SEMANTIC SUPPORT FOR AUTOMATED NEGOTIATION

WITH ALLIANCES

Stanislav Pokraev

, Zlatko Zlatev

, Rogier Brussee

, Pascal van Eck

Telematica Instituut, P.O.Box 589, 7500 AN, Enschede, The Netherland

Centre for Telematics and Information Technology,

University of Twente, P.O. Box 217, 7500 AE, Enschede,The Netherlands

Keywords: Automated negotiation, Alliances, Ontology, Semantics, OWL, Agents, e-Commerce

Abstract: Companies can form alliances on the Internet to aggregate buying or selling power and create value. More

concretely, together resources are shared or new possibilities are exploited that cannot be exploited

individually. Most alliances are formed as a result of a negotiation process between the companies that form

an alliance. This paper proposes a software framework that enables automated negotiation between

alliances. Our framework allows for the semantic description of negotiation objects and their attributes, and

provides a mean for the exchange of negotiation messages unambiguously interpretable by all parties

involved. The proposed framework supports ad-hoc alliances by allowing parties with a common interest to

negotiate on the proposal they want to make to other market participants first.

1 INTRODUCTION

The Internet provides opportunities to create value

by forming alliances to aggregate buying or selling

power. More specific, companies form alliances

either to share current resources and competences or

to exploit new possibilities. The motives for forming

alliances are diverse. Alliances are either formed by

co-option of competitors or by providers of

complementary products or services to reduce cost

and improve customer offerings. Others allow each

partner to concentrate on activities that best match

their resources and competences or to learn from the

other partners and develop competences that may be

more widely exploited..

Kaplan and Sawhney (2000) have investigated

allian

ces in their appearance as so-called ‘e-hubs’:

B2B marketplaces that aggregate or match buyers or

sellers. E-hubs can be neutral third parties, or buyer

or seller biased. In their paper, Kaplan and Sawhney

indicate which model - aggregation or matching,

biased or neutral - works best in a specific market

setting.

In our paper, we develop a software framework

r the dynamic aggregation of buyers and/or sellers

in an electronic market. The framework provides an

architecture for automated negotiation between

alliances. The framework relies entirely on the use

of ontologies as a mean of formally representing the

knowledge on a particular domain of interest. This

provides clear semantics and represents shared

understanding of the issues being negotiated on.

Common understanding of the market, knowledge of

the assets that the potential partners are ready to

commit as well as shared expectations is a crucial

prerequisite for choosing the right form of alliance

or the discovery of the pitfalls in time.

2 SEMANTIC DESCRIPTIONS

Present e-commerce applications extend over the

boundaries of an enterprise. New business scenarios

include advertising of services or products on the

Internet, matchmaking of sellers and buyers,

negotiation with potential partners, etc. These

scenarios presume the use of information outside its

original context where it tends to lose its semantics.

It is therefore widely believed that semantically

richer descriptions are needed for tasks such as

service and request matchmaking, negotiation,

contracting and contract enactment.

Ontologies will play a crucial part in facilitating

e sharing of information between communities,

both of people and of software agents. To support

244

Pokraev S., Zlatev Z., Brussee R. and van Eck P. (2004).

SEMANTIC SUPPORT FOR AUTOMATED NEGOTIATION WITH ALLIANCES.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 244-249

DOI: 10.5220/0002628702440249

 SciTePress

the use of ontologies, a number of representational

formats have been proposed, including RDF Schema

(re: RDF Vocabulary Description Language), the

Ontology Interchange Language (OIL) (re: Ontology

Inference Layer) and the DARPA Agent Markup

Language (DAML) (re: DARPA Agent Markup

Language). These last two have been unified to form

DAML+OIL (DAML+OIL, 2001), which is the

basis of the ontology web language (OWL) (re: Web

Ontology Language). The OWL language is now a

W3C standard for ontology and metadata

representation. OWL exploits existing web standards

such as XML, RDF and RDFS and adds primitives

of object oriented and frame based systems, as well

as the formal thoroughness of description logic. Its

formal and rational basis provides powerful means

for knowledge representation and reasoning services

that are essential in the negotiation process.

3 NEGOTIATION AND

ALLIANCES

Negotiation is a process whereby different entities

reach an agreement on joint future behaviour. The

need for negotiation arises when more than one

entity (actor) has overlapping interests. The entities

communicate their expectations about a potential

mutually acceptable agreement. The result of the

negotiation process is an agreement to which actors

commit themselves for a certain future course of

action.

