AGENTS FOR MANAGING BUSINESS-TO-BUSINESS

INTERACTIONS

Software Agents for Managing Business-to-Business Collaborations

Edgar Tello-Leal

Universidad Aut

onoma de Tamaulipas, Matamoros entre J.B. Tijerina y C. Col

on, 87000, Victoria, Tamaulipas, Mexico

Omar Chiotti

INGAR-CONICET, Avellaneda 3657, S3002GJC, Santa Fe, Argentina

Pablo D. Villarreal

CIDISI, Universidad Tecnol

ogica Nacional-FRSF, Lavaise 610, S3004EWB, Santa Fe, Argentina

Keywords:

Software agents, Business-to-Business, Collaborative business process, Model-driven architecture.

Abstract:

Current market opportunities and the growth of new Internet technologies encourage organizations to dynam-

ically establish Business-to-Business (B2B) collaborations. B2B interactions are carried out by executing

collaborative business processes among the parties. In this work we propose B2B collaboration agents for

managing B2B interactions that allow organizations to dynamically establish collaborations and execute col-

laborative processes with their partners. The planning and execution of the actions of the agents that execute

collaborative processes are driven by a Petri Net engine embedded in these agents. The role an organiza-

tion fulﬁlls in a collaborative process is represented by a high-level Petri Net model which is used to drive

the behavior of the B2B collaboration agents representing the organization. Moreover, interaction protocols

representing collaborative processes are executed by these agents without the need for protocols deﬁned at

design-time. Finally, an implementation of the B2B agents is presented.

1 INTRODUCTION

The growth of new Internet technologies has led to

the opportunity of much greater connectivity among

organizations. Then, organizations are focusing on

the setting up of Business-to-Business collaborations

with their business partners in order to manage inter-

organizational collaborations and improve their per-

formance and competitiveness. A B2B collaboration

entails a process-oriented integration among hetero-

geneous and autonomous organizations which must

be achieved at a business level and at a technological

level (Villarreal et al., 2007b). Inter-organizational

collaborations are carried out through the execution

of collaborative business processes. A collabora-

tive business process deﬁnes the global view of the

interactions among enterprises to achieve common

business goals (Villarreal et al., 2007b; Roser and

Bauer, 2005). The design and implementation of

collaborative business processes (CBPs) implies new

challenges, such as participants autonomy, decentral-

ized management, peer-to-peer interactions, negotia-

tion, and alignment between the business solution and

the technological solution (Villarreal et al., 2006b;

Weske, 2007). Therefore, the software applications

must be developed that can interoperate effectively in

this new distributed, heterogeneous, and sometimes

unreliable environment.

Software agent technology is seen as a poten-

tially robust and scalable approach to meet this chal-

lenge. Since the features of software agents such as

autonomy, heterogeneity, decentralization, coordina-

tion and social interactions are also desirable for or-

ganizations involved in B2B collaborations (Villarreal

et al., 2007b), for which the use of this technology

can be considered as appropriate to be used in this

domain. The use of software agents to execute CBPs

helps to improve process integration, interoperability,

238

Tello-Leal E., Chiotti O. and D. Villarreal P..

AGENTS FOR MANAGING BUSINESS-TO-BUSINESS INTERACTIONS - Software Agents for Managing Business-to-Business Collaborations.

DOI: 10.5220/0003180002380244

In Proceedings of the 3rd International Conference on Agents and Artiﬁcial Intelligence (ICAART-2011), pages 238-244

ISBN: 978-989-8425-41-6

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

reusability and adaptability (Trappey et al., 2009; Zin-

nikus et al., 2008; Guo et al., 2005).

Therefore, in this work we propose B2B collabo-

ration agents for managing B2B interactions that al-

low organizations to establish dynamic collaborations

and execute collaborative processes with their part-

ners. Interaction protocols are better intuitive mod-

els of how agents will interact through a message-

based communication. The interaction protocols

representing collaborative processes are executed by

these agents without the need for protocols deﬁned

at design-time. The planning and execution of the

actions of these agents are driven by a Petri Net en-

gine embedded in them. The role an organization

fulﬁlls in a collaborative process is represented by

a high-level Petri Net model which is used to drive

the behavior of the B2B collaboration agent repre-

senting the organization. Our proposal is consistent

and integrated with the model-driven development ap-

proach for CBPs and B2B information systems pro-

posed in (Villarreal et al., 2006a; Villarreal et al.,

2007b). The Petri Net models of the agents’ behav-

iors are derived from collaborative process models

which describe business interaction protocols deﬁned

with the UML Proﬁle for Collaborative Business Pro-

cesses based on Interaction Protocols (UPColBPIP)

language (Villarreal et al., 2006a; Villarreal et al.,

2007b).

