SPECIFICATION AND IMPLEMENTATION OF MULTI-AGENT

ORGANIZATIONS

Fatemeh Ghassemi, Naser Nemat Bakhsh, Behrouz Tork Ladani

Department of Computer Engineering, Isfahan Universityt,Isfahan,Iran

Marjan Sirjani

Department of Electrical and Computer Engineering, University of Tehran, Tehran, Iran

Keywords: Organization, Formal Model of Organization.

Abstract: Multi-agent systems are used as a solution for complex and distributed systems. Since agents are autonomous they

can be coordinated exogenously by a coordination language Reo. Reo coordinates agents without having any

knowledge about agents. We apply organizational concepts to analyze and design such systems. In this paper, we

propose a formal model to specify the results achieved during these phases. This formal model helps in designing a

coherent and consistent system. The formal model is applied to make the implementation of system by Reo

systematically. We will specify and implement system by Reo according to the formal model. This paper also

defines how to convert the formal specification to a Reo circuit by providing Reo circuits for the different patterns

of interaction protocols and how to compose simpler circuits to support more complex patterns.

1 INTRODUCTION

Autonomous agents and multi-agent systems

(MASs) are widely used by developers to design

complex and distributed systems such as e-learning

and e-marketing systems and business-to-business

applications. An agent provides a behaviour

abstraction which allows the developers to naturally

model and construct complex systems.

In the context of MASs, the autonomous and

proactive behaviour of agents suggests that applications

can be designed by mimicking the behaviour and

structure of human organizations. Thus the architecture

of a multi-agent system can be naturally viewed as a

computational organization, which consists of a

multitude of autonomous and interacting agents. Each

agent plays one (or more) specific roles. However, the

organization of a multi-agent system is distinct from

the individual agents that populate the system

(Zambonelli et al., 2000; DeLoach, 2002; DeLoach and

Matson, 2004). While agents play roles within the

organization, their roles do not constitute the

organization. Roughly speaking organizations are

characterized by the organizational structures as well as

organizational rules that define the requirements for the

instantiation and operation of the organization as well

as constraints on agent behaviours and interactions

(Zambonelli et al., 2000; DeLoach, 2002).

Organization defines and coordinates agent

interactions. So a multi-agent system is defined by a set

of agents and its coordination (Dastani et al., 2005). We

only consider the external behaviour of agents. Agent

organization can be open, where agents enter or leave

dynamically. Thus agents are not known to each other

and they may not be honest to each other. However

there are some problems in the specification and

implementation of an open organization as follows:

• Participants in an organization should have a

common understanding of the organizational rules and

organizational structure. We consider the source of this

problem the use of protocol specification languages

with poor formal semantics.

• Some of agents in a large organization may make a

new sub-organization to cooperate for a specified task.

Since agents are autonomous, they are not affected by

this organizational changes and it is the responsibility

of the organization to manage dynamic changes.

Supporting such characteristics for an organization

requires implementation to have the ability to adapt

changes in organization policies.

447

Ghassemi F., Nemat Bakhsh N., Tork Ladani B. and Sirjani M. (2006).

SPECIFICATION AND IMPLEMENTATION OF MULTI-AGENT ORGANIZATIONS.

In Proceedings of WEBIST 2006 - Second International Conference on Web Information Systems and Technologies - Internet Technology / Web

Interface and Applications, pages 447-453

DOI: 10.5220/0001256204470453

 SciTePress

• There is always a gap between system specification

and implementation.

In this paper, we apply the coordination language

Reo to address above problems. Reo has a formal

semantics and is dynamically reconfigurable. To

implement an organization using Reo language, we

propose a formal model to specify the organization of

multi-agent systems. Then we use Reo to implement

the specification of the organization. Applying the

formal specification makes the implementation process

by Reo language systematically. We can use the formal

semantics of Reo to evaluate the properties of the

organization and overall system performance, security,

flow of information, etc.

Structure of the paper: The organization metaphor is

described in Section 2. In Section 3 we explain the Reo

concepts and in Section 4, we explain our formal model

for an organization. In Section 5, we explain how to

implement an organization by Reo language. In Section,

6,we specify and implement an example system using

our formal approach. Finally in Section 7, we explain our

concluding remarks and future works.