Negotiation is usually decomposed into the

following elements: negotiation objects, negotiation

protocols and negotiation strategy (Jennings et al.,

2001). Because the third element is considered

private for every actor (Bartolini & Preist, 2001), we

will discuss the first two ones only. The negotiation

object, represents the subject (product or service) the

participants negotiate on. The negotiation protocol,

represents the rules that govern the negotiation

process.

Alliances are groups of business actors formed to

make collective use of resources or possibilities. The

process of forming an alliance can include

negotiation. An alliance can last for the duration of a

single deal or it can last for many deals. We are

interested in alliances formed to take part in a single

negotiation agreement which we will call ad-hoc

alliances.

Down below in this section we will discuss the

problems of automating the process encountered by

the participants of a negotiation process, which

motivate the approach proposed in this paper.

Problem one: lack of common understanding.

A negotiation process is an exchange of messages.

The messages are created by different actors and

therefore (potentially) different meaning is given to

the concepts used in them. The problem is the lack

of common understanding of issues being discussed

and the meaning of the exchanged messages. In our

approach, we present a solution for this problem

using ontologies.

Problem two: lack of common understanding

of the protocol. Participants in a negotiation process

should have a common understanding not only of

the negotiation issues but also of the protocol they

have to follow. One approach to this problem is to

use a formal description of the negotiation protocol.

Giving formal semantics to the negotiation protocol

is an issue that we want to address in future work. In

our current approach, we present a partial solution to

this problem.

Problem three: loose connection among

participants. In a many-to-many negotiation,

participants have the possibility to form alliances. If

the shared interests are only temporary, there is a

loose connection among participants and the

alliance is not stable. Therefore, it is not possible to

treat the alliance as one participant and simplify the

negotiation. We propose a solution that allows the

formation of ad-hoc alliances.

Problem four: negotiation within an alliance.

Apart from the problems in negotiating between

alliances, there is a problem of similar complexity in

forming the alliance. Alliance formation requires

negotiation. We look at the alliance formation

process as a separate negotiation process. The result

of an alliance formation is a mutually agreed

proposal that the alliance commits to. In our

approach, we propose a separate negotiation process

to form the alliance.

4 OUR APPROACH

C. Bartolini, C. Preist and N. Jennings (2002)

present an abstraction of a negotiation process. The

main concepts of the process are the following: To

negotiate with one another, parties must have a

common understanding of the different parameters

of the negotiation (e.g. price, quantity, delivery

terms, etc.). The negotiation process consists of

exchanging proposals representing the agreements

currently acceptable to the sending party involved.

The submitted proposal should be valid with respect

to the restrictions defined and it should be submitted

according to the set of rules governing the

negotiation.

In our approach, we extend the framework of

Bartolini et al. by (1) semantic annotation of

negotiated issues and messages and (2) providing

SEMANTIC SUPPORT FOR AUTOMATED NEGOTIATION WITH ALLIANCES

245

means for ad-hoc alliances. Our approach

investigates the use of OWL for describing the

negotiation objects and messages by the means of an

upper negotiation ontology. The proposed formal

semantic descriptions are a solution to the problem

of common understanding over the negotiated matter

(see section 4.1). Next, the proposed negotiation

ontology defines the structure of the exchanged

negotiation messages (see section 4.2). This solves

only part of problem two (section 0), that is, giving a

meaning to the messages. We present a partial

solution to the rest of the problem - a two-step

iterative exchange of negotiation proposals.

Furthermore, our approach allows the negotiating

parties to form ad-hoc alliances (see section 0 and

4.4). Actors that cannot satisfy the request as a

whole, can join an ad-hoc alliance and contribute to

the assembly of the requested package of services or

products. Finally, we enable negotiation within an

alliance by providing the means for negotiation over

a joint proposal among the allied actors (see section

4.4). The following sections give more insight into

our approach.

4.1 Negotiation objects

The process of negotiation requires that all parties

involved have a common understanding about the

negotiation issues (e.g. objects and their parameters).

Furthermore, support for reasoning is necessary to

ensure the validity of the proposals (for instance to

test whether two negotiation issues are non-

contradictory) or to derive implied relations between

negotiation objects. Finally, a particular negotiation

strategy might need to compute the concept

hierarchy. Information about which concept is a

specialization of another and which concepts are

synonyms can be used to support the decision-

making process.

To provide the desired functionality described

above, we propose the usage of ontologies to

describe the negotiation objects. This section gives

an idea of how this can be done using the OWL

language.