The paper is organized as follows. Section 2 de-

scribes the architecture and agents that compose the

proposed B2B collaboration approach. Section 3 de-

scribes the MDA-based method for generate Petri Net

models. Section 4 describes the implementation of

B2B collaboration agents. Section 5 presents conclu-

sions and future work.

2 B2B COLLABORATION BASED

ON SOFTWARE AGENTS

In this section we propose a B2B Collaboration

Agents architecture in order to enable organizations

to establish collaborations in a dynamic way and exe-

cute CBPs. Since the communication among software

agents is based on interaction protocols, a straight-

forward implementation of UPColBPIP collaborative

process models based on interaction protocols is pro-

vided by this approach. Two main functionalities are

supported in the B2B collaboration agents approach.

The ﬁrst functionality is the management of dynamic

agreements on collaborations among organizations to

execute one or more CBPs. This implies that mech-

anisms for organizations can instantiate and execute

CBPs whose models are provided by an organization

or stored in a public repository. The second function-

ality refers to the capabilities to execute and monitor

CBPs that organizations agreed to carry out.

The B2B collaboration agents architecture con-

sists of two types of agents to support the above func-

tionalities:

• A Collaboration Administrator Agent (CAAgent),

which represents an organization or participant of

a B2B collaboration. It is responsible for estab-

lishing communications with other CAAgents in

order to agree, in a dynamic way, on the CBPs

to be executed, and responsible for instantiating

CPAgents, which execute CBPs (see Fig. 1).

• A Collaborative Process Agent (CPAgent), which

executes the role an organization fulﬁlls in a CBP,

and thus, it is the responsible for the jointly ex-

ecution of a CBP along with the other CPAgents

from the partners involved in the CBP. Thus, at a

speciﬁc moment or time, an organization can have

as much CPAgents as CBPs being executed on the

B2B collaboration (see Fig. 1).

Figure 1: B2B Collaboration Agents architecture.

2.1 Collaboration Administrator Agent

CAAgents are used to create dynamic collaboration

agreements and instantiate the CPAgents, which exe-

cute CBPs. When an organization wants to establish a

dynamic agreement with another organization to col-

laborate and execute a CBP, its CAAgent is used to

initiate a conversation with the CAAgent of the other

organization. This conversation among agents is car-

ried out by executing a predeﬁned CBP Request for

Collaboration. Figure 2 shows the UPColBPIP model

that represents the interaction protocol of this CBP.

The initiator role is performed by the CAAgent

of the organization that initiates the conversation. It

sends a request for collaboration to responder role,

which is performed by the CAAgent of the other or-

ganization and receives the request. This means that a

CAAgent can perform the initiator or responder role

according to the way that the organization engages in

a collaboration agreement.

AGENTS FOR MANAGING BUSINESS-TO-BUSINESS INTERACTIONS - Software Agents for Managing

Business-to-Business Collaborations

239

Figure 2: Request for Collaboration Interaction Protocol.

The request sent by the initiator conveys a Col-

laborationRequest document which contains the UP-

ColBPIP model of the CBP that the initiator wants

to execute with a responder. The responder evalu-

ates the request and responds with a successful agree

message or a refuse message. In the last case, it im-

plies that the responder does not want to establish a

dynamic collaboration agreement to execute the CBP,

and the protocol ﬁnishes. If the responder sends an

agree message, it waits for the process model that

contains the activities that the role is going to exe-

cute in the CBP. When the initiator receives the agree

message, it creates an instance of a CPAgent, and if

it is successful, it sends an inform message with the

responder’s process model. Otherwise, the initiator

sends a failure message indicating that it could not in-

stantiate its CPAgent.