2 ORGANIZATION

In the traditional design of concurrent and distributed

systems, the architecture is derived from the

decomposition of functionalities and the data required by

system to achieve its goals as well as the definition of

their inter-dependencies. However, using organizational

concepts to design such systems, leads to a number of

agents, each with specific roles in the system. In this

model, agents interact to accomplish their tasks and,

agents embed most of the functionalities they need, so

the interactions of agents are reduced which makes the

design less complex and easier to manage. Most MASs

are intended to support or control some real-world

organizations. In such cases, an organizational-based

MAS design reduces the conceptual distance between

the software system and the real world system it has to

support.

An organizational structure defines the specific class

(among the many possibilities) of organization and

control regime to which the agents/roles have to conform

in order for the whole MAS to work efficiently and

according to its specified requirements (Zambonelli et

al., 2000). These organizational structures are usually

described in terms of a variety of social and

organizational concepts such as norm, trust, power,

delegation of task, responsibilities, permission, access to

resources and communication (Dastani et al., 2005). The

organization structure defines admissible actions of agent

interactions. For instance when there is delegation

relation between two agents, one agent can delegate task

to another agent. So, the delegating agent has a

delegation action in its interaction protocol.

Organizational rules express general, global (supra-

role) requirement for the proper instantiation and

execution of a MAS (Zambonelli et al., 2000; DeLoach,

2002). These rules indicate some constraints between

two communicating agents or an agent and organization.

3 REO CONCEPTS

Reo is a channel-based exogenous coordination

language based on the calculus of channels (Arbab,

2004; Arbab and Rutten, 2003; Arbab, 2003). Reo

consists of components that are connected via complex

coordinators, called connectors or networks, which

coordinate their activities. Connectors are

compositionally built out of simpler ones. The simplest

connectors in Reo are a set of channels with well-

defined behaviour supplied by the users (Arbab, 2004).

Agents communicate with each other by means of I/O

operations they perform through the I/O interfaces of

the connectors. The connector imposes a specific

coordination pattern on agent actions without any

knowledge about their internal communications. A

channel has precisely two channel ends. There are two

types of channel ends: sink and source. A sink channel

end dispenses data out of its channel and a source

channel end accepts data into its channel.

A connector is a set of channels and channel

ends organized in a graph of nodes and edges.

Channels are joined together in a node, so, a node is

a construct which consists of a set of channel ends.

Reo provide two types of operations: topological

–ones that allow manipulation of connector topology

and IO – ones that allow input/output of data. Reo

enables components to connect and perform I/O on

the connector, namely read, take and write.

Topological operations are join and split, because of

space limitation we do not explain them here.

As we mentioned earlier, Reo has operational

semantics (Arbab, 2003). The semantics of a Reo

connector is defined by the composition of the semantics

of its channels and nodes. Users define the semantics of

channels and Reo defines the semantics of nodes.

Arbab has defined a set of complete channel

types (Arbab, 2004), namely Sync, Filter,

SyncDrain, LossySync, and FIFO-1. Figure 1 shows

the visual notation for these channels.

Figure 1: Visual notation for basic channels.

WEBIST 2006 - WEB INTERFACES AND APPLICATIONS

448

The “Exclusive Router 2” connector is shown in Fig 2.

This connector has one input and two output ports.

When a data is written on the input port, it can be read

only by one of the components reading from the output

ports.

Figure 2: Exclusive Router 2 connector.

We can use abstraction notation to hide the internal

structure of the connectors using a box and interface

ports on its border. The abstract notation of “Initially

Closed Valve” connector is shown in Figure 3. This

connector is initially closed which implies that when a

data is written on its input port ‘a’, it won’t flow through

the connector until a data is written on the ‘c’ port (by

the administrator) and the valve is opened.

Figure 3: Abstract notation of “Initially Closed Valve” connector.

The “Initially Opened Valve” connector is the same as

“Initially Closed Valve” but data can flow from producer

to consumer until administrator closes the valve.

4 FORMAL MODEL

As described in Section 1, an organization coordinates

agent interactions. Thus an organization can be viewed

as a coordination artifact that coordinates the behaviour

and interactions of agents in terms of long-term goals

of system. We consider organization as an open

system, where agents are self-interested and can enter

and leave organization dynamically.

A formal model has been proposed in (Omicini et al.,

2004) to specify an environment-based coordination

artifact. In the environment-based coordination model,

agents are coordinated via data existed in the

environment. Thus, agents do actions using operations

defined by the user interface of the artifact. When artifact

receives an action, it is responsible to execute the action,

and reify proper data to keep track of agent actions.