An ontology consists of object descriptions and

their properties. To describe objects we first need to

describe object types. This is done by defining

general classes (using owl:Class). If we need to

specify that a certain type is a subtype of another we

use the rdfs:subClassOf property. Once defined, the

object classes are used to describe the objects (e.g.

a1 is rdf:type of class A). OWL uses properties to

relate individuals to each other. Properties allow

introducing associated object characteristics, such as

color or weight. To limit the kinds of individuals

that can fill a certain property, OWL allows for

stating the property’s range (using rdfs:range

property). This mechanism allows new classes to be

created by restricting the properties of the existing

ones (e.g. all objects from class Wines that have

property color restricted to value red define the new

subclass of Red wines).

The expressive power and the formal rigor of

OWL allows negotiation objects to be described in

unambiguous way and therefore understood by all

the participants in a negotiation process. Its basis for

reasoning services provides a powerful mean to

check the validity of a proposal or to infer implicit

knowledge.

4.2 Negotiation messages

The process of negotiation in general is the iterative

exchange of negotiation proposals until agreement is

reached. Agreement is reached when the response of

a proposal is an identical proposal or when one party

repeats his proposal a second time in a row.

The proposed message structure is inspired by

the work of Jennings et al. (Jennings et al., 2001)

and Bartolini and Preist (Bartolini & Preist, 2001).

Bartolini and Preist’s messages are based on the

instantiation of a template message, which is the

proposal sent to the opponent. Jennings et al. see the

negotiation objects as a collection of negotiation

issues. We combine these two approaches and

propose the message structure as shown in Figure 1.

1..*

0..1

Negotiation

template

Negotiation

issue

owl:Class owl:OntologyProperty

Negotiation

agreement

Negotiation

proposal

rdfs:subClassOf

owl:collection

owl:Thing

neg:object

neg:attribute1

neg:value

owl:Class

Figure 1: Negotiation message

Prior to any negotiation, participants must agree

on the structure of the negotiation proposal, which is

defined by the Negotiation template (see Figure 1).

The Negotiation template specifies that every valid

negotiation proposal is a collection of Negotiation

issues. The Negotiation issue specifies the

negotiation object, an attribute of a negotiation

object and the value of that attribute. The

Negotiation template does not prescribe the number

of Negotiation issues, which allows parties to

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

246

change the structure of the proposal during the

negotiation process. The Negotiation template is a

blank proposal, i.e. neither attributes nor values are

specified. Nevertheless, the Negotiation template

does not forbid such specification. Its role is to

‘reserve’ space for participants to make their

concrete proposals. The Negotiation proposal is a

subclass of the Negotiation template. The

Negotiation proposal follows the structure of the

Negotiation template and the Negotiation issues are

fully specified. By fully specified, we understand

that to every negotiation object attribute there is a

corresponding value. Once accepted, a Negotiation

proposal becomes a Negotiation agreement.

4.3 Ad-hoc alliances

When businesses enter a negotiation process they

usually have expectations that do not match exactly.

Requestors ask for a complex product or service that

cannot be fulfilled by any party. Conversely,

providers may offer large discounts but for a

quantity that no one can handle separately.

Businesses deal with such problems by forming

alliances. Alliances are formed by an additional

negotiation among the future partners. In our

approach, we provide a means for forming alliances.

We call them ad-hoc alliances, because they are

formed dynamically in the beginning of the

negotiation process and disappear after an agreement

is reached.

The main components that conduct the

negotiation process in our framework are

negotiation host and ad-hoc alliance host (see

Figure 2). The negotiation host provides a medium

for negotiation among single business parties and

potential ad-hoc alliances. The ad-hoc alliance host

represents the ad-hoc alliance in the negotiation

process that takes place in the negotiation host. The

ad-hoc alliance host provides allied parties with

medium for negotiation of a common strategy. The

ad-hoc alliance host represents an ad-hoc alliance in

the negotiation process, which takes place in the

negotiation host. Both the ad-hoc alliance host and

the negotiation host have the same internal structure

(discussed in details in the next section). These hosts

have components (discussed in section 5.1)

responsible for admission of participants, validation

of proposals, enforcement of the protocol and the

update of negotiation information. What makes them

different is the template they use in the course of the

negotiation process. The ad-hoc alliances are formed

by sending invitations from the ad-hoc alliance host

to the matched service or product providers. They

can join the ad-hoc alliance by responding to these

invitations.

4.4 Negotiation process

In our work, we adopt the approach proposed by

Bartolini, Preist and Jennings (2002) and extend it

further with a new concept of ad-hoc negotiation

host. We propose a two-step negotiation process

(Figure 2). The first step is negotiation among allied

parties about their joint proposal to their opponent.