Once the responder receives the process model, it

immediately instantiates a CPAgent, and if it is suc-

cessful, it sends an inform message indicating that the

collaboration was established, and the protocol ﬁn-

ishes. Otherwise, if an error occurs in the instantiation

of the CPAgent, the responder sends a failure mes-

sage indicating that the collaboration was not estab-

lished, and the protocol ﬁnishes. Thus, by executing

this protocol CAAgents establish a dynamic agree-

ment to execute a CBP. In case the CBP involves sev-

eral organizations, the initiator executes the Request

for Collaboration protocol for each organization that

can fulﬁll the roles of the CBP. The CAAgents can

also remove their CPAgents in case of the execution

of a CBP needs to be interrupted.

2.2 Collaboration Process Agent

CPAgents carry out the jointly and decentralized ex-

ecution of CBPs from an established dynamic agree-

ment (which was achieved by CAAgents) among the

organizations involved in the CBP. A CPAgent is re-

sponsible of performing the role an organization ful-

ﬁlls in a CBP. To perform the role of an organization, a

CPAgent takes a process model containing the behav-

ior of the organization’s role, which is known as inte-

gration process or abstract process model (Villarreal

et al., 2006a; Villarreal et al., 2007b). An integration

process contains the public and private activities and

logic that support the role the organization performs

in the CBP. Thus, a CBP is executed by the enact-

ment of the integration process of each organization

by means of the CPAgents of the organizations.

In order to interpret and execute an integration

process model by a CPAgent, this model is deﬁned in

terms of a High-Level Petri Net (PN) model, which is

derived from the UPColBPIP model of the CBP to be

executed. A CPAgent has an embedded process ma-

chine that interprets the PN model and executes the

organization’s role in a CBP (see Fig. 3). The activi-

ties or transitions deﬁned in a PN model represent the

actions (i.e. send a message to another agent, wait for

the reception of a message from another agent, exe-

cute an internal action) that a CPAgent has to perform

to support the message exchange deﬁned in the inter-

action protocol of the CBP that is executed, according

to the role that the agent performs in the CBP.

Therefore, the behavior of this agent is driven by

the process machine according to a PN model and

it invokes the behavior action manager component

which is responsible for planning and executing the

agent’s actions derived from the transitions of the PN

model. The sending or receiving of a message from

another agent is carried out by the communication

manager component of the CPAgent when the behav-

ior action manager executes a message sending or re-

ceiving action. Tokens in the execution of a PN model

contain the business documents that are exchanged

between the CPAgents. Thus, business documents

move through the organizations’ CPAgents when the

actions for the sending of messages are executed.

In this way, CPAgents only have generic actions

(such as send or receive) deﬁned at design-time and

they do not require to have the behavior of prede-

ﬁned interaction protocols implemented. The transi-

tions of the PN models execute actions. Such actions

are used to activate the behavior of the CPAgent ac-

cording to the type of the transition to execute. Differ-

ent types of transitions (such as send message, receive

message, invoke an internal process) were deﬁned to

plan and assign the behavior to be executed by the

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

240

Figure 3: Collaboration Process Agents executing a CBP.

Figure 4: Interaction protocol for Collaborative Demand Forecast scenario.

agent. Tokens maintain information about the process

in which they were generated (process ID) and busi-

ness documents which are sent, received, or generated

internally. The transitions are also allowed to have a

guard, which is a boolean expression. When a guard

is present, it must evaluate to true to enable the tran-

sition, otherwise the transition is disabled and cannot

occur (Jensen and Kristensen, 2009). Hence, a guard

puts an extra constraint on the enabling of a transi-

tion. Therefore, the concrete actions (speech acts and

business documents) of these agents are obtained at

run-time according to the logic deﬁned in PN models.

This enables CPAgents to execute any interaction pro-

tocol representing the CBP that organizations want to

execute as part of a dynamic collaboration agreement.