These data define the coordination status of the

coordination artifact. In this model, agents do not have a

direct communication and they communicate via data

reserved in the environment. The operating instructions

of the artifact define for each agent how to exploit

coordination service.

We have extended the formal model proposed in

(Omicini et al., 2004) with organizational concepts to

specify an organization. The usage interface defines

what actions an agent can do and the set of operating

instructions defines the interaction protocol between

agent and organization. The coordination behaviour

of the artifact defines how the organization

coordinates the interactions of agents. In our model,

agents can communicate directly, which may be

synchronous or asynchronous.

An organization is specified by a tuple 〈R, A, ψ,

α, β, ρ, δ, →

, γ〉. Some of these parameters are in

common with the model in (Omicini et al., 2004).

The set R defines the set of roles required within the

organization to reach its goals. The set A defines the

set of agents and the roles they play.

Agent-oriented methodologies such as Gaia

(Wooldridge et al., 2000) and Tropos (Giorgini et al.,

2004), specify an organization in terms of roles and their

interaction relation/structure, which are usually modelled

as interaction protocols. In (Grossi et al., 2005) three

relations are distinguished between roles, i.e., power,

control, and coordination. So, agents can interact by

delegating tasks to each other, passing information to

each other, or taking responsibility for each other. The

meta-variable ψ is the set of binary relations, which

defines the organizational structure of MAS in three

dimensions of control, power and information:

ψ::= {power(r,s), control(r,s), inform(r,s), r,s

∈

These relations are not limited and users can define

other (social) relations. For instance, the power relation

specifies the agent enacting role r delegates tasks to the

agent enacting role s. Note ψ is exploited to cross-check

the consistency between organizational structure, agent

interaction protocols and the coordination behaviour. For

example when a power relation exists between two roles,

the delegating agent is allowed to delegate a task and the

delegated agent should receive the task either

synchronously or asynchronously.

The meta-variable α ranges over the operations

allowed by the organization to the agents and it

defines the actions an agent can do/initiate. The

meta-variable β ranges over the perceptions of action

completion and it may contain some information

about the outcome of the action. Therefore, the set L

of interactions between agents and the organization,

ranged over by l, is defined by the syntax as follows:

SPECIFICATION AND IMPLEMENTATION OF MULTI-AGENT ORGANIZATIONS

449

l ::= id!α | id?β

The id!α represents an agent identifier id,

executes an action α, and id?β represents agent id

perceives the completion β for the action α.

The function ρ associates to each agent identifier id

the usage instruction I he is committed to follow in the

organization and it defines the admissible actions and

perceptions. Instructions can be defined by exploiting

typical process algebra operators, i.e. by the syntax:

I ::= 0 | !α | ?β | I+I | I;I | I||I

Where, 0 is the void instruction, !α is execution of

an action, ?β is perception of a completion, operator

“+” is used for choice between instructions, “;” for

sequential composition of instructions and “||” for

parallel composition of instructions. The definition can

be recursive. As an example, the definition I:= !α ; (?β

|| I) means that the agent is initially allowed to do an

action α and later, while it can do the whole protocol

again, doing another action of α, it can perceive the

completion of previous actions (i.e. β) of α.

The meta-variable δ ranges over the data reified

into the organization (like databases or temporary

containers) to possibly keep information of

organization. Agents may not communicate directly

with each other to coordinate their actions, thus they

reify data into the organization which then taken by

another agent to coordinate their behaviours. The

meta-variable σ ranges over the set of Σ of states of

the organization, which is defined as follows:

σ ::= 0 | δ | l | (σ || σ )

The operator || is characterized by the following rules:

σ||0

↑

σ , σ || σ'

↑

σ' || σ , σ || (σ' || σ'')

↑

(σ || σ') || σ''

Thus, each state σ is defined by the parallel

composition of elements δ and interactions l. The l is

used to represent the pending actions to be executed

and pending completions waiting to be perceived.

The state of organization is changed when an

interaction occurs and is modelled by the transition

relation →

⊆

Σ, representing the fact that a state σ

may eventually move to another σ', when a new pending

action has to be computed which typically causes a

change in the data reserved into the organization.

The meta-variable γ ranges over the first order

predicates to define the organizational rules using

prepositional logic. It is defined by the syntax as follows:

γ::= a |

γ | γ

∧

γ | γ

∨

Where “a” is the set of atomic propositions existed

in the organization. These rules usually define the

pre-conditions required for the interactions between

agents, or an agent and the organization.