The second step takes place in the negotiation host

where parties negotiate over the joint proposal made

by the ad-hoc alliance. We discuss our approach in

details below.

Buyer

Negotiation process

Sellers

hoc

alliance

host

egotiation

host

Figure 2: Negotiation process

Sellers register their service or product

advertisements. The validity of the advertisement is

checked and if it is valid, sellers are notified that the

process of registration has been successful. When a

buyer wants to search for a service or product, he

creates a request – a virtual description of the

package of products or services he wants to have –

and files it. If the request is valid, the buyer is

notified that the registration has been successful and

the request is matched with the registered

advertisements.

In case of a partial or complete match, an ad-hoc

alliance host is created and all matched sellers are

invited to join in. In this way, the ad-hoc alliance

host serves as a forward aggregator in the

terminology of Kaplan and Sawhney (2000). At the

same time, a negotiation host is created, connected

to the ad-hoc alliance host and the buyer is invited to

join in. To be admitted to the negotiation process,

sellers must accept an ad-hoc alliance negotiation

template proposed by the ad-hoc alliance host and

conversely, the buyer must accept a negotiation

template proposed by the negotiation host. Both

negotiation templates are dynamically created and

specify the parameters of the negotiation (e.g.

product types, prices, delivery dates, etc) taking into

account the buyer’s request. Additionally, a new

negotiation issue is added to the ad-hoc alliance

negotiation template – the direction that all sellers

want to give to the buyer. When the participants are

admitted, the negotiation process begins at the ad-

hoc alliance host. Once all sellers agree on the

SEMANTIC SUPPORT FOR AUTOMATED NEGOTIATION WITH ALLIANCES

247

proposal they want to make to the buyer, the ad-hoc

alliance host sends this proposal to the negotiation

host. The buyer then agrees or makes a new proposal

that is sent back to the ad-hoc alliance negotiation

host.

5 ARCHITECTURE

This section presents an architecture that realizes our

approach. We define the main components and

describe the interactions among them. The

architecture is depicted in Figure 3.

5.1 Components

The communication layer is responsible for

providing the means for communication between the

components representing participants and the system

components. It provides secure and reliable

communication, while hiding the complexity of the

underlying network protocols.

The participant’s component represents a

requestor or provider of a product or service. It

contains facts and rules that represent the negotiation

strategy of the involved party and uses a reasoning

engine to carry out the negotiation process.

Figure 3: Architecture

The advertisement registrar is a component

responsible for registering a particular service or

product description. This component communicates

with a request validator component to check

whether a particular request or advertisement is valid

and if so, it registers the request in its advertisement

database.

The request validator component is responsible

for checking whether an advertisement or request

contains instances of agreed upon, domain-specific

concepts.

The matchmaker component is responsible for

matching requests with product or service

advertisements. Initially, it uses the request

validator component to check whether the issued

request is valid. If so, the matchmaker uses its

reasoning engine to match facts of the request with

the facts of the registered advertisements. If there is

no advertisement that completely satisfies the

request, the component tries to find advertisements

whose combination complies with the request. If

any, the matchmaker creates an ad-hoc alliance host

and initializes its admission control component to

accept only participants associated with the matched

advertisements. In addition it creates a negotiation

host and initializes its admission control component

to accept only the requestor and the ad-hoc alliance

host.

Once created and initialized, the ad-hoc alliance

host sends an invitation to all admitted participants

to submit their proposals. When a proposal is

received, the communication gateway checks

whether this proposal comes from an admitted

participant. If so, it sends it to the protocol enforcer.

The protocol enforcer checks the validity of the

proposal with the negotiation template. Then it

checks whether the proposal is submitted according

to the negotiation rules. If all conditions are met, it

uses the information updater to assert or retract the

new facts from the proposal to the negotiation

history database. Finally, it updates the negotiation

template. Once this is done, the protocol enforcer

sends the updated negotiation template to all

participants and invites the next participant to place

its proposal. After all participants reach an

agreement on the offer they want to make to the

requestor, the ad-hoc alliance host submits the

agreement as a proposal to the negotiation host.

Communication layer

Request

validator

Registrar

Matchmaker

Proposal

validator

Protocol

enforcer

Information

updater

Negotiation

templates

Negotiation

rules

Negotiation

history

Communication

gateway

d-hoc

alliance host

Negotiation host

Admission

control

Participant

Communication

Request

validator

Registrar

Matchmaker

Proposal

validator

Protocol

enforcer

Information

updater

Negotiation

templates

Negotiation

rules

Negotiation

history

Communication

gateway

d-hoc

alliance host

Negotiation host

Admission

control

Participant

The Negotiation host repeats the structure of the

ad-hoc alliance host. The same components enforce

the rules of negotiation between the requestor and

the ad-hoc alliance host.

