assignment

of two access codes (an agent name and a password)

to the agent. Each registered agent can enter into

the marketplace by accessing the agency site and au-

thenticating itself by means of the above codes. The

agency maintains a list of all the registered agents, and

is able to provide information about them. By means

of the agency, an agent of the virtual marketplace can

ﬁnd those agents that have similar interests, in order

to form a coalition of buyers. It is worth pointing out

that the agency does not make public the information

about the registered agents, in order to protect cus-

tomer privacy. Thus, the agency behaves as a media-

tor that supports the agent communication. As an ex-

ample, an agent a that desires to know the agents that

have similar preferences, send to the agency those in-

formation about itself (personal preferences, buying

behaviour, etc.) it wants to make public. Automati-

cally, the agency contacts all the agents of the market-

place, sends them the above information, receives the

answers by the contacted agents and ﬁnally transmits

these answers to a. Agency also mediates virtual auc-

tions, by allowing the seller agents of the marketplace

to create auctions for selling goods and the customer

agents to bid offers for desired goods.

In the client-agent architecture of the OBA-B2C

system, the customer agent is a client that a real cus-

tomer exploits for participating to the virtual market-

place. This agent (see Figure 5) is composed by two

knowledge bases called ontology and address book,

and by two main programs, called ontology manager

and communication manager, respectively. The core

of the agent architecture is the ontology, that has the

structure deﬁned in the Section 3. The ontology is

handled by the ontology manager, that exploits the

information contained in it for supporting the various

stages of the e-commerce. For this reason, the ontol-

ogy manager is composed by ﬁve program modules

that mediate, respectively, need identiﬁcation, prod-

uct brokering, buyer coalition formation, merchant

brokering and negotiation. Each module performs, on

the behalf of the customer, the associated activity. For

instance, the merchant brokering manager exploits the

information contained in the customer ontology rela-

tive to the products of interest, and ﬁnds in the vir-

tual marketplace the best merchants for those prod-

ucts. Each program module also monitors the cus-

tomer behaviour in the associated activity. All the

program modules of the ontology manager need to

contact the agents of the marketplace for performing

their tasks. For instance, the buyer coalition forma-

tion has to contact all the agents of the marketplace

in order to ﬁnd possible collaborators. The commu-

nication among the agents is handled by a component

called communicator manager . This module exploits

a database of agent addresses, called address book,

for performing its tasks.

The seller agent is a client software that a real seller

can exploit for participating in the virtual market-

place. This agent is composed by an ontology, an ad-

dress book and tree main programs, called ontology

manager, site builder and communication manager,

respectively. Similarly to the customer agent, the core

of the seller agent architecture is the ontology that,

in this case, is not built by observing the owner be-

haviour but, obviously, the seller itself is able to build

it by specifying the categories of interests and the

products to sell. This activity is performed by a dedi-

cated component called site builder. The site builder

operates as a seller assistant, allowing the seller to

build his Web store by using a uniform representa-

tion w.r.t. the virtual marketplace. For instance, if the

seller tries to insert in his ontology a product (resp., a

product category) that has the same characteristics of

a product (resp., a product category) always present in

the ontology of another seller of the marketplace, the

site builder identiﬁes the product(resp., the product

category) in a unique manner, by assigning it a unique

identiﬁer, and thus avoiding heterogeneous represen-

tations of the same good.

The ontology manager and the communication

manager have the same role than in the customer

agent architecture. Obviously, the ﬁve modules that

support the CBB stages operate in this case on the

seller’s viewpoint.

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

E={Book,Cd,Negotiation}

C={JOHN,BOOKS,NARRATIVE,POETRY,BEHAVIOUR}

K={k

, k

}

A={a

J ohn

}

Figure 3: The J ohn’s ontology

E={(Book,{Book1,Book2,Book3,Book4}), (Cd,Cd1),(Negotiation,Negotiation1)}

C={(JOHN,JOHN1),(BOOK,BOOKS1),

(NARRATIVE,NARRATIVE1),(POETRY,POETRY1),

(BEHAVIOUR,BEHAVIOUR1)}

K={k

, k

}

A={a

J ohn

}

Figure 4: An instance of the J ohn’s ontology

Figure 5: The customer agent’s architecture.

5 CONCLUSIONS

This paper describes a new model for representing

agent ontologies in a B2C e-commerce scenario, suit-

able to support the various stages of the Consumer

Buyer Behaviour (CBB) model. The proposed model

is capable of representing both the concepts involving

in customers and sellers realities as well as the be-

haviour of customers and sellers in performing their

activities. An architecture of a multi-agent system

based on the presented ontology model is also brieﬂy

described. Our ongoing research mainly deals with

the deﬁnition of machine learning techniques for efﬁ-

ciently extracting the knowledge patterns of the agent

ontologies by directly observing customer and seller

behaviours.

REFERENCES

Bergamaschi, S., Castano, S., Vincini, M., and Beneven-

tano, D. (2001). Semantic integration and query of

heterogeneous information sources. Data & Knowl-

edge Engineering, 36(3):215–249.

Buccafurri, F., Lax, G., Rosaci, D., and Ursino, D. (2002a).

A user behavior-based agent for improving web usage.

In Proceedings of the International Conference on

Ontologies, Databases and Applications of Semantics

Conference (ODBASE 2002), Lecture Notes in Com-

puter Science, pages 1168–1185, Irvine, CA, USA.

Springer-Verlag.

Buccafurri, F., Rosaci, D., Sarn

e, G., and Ursino, D.

(2002b). An agent-based hierarchical clustering ap-

proach for e-commerce environments. In Proceedings

of the 3th E-Commerce and web Technologies (EC-

Web 2002), Lecture Notes in Computer Science, pages

115–118, Aix-en-Provence, France. Springer-Verlag.

He, M., Jennings, N., and Leung, H. (2003). On agent-

mediated electronic commerce. To appear in IEEE

Transactions on Knowledge and Data Engineering.

Maes, P., Guttman, R., and Moukas, A. (1999). Agents that

buy and sell. Communications of the ACM, 42(3):81–

91.

OAS 2002 (2002). Workshop on ontologies in

agent systems, ﬁrst international joint con-

ference on autonomous agents and multi-

agent systems (aamas:oas 2002), available at

http://autonomousagents.org/oas/2002/proceedings.pdf.

Omelayenko, B. (2001). Syntactic-level ontology inte-

gration rules for e-commerce. In Proceedings of

the 14th International FLAIRS Conference (FLAIRS-

2001), pages 324–328, Key West, Florida, USA.

AAAI Press.

Rosaci, D., Sarn

e, G., and Ursino, D. (2002). A multi-agent

model for handling e-commerce activities. In Pro-

ceedings of the International Database Engineering

and Applications Symposium (IDEAS 2002), pages

202–211, Edmonton, Canada. IEEE Computer Soci-

ety.

Terracina, G. and Ursino, D. (2000). Deriving synonymies

and homonymies of object classes in semi-structured

information sources. In Proceedings of International

Conference on Management of Data (COMAD 2000),

pages 21–32, Pune, India. McGraw Hill.

Yang, J., Seo, H., and Choi, J. (2001). Morpheus: a more

scalable comparison-shopping agent. In Proceedings

of the 5th International Conference on Autonomous

Agents (AGENTS 2001), pages 63–64, Montreal, Que-

bec, Canada. ACM Press.

A MODEL OF AGENT ONTOLOGIES FOR B2C E-COMMERCE