The coordination behaviour of organization is

described by a transition system

〈

C, →, L

{τ}

〉

. C is

the set of configurations of the organization, which is

defined by the composition of ρ and σ shown by ρ

⊗

σ,

where the function ρ associate to each agent the

instruction it currently has to follow, and the σ defines

the current state the organization. The transition

function →

C is defined by below rules:

ασρσρ

!||][

)(

idIid

Iid

⊗⎯⎯→⎯⊗

⎯→⎯

Rule 1

σρβσρ

⊗⎯⎯→⎯⊗

⎯→⎯

][?||

)(

Iidid

Iid

Rule 2

σρσρ

σσ

′

⊗⎯→⎯⊗

′

⎯→⎯

Rule 3

The first rule defines an agent id can do/initiate an

action, if the ρ(id) allows this action and then this action

will reified in the state σ. The second rule defines the

completion β to action α; if this is reified into the state σ

and the ρ(id) allows perception of the completion. The

third rule is derived from the actual coordination task

inside the organization; when the →

defines changes in

the states of the organization, there is a silent change in the

system configurations, which is shown by a silent

transition (τ).

5 MAPPING OF THE FORMAL

MODEL TO REO

Organization is a coordination artifact that can be

implemented using Reo, an exogenous coordination

language as explained in Section 3. We apply the formal

model of organization explained in Section 4 to make the

implementation of organization by Reo circuits

systematically. In this section, we show how to implement

an organization given the tuple 〈R, A, ψ, α, β, ρ, δ, →

, γ〉.

The set of agent operations within an organization

is restricted to the operations that are allowed to the

agent by Reo on the connector interfaces: write, read

and take. Thus α is a proper subset of I/O operations

allowed by Reo for an organization. In Reo, an

operation is not started unless it can be completed, so

for these completions the β is trivial. These perceptions

that contain information should be defined by an extra

write and read operations that will be included in β.

An interaction connector is implemented for the

interaction protocol of each role and organization

connector is defined according to the transition relation.

The organizational rules are implemented by control

WEBIST 2006 - WEB INTERFACES AND APPLICATIONS

450

connector. When an agent enters into an organization, it

is committed to an interaction protocol, which can be

implemented by an interaction connector. Agents

initiate actions (read, take or write) via Reo circuits

which accepts an action if it is admissible, otherwise it

cannot be initiated. For the different interaction

protocol patterns commonly used (Zlatev et al., 2004),

we describe their corresponding Reo circuits. The Reo

circuit for the interaction protocol I:=(a;b);I is shown

in Figure 4. We call this circuit Sequencer The Reo

circuit for the interaction protocol I:=(a+b);I is shown

in Figure 5. We call it Choicer during the paper.

Figure 4: The Reo

circuit for I:=(a;b);I

protocol.

Figure 5: The Reo

circuit for I:=(a+b);I

protocol.

If an “Initially Open Valve” circuit placed on the

way of the output c, the protocol changes into the

I:=(a;b) protocol, which is shown in Figure 6.

Figure 6: The Reo circuit implements the I:=(a;b)

protocol.

The Choicer circuit has two parts, namely

competitors and choicer parts as shown in Figure 7.

Figure 7: Different parts of the Choicer circuit.

The Reo connector implementing parallel

composition is defined in (Ghassemi, 2006).The →

can

be implemented by Sync, SyncDrain and FIFO channels.

The FIFO channel has the ability to store data. Thus we

can use different capacities of FIFO to store data reified

into the organization. The δ meta-variables define the

outputs of agents that should be stored in a FIFO

channel. The →

defines when a FIFO gets full and

empty or which actions should be synchronized.

The meta-variable γ place some restrictions

between an agent interaction and the organization or

between agent interactions. They can be implemented

by control connectors, which are placed between agents

and interaction connectors or within organization

connector and control the flow of data in the

connectors. Control connectors are usually

implemented by “Initially Open Valve”, “Initially

Closed Valve” and SyncDrain channel (Ghassemi,

2006).

Reo is a compositional model and complex circuits

are composed of simpler ones, which enable us to define

interaction protocols compositionally. We can define

complex interaction protocols by composing simpler

ones. We define three functions over a protocol namely

first, last and tail. The first function returns actions that

initiate the protocol. The tail function returns the

protocol by omitting the first actions from the protocols.