6 RELATED WORK

Bartolini, C. Preist, C and Jennings, N. (2002)

present a general negotiation protocol, which can be

parameterized by different rules to implement a

variety of negotiation mechanisms. However, this

work does not provide support for multi-party

negotiations. Moreover, the lack of a formal

semantics of the negotiation objects and messages

could lead to misinterpretation of the issues being

discussed.

The work of Tamma, Wooldridge and Dickinson

(2002) presents an approach to ontology-based

ICEIS 2004 - SOFTWARE AGENTS AND INTERNET COMPUTING

248

automated negotiation. The approach complements

the one presented in the previous paragraph in a

sense that negotiation issues and protocol are

described in terms of a shared ontology. This

approach, as the previous one, does not address

multi-party negotiation.

7 FUTURE WORK

A number of issues raised in this paper require

further investigation. First, heterogeneous agents

meet difficulties joining an unknown negotiation

process because they need to know the protocol in

advance. In order to allow agents to participate in

arbitrary negotiations, the rules that govern a

particular negotiation should be described by the

means of formal semantics. This requires

investigation of the possibility to use DAML-Rules

(re: DAML Rules) for protocol description. Second,

the proposed solution for alliances requires further

refinements into the following directions: (1)

support for more that one alliance, including buyers’

and sellers’ sides. This will allow negotiation

between alliances and freedom for actors to choose

the alliance that suits them best; (2) better support

for actors inside an alliance for acceptance and

execution of a shared negotiation strategy.

8 CONCLUSIONS

In this paper, we address the problems of semantic

interoperability in existing systems for automated

negotiation. More, we discuss the lack of support for

alliances in a negotiation process. We use the

description logic provided by OWL as a means for

formal representation of agreed upon concepts and

relations from a particular domain of interest. The

OWL language also allows for complex reasoning,

while exploiting existing web standards such as

XML and RDF. Our work extends existing work in

the field by proposing a more advanced negotiation

template defined by means of OWL. Such a template

can significantly improve the process of negotiation

by removing the possible ambiguities. The latter is

achieved by restricting the range of the attributes

that a certain negotiation object may take.

Furthermore, in our approach we introduce the

notion of “ad-hoc alliance host” – a component that

enables multiple parties to agree on a joint proposal

when negotiating with other party or alliance.

ACKNOWLEDGMENTS

We would like to express our gratitude to prof. Roel

Wieringa from the University of Twente for his

comments and suggestions.

This work is partially funded by the WASP project

(re: WASP).

REFERENCES

Bartolini, C. & Preist, C., 2001, A Framework for

Automated Negotiation, HPL-2001-90, HP

Laboratories Bristol

Bartolini, C., Preist, C. & Jennings, N., 2002, A Generic

Software Framework for Auto-mated Negotiation,

http://www.hpl.hp.com/techreports/2002/HPL-2002-

2.pdf

DAML+OIL (March 2001) Reference Description,

http://www.w3.org/TR/daml+oil-reference

DAML Rules, http://www.daml.org/rules/

DARPA Agent Markup Language (DAML),

http://www.daml.org/

Jennings, N., Faratin, P., Lomuscio, A., Parsons, S.,

Sierra, C. & Wooldridge M. , 2001, Automated

Negotiation: Prospects, Methods and Challenges, Int

Journal of Group Decision and Negotiation, vol. 10,

no. 2, pp. 199-215

Kaplan, S. & Sawhney, M., 2000, E-Hubs: The New B2B

Marketplaces, Harvard Business Review, May-June

Ontology Inference Layer (OIL),

http://www.ontoknowledge.org/oil/

RDF Vocabulary Description Language: RDF Schema,

http://www.w3.org/TR/rdf-schema/

Tamma, V., Wooldridge, M., and Dickinson, I., 2002, An

ontology based approach to automated negotiation,

Proceedings of the IV workshop on agent mediated

electronic commerce, AAMAS 02 conference,

Bologna, Italy.

WASP project,

http://wasp.freeband.nl/

Web Ontology Language (OWL),

http://www.w3.org/TR/owl-ref/

SEMANTIC SUPPORT FOR AUTOMATED NEGOTIATION WITH ALLIANCES

249