For each individual role there is a separate Petri

net. The collection of individual Petri nets associated

with all the relevant roles represents the entire interac-

tion protocol. Figure 4 shows the PN that represents

the interaction protocol that the CPAgents have to ex-

ecute the Collaborative Demand Forecast collabora-

tive process (Villarreal et al., 2010). The CPAgents

involved in this collaborative process are the customer

CPAgent and the supplier CPAgent. The interaction

protocol of the process supports a negotiation pro-

cess between a customer and a supplier to agree on

a demand forecast. The protocol begins with the cus-

tomer, who requests a demand forecast. The supplier

processes the request and may respond by accepting

or rejecting it, as it is indicated by the Xor control

ﬂow segment. If it is accepted, the supplier under-

takes to realize the required forecast; otherwise, the

process ﬁnishes with a business failure. If the sup-

plier accepts the request, the customer informs, in par-

allel, a sale forecast of its points of sales (POS) and

its planned sales, as it is indicated by the And control

AGENTS FOR MANAGING BUSINESS-TO-BUSINESS INTERACTIONS - Software Agents for Managing

Business-to-Business Collaborations

241

ﬂow segment. Finally, with this information, the sup-

plier generates a demand forecast and sends it to the

customer. Then, the process ends. Each agent in the

scenario described above is an instance of a CPAgent.

The CPAgent is capable of executing CBPs models.

In this case, we have two CPAgents that are capa-

ble of executing interaction protocols. Figure 4 shows

the Petri Nets representing the customer and supplier

roles in a Collaborative Demand Forecast. The dotted

arrows that emanate from the transitions of individual

Petri nets show the exchange of messages and busi-

ness documents.

3 MDD-BASED METHOD TO

GENERATE PN MODELS

A Model-Driven development approach for generat-

ing a technological solution for B2B Collaborations

should support the derivation of the software arti-

facts from technology-independent models. In pre-

vious work, we proposed a model-driven architec-

ture (MDA) (OMG, 2003) methodology that consists

of three phases: analysis and design of collabora-

tive processes, veriﬁcation of collaborative processes

and generation of B2B speciﬁcations (Villarreal et al.,

2006a). The ﬁrst phase consists in the modeling of

CBPs from a business perspective, i.e. using concepts

that are less bound to the implementation technology

and are closer to the B2B collaboration domain. The

UPColBPIP language (Villarreal et al., 2007b; Villar-

real et al., 2010) is used for modeling technology-

independent CBPs. The behavior of CBPs is repre-

sented through the deﬁnition of interaction protocols

from a business point of view.

The second phase consists in verifying the correct-

ness of CBPs deﬁned in a UPColBPIP model. To sup-

port this, the MDA-based method for generating Petri

Net speciﬁcations from a UPColBPIP model is ap-

plied (Villarreal et al., 2007a). Interaction protocols

are formalized, transformed and mapped into Colored

Petri Net (CPN)(Jensen and Kristensen, 2009) spec-

iﬁcations, which are then veriﬁed with CPN Tools.

Thus, the veriﬁcation of CBPs is carried out at an

early stage of the development, when most of the fun-

damental decisions of a B2B collaboration are carried

out, and previous to the generation of the technologi-

cal solution.

The third phase consists in selecting the target im-

plementation technology to generate the technolog-

ical solution and ﬁnal software artifacts from CBP

models.

In this work, we propose a MDD-based frame-

work to generate technological solutions based on

software agents in which process models are inter-

preted by software agents. To generate the PN model

that each CPAgent require to execute a CBP, a pro-

cess transformation approach is applied to derive the

PN models of the organizations from a UPColBPIP

model (see Fig. 3). Thus, the behavior required for

the agents to execute a CBP is generated in an agile

way and represented in the PN models, which are the

software artifacts that interprets the agents.

4 IMPLEMENTATION OF B2B

COLLABORATION AGENTS

In this section we describe the implementation of B2B

collaboration agents architecture (see Fig. 5). The

software agents were built by using the Java Agent

DEvelopment Framework (JADE) (Bellifemine et al.,

2007), which is a physical multi-agent development

framework which complies with FIPA speciﬁcations

and aims at simplifying the development and imple-

mentation of multi-agent systems (Bellifemine et al.,

2007). The proposed B2B collaboration agent is com-

posed of a message transport system (MTS) used for

communicating with other agents or agent platforms,

an agent management system (AMS) used for man-

aging the agent life cycles such as starting and stop-

ping, and a directory facilitator (DF) used for record-

ing the services provided by an agent. Therefore, the

CAAgent and CPAgents are executed on the JADE

platform and they use ACL (FIPA, 2002) messages to

communicate among them.