The definition of first function is shown as follows:

⎪

⎩

⎪

⎨

⎧

∨+++

∪∈

≠

);...;;()(

)||...||||()...()()(

211

2121

IIIIfirst

IIIIIIIfirst

Ifirst

βα

The last function defines the actions should be

done in order a protocol get finished. The definition

of last function is shown as follows:

⎪

⎩

⎪

⎨

⎧

∨+++

∪∈

≠

);...;;()((

)||...||||()...())(()(

2121

IIIItaillast

IIIIIIItaillast

Ilast

βα

To define the composition of protocols under an

operator we specify the Compose(I

,…,I

,op) function

for different op operations. Thus the Reo circuit for

Compose(I

,…,I

,+) is implemented by implementing

the Reo circuit of Compose(first(I

),first(I

),…,first(I

),+)

where Compose(a,b, …,+) is implemented using

Choicer-n circuit. This circuit is easily implemented by

connecting competitor parts to the choicer part of a

Choicer circuit The Reo circuit for Compose(I

,…,I

,;)

is implemented by implementing the

Compose(last(I

),first(I

),;) to Compose(last(I

),first(I

),;), where Compose(a,b,;) is implemented by

Sequencer circuit.

The transition relation defines how the agent actions

should be coordinated with each other. In this step,

designer can decide to coordinate two actions

synchronously or asynchronously. For example when a

manager delegates a task to his employer, the send and

receive actions of manager and employer should be

SPECIFICATION AND IMPLEMENTATION OF MULTI-AGENT ORGANIZATIONS

451

coordinated. If they are synchronous, the appropriate ports

of their interaction protocols are connected directly to each

other; otherwise, a FIFO-1 channel is used to keep the

delegation, which is later taken by the employer.

Each agent that enacts a set of roles needs a set of

Reo circuits implementing the interaction protocols

required for each role.

6 EXAMPLE

In this section we specify and implement an online

store, where seller and buyer interact with each

other. The buyer can ask about the price of items and

pay money for an item and get the item. In return the

seller answers client questions and receives money.

The seller should opens the store before any client

can enter and does any interaction.

6.1 Formal Specification

The online store is an organization where the two roles,

seller and buyer exist (R). Thus agents playing role

within the store are seller and buyer each have roles of

seller and buyer respectively (A). There is a flow of the

information from the seller to the buyer in order to

inform the buyer about the price of items and there is

also a flow of money from the buyer to seller. Thus the

organizational structure is defined as follows:

ψ::= {inform(seller, buyer),pay(buyer, seller)}

The actions that the buyer and the seller agents

can do in the organization are shown as follows:

α ::= answer| ask(item) |get_money |pay_money(amount)| open

β::=OK

answer

|OK

ask

(price)|receive_money(amount)|OK

pay

(item)| Ok

open

The interaction protocols (ρ) for each agent is

defined according to the organizational structure as

follows:

Seller:=((!answer;?OK

answer

)+(!get_money;?

receive_money(a))+ open);Seller

Buyer := ((ask(item);?OK

ask

(price))+(pay_money(a);

?OK

pay

(item))) ;Buyer

The δ contains “start” which indicates that the

seller has opened the store. The transition relation

→

is defined by the two simple rules:

seller

!open → Start || idSeller?open

Start || id

seller

!answer(item) || id

buyer

!ask(item) → Start ||

seller

?OK

answer

|| id

Buyer

?OK

ask

(a)

Start || id

Seller

!get_money || id

Buyer

!pay_money(a) → Start ||

Seller

?receive_money(a) || id

Buyer

?OK

pay

(item)

We can consider an organizational rule that when

a buyer can ask or pay money if the seller opens the

store. This rule is defined in the following:

Start ƒ(id

Buyer

!askϖid

Buyer

!pay)

In this formal specification, we can check the

correctness of interaction protocols and the

coordination behaviour of the organization

according to the organizational structure.

6.2 Reo Implementation

In this section we specify and implement online

store according to the formal specification and

mapping rules explained in Section 5. The α and β

are redefined as follows:

α ::= read

answer

(item)| write

ask

(item) | read

get-money

(amount) |

write

pay

(amount)

β ::= read

ask

(price) | read

pay

(item)

The interaction protocol for agents is redefined

as follows:

Seller:=((read

get-qu

(item);write

answer

(price))+(read

get-

money

(amount);write

give-item

(item))+write

open

);Seller

Buyer:=((write

ask

(item);read

answer

(price))+(write

pay

(amount)

; read

get-item

(item)));Buyer

The interaction connectors for each agent are defined

according to the interaction protocols and the interaction

connectors are connected to each other according to the

transition relation (organization connector). The

implementation of the store is shown in Figure 8.