The process machine component of the CPAgent

was implemented by using the Java-based Petri Net

framework (JFern). JFern provides an object-oriented

Petri Net simulator (Nowostawski, 2003). It consists

of a lightweight Petri Net kernel, providing methods

to store and execute Petri Nets in realtime, and for

simulation. Furthermore, JFern supports XML based

persistent storage of Petri nets and markings. The to-

kens of PNs contain business documents, which are in

XML format and they are transported as parts of the

content of ACL messages that agents exchange.

The B2B collaboration agents architecture in-

cludes an administration tool based on Eclipse Rich

Client Platform. This tool has to be deployed in each

organization and provides the support to:

1. Instantiate the JADE platform,

2. Instantiate the CAAgent of the organization,

3. Search, select and retrieve the UPColBPIP and

Petri Net models stored in the local repository of

the organization; and

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

242

Figure 5: Agent-based architecture for B2B collaboration.

4. Administrate and monitor the CPAgents instanti-

ated.

5 RELATED WORK

There are several approaches that exploit the bene-

ﬁts of software agents in the execution of the collab-

orative business processes. In (Kahl et al., 2007), a

complete architecture for the design and agent-based

implementation of cross-organizational business pro-

cesses was deﬁned. The principles of MDA are ap-

plied in three levels: business, technical, and execu-

tion. An advantage is shown by the generating of

agents from a platform-independent model of a ser-

vice oriented architecture (PIM-SOA). In addition,

they consider the potentials of their execution offered

by agents in comparison to other techniques like WS-

BPEL engines.

Another approach (Zinnikus et al., 2008), supports

a rapid prototyping by combining a model-driven

framework for CBPs with an agent-based approach

for ﬂexible process execution, including a semantic

service discovery for ﬂexible service composition.

In these approaches the behavior and interaction

protocols of the agents are deﬁned in design and de-

velopment time. However, they do not consider the

generation of the agents’ behaviors at run-time, which

makes more ﬂexible the architecture of the agents for

executing CBPs and allows the establishing of dy-

namic B2B collaborations.

6 CONCLUSIONS

In this work we have proposed a B2B collaboration

agents architecture for managing B2B interactions

that allow organizations to dynamically establish col-

laborations, and carry out a decentralized execution

of collaborative business processes (CBPs) with their

partners. B2B collaboration agents architecture are

composed by CAAgents and CPAgents.

We apply High-Level Petri Net models to repre-

sent the required behavior of a CPAgent to perform

the role an organization fulﬁlls in a CBP. Petri Net

(PN) models are derived from UPColBPIP models of

CBPs. Therefore, the decentralized execution of a

CBP is achieved by the enactment of a PN model car-

ried out by each CPAgent representing the role of an

organization in the CBP.

In addition, due to the behavior of CPAgents is

driven by the Petri Net-based process machine, their

actions are generated in run-time to support the exe-

cution of any interaction protocol representing a CBP.

These concrete actions are composed of speech acts

and business documents, the concrete actions are ob-

tained at run-time according to the logic deﬁned in PN

models. This makes more ﬂexible the architecture of

agents for executing CBPs. This is different from the

traditional development of agents where actions and

interaction protocols that can be executed are before-

hand implemented and deﬁned in design-time.

Finally, an implementation of the B2B collabora-

tion agents architecture is based on the JADE agent

platform. A Petri Net simulator was used to imple-

ment the process machine of the CPAgent. Also, a

distributed B2B collaboration management tool based

on the Eclipse platform was developed that enables

organizations to manage their agents and their own

repository of CBP models.

Future work is about how establish an algorithm

and automatic mechanism for the CBP Request for

Collaboration, in which the CAAgent will determine

if agree or refuse to establish a dynamic agreement

with another organization to collaborate and execute

a CBP. Another future work is about the addition of

mechanisms and tools to discover CBP models either

in a public or organizations’ repositories. In the cur-

rent implementation of the CPAgent the private activ-

ities or processes of the organizations, which gener-

ate or processes the exchanged messages, are simu-

lated. However, a support for a real integration and

execution will be developed by integrating Web ser-

vice technology with agents.