Figure 8: Implementation of online store by Reo.

WEBIST 2006 - WEB INTERFACES AND APPLICATIONS

452

7 CONCLUSIONS

In this paper, we propose a formal model for the

specification of multi-agent organizations. This

specification formally defines what tasks an agent is

allowed to do in an organization and in

synchronization with which actions in the system. It

also defines what the pre-condition of each task is

and how the organizational structure affects on the

interactions between agent and the coordination

behaviour of organization.

We apply Reo coordination language to

implement organizations. To make the

implementation systematically, we use our formal

model to specify the system according to the Reo

I/O operations. We define how to convert

specification to the implementation by introducing

some Reo circuit for common interaction protocols

and how they can be composed to make complex

protocols. Reo not only provides a formal

specification but also provide an implementation.

Thus the Reo circuits are executable. There is a tool

to run Reo circuits (Dave, 2005).

We are going to find a mapping between process

algebra expressions into the Reo circuit. This

mapping enables us to automate the conversion of

specification by our formal model to a Reo circuit.

REFERENCES

Arbab, F., 2003. Abstract behavior types: A foundation model

for computers and their composition. In Proceeding of the

First International Symposium on Formal Methods for

Components and Objects(FMCO 2002), vol. 2852, pp 33–

70.

Arbab, F., 2004. Reo: A channel-based coordination model for

component composition. Mathematical Structures in

Computer Science, 14:329 – 366.

Arbab, F. and Rutten, J., 2003. A coinductive calculus of

component connectors., Recent Trends in Algebraic

Development Techniques, Proceedings of 16th

International Workshop on Algebraic Development

Techniques (WADT 2002), volume 2755, pages 35–56.

Clarck, D., 2005. Reo Lite. Internal report, Software

Engineering Department, CWI.

Dastani, M., Arbab, F., and de Boer, F. S., 2005. Coordination

and composition in multi-agent systems. In4rd

International Joint Conference on Autonomous Agentsand

Multiagent Systems (AAMAS 2005), July 2005, Utrecht,

Netherlands.

DeLoach, S., 2002. Analysis and design of multi-agent systems

using hybrid coordination media. In proceedings of

Software Engineering in Multiagent Systems.

DeLoach, S. and Matson, E., 2004. An organization model for

designing adaptive multiagent systems. In The AAAI-04

Workshop on Agent Organizations: Theory and practice.

Ghassemi, F., 2006. Analysis and Design of Multi-agent

Systems using Reo. Master Thesis, Isfahan University.

Giorgini, P., Kolp, M., Mylopoulos, J., and Pistore, M., 2004.

The Tropos Methodology: An Overview. In Methodologies

and Software Engineering for Agent Systems, Kluwer.

Grossi, D., Dignum, F., Dastani, M., and Royakkers, L. M. M.,

2005. Foundations of organizational structures in

multiagent systems. In 4rd International Joint Conference

on Autonomous Agents and Multiagent Systems, pp 690–

697.

Omicini, A., Ricci, A., Viroli, M., Castelfranchi, C., and

Tummolini, L., 2004. Coordination artifacts: Environment-

based coordination for intelligent agents. In proceeding of

3rd International Joint Conference on Autonomous Agents

and Multiagent Systems (AAMAS 2004), pages 286–293.

Wooldridge, M. and Jennings, N., 2000. The Gaia

Methodology for agent-oriented analysis and design.

Autonomous Agents and Multi-Agent Systems, 3(3):285-

312.

Zambonelli, F., Jennings, N. R., and Wooldridge, M., 2000.

Organizational abstractions for the analysis and design of

multi-agent systems. In Agent-Oriented Software

Engineering, First International Workshop, pages 235–

251.

Zlatev, Z., Diakov, N., and Pokraev, S., 2004. Construction of

negotiation protocols for e-commerce applications. ACM

SIGecom Exchanges, 5(2):12-22.

SPECIFICATION AND IMPLEMENTATION OF MULTI-AGENT ORGANIZATIONS

453