REFERENCES

Bellifemine, F., Caire, G., and Greenwood, D. (2007). De-

veloping Multi-Agent Systems with JADE. Wiley.

FIPA (2002). FIPA Agent Communication speciﬁca-

tions deal with Agent Communication Language

(ACL). Available: http://www.ﬁpa.org/repository/

aclspecs.html.

AGENTS FOR MANAGING BUSINESS-TO-BUSINESS INTERACTIONS - Software Agents for Managing

Business-to-Business Collaborations

243

Guo, L., Robertson, D., and Chen-Burger, Y. (2005). A

novel approach for enacting the distributed business-

workﬂows using BPEL4WS on the multi-agent plat-

form. In The 2005 IEEE International Conference on

e-Business Engineering.

Jensen, K. and Kristensen, L. (2009). Coloured Petri

Nets, Modelling and Validation of Concurrent Sys-

tems. Springer Heidelberg, USA.

Kahl, T., Zinnikus, I., Roser, S., Hahn, C., Ziemann, J.,

uller, J. P., and Fischer, K. (2007). Architecture for

the design and agent-based implementation of cross-

organizational business processes. In Gonalves, R. J.,

uller, J. P., Mertins, K., and Zelm, M., editors, En-

terprise Interoperability II, pages 207–218. Springer

London.

Nowostawski, M. (2003). JFern - Java-based petri net

framework.

OMG (2003). MDA guide v1.0.1, 03-06-01.pdf. Available:

http://www.omg.org/cgi-bin/doc?omg/03-06-01.

Roser, S. and Bauer, B. (2005). A categorization of collab-

orative business process modeling techniques. In 7th

IEEE International Conference on E-Commerce Tech-

nology Workshops, pages 43–54.

Trappey, C. V., Trappey, A. J., Huang, C.-J., and Ku,

C. (2009). The design of a JADE-based au-

tonomous workﬂow management system for collabo-

rative SoC design. Expert Systems with Applications,

36(2):2659–2669.

Villarreal, P., Roa, J., Salomone, H., and Chiotti, O.

(2007a). Veriﬁcation of models in a MDA approach

for collaborative business processes. In 10th Ibero-

American Workshop of Requirements Engineering and

Software Environments.

Villarreal, P., Salomone, E., and Chiotti, O. (2006a). A

MDA-based development process for collaborative

business processes. In European Workshop on Mile-

stone, Models and Mappings for Model-Driven Archi-

tecture, 3M4MDA 2006.

Villarreal, P., Salomone, E., and Chiotti, O. (2006b). Trans-

forming collaborative business process models into

web services choreography speciﬁcations. In Lee, J.,

Shim, J., Lee, S., Bussler, C., and Shim, S., editors,

Data Engineering Issues in E-Commerce and Ser-

vices, Second International Workshop, DEECS 2006,

LNCS. Springer, Heidelberg.

Villarreal, P., Salomone, E., and Chiotti, O. (2007b). Mod-

eling and speciﬁcations of collaborative business pro-

cesses using a MDA approach and a UML proﬁle. In

Rittgen, P., editor, Enterprise Modeling and Comput-

ing with UML, pages 13–45. Idea Group Inc, USA.

Villarreal, P. D., Lazarte, I., Roa, J., and Chiotti, O. (2010).

A modeling approach for collaborative business pro-

cesses based on the up-colbpip language. In Aalst, W.,

Mylopoulos, J., Sadeh, N. M., Shaw, M. J., Szyperski,

C., Rinderle-Ma, S., Sadiq, S., and Leymann, F., edi-

tors, Business Process Management Workshops, vol-

ume 43 of LNBIP, pages 318–329. Springer Berlin

Heidelberg.

Weske, M. (2007). Business Process Management. Con-

cepts, Languages, Architectures. Springer, Heidel-

berg.

Zinnikus, I., Hahn, C., and Fischer, K. (2008). A model-

driven, agent-based approach for the integration of

services into a collaborative business process. In 7th

Int. Conf. on Autonomous Agents and Multiagent Sys-

tems (AAMAS 2008), pages 241–248.

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

244