A SOA-BASED MULTI-AGENT APPLICATION FOR THE

CROSSLINGUAL COMMUNICATION PROBLEM

Hércules Antonio do Prado

Embrapa Food Technology, Catholic University of Brasília, Brazil

Aluizio Haendchen Filho

Anglo-Americano - Faculdades Anglo-Americano, Rio de Janeiro, RJ, Brazil

Míriam Sayão

Pontifical University Catholic of Rio Grande do Sul, Brazil

Fénelon do Nascimento Neto

Embrapa Food Technology, Brazil

Keywords: Crosslingual Communication, Multi-Agent Systems, Communities Mediation, SOA.

Abstract: We present a multi-agent system (MAS) approach to deal with the communication mediation among com-

munities with disparate levels of language. Service-oriented architecture (SOA) is adopted as the basis for

designing our proposal that has double purpose: (a) to present an alternative to the crosslingual communica-

tion problem and (b) to study an agent organization under the SOA specifications. A case is presented in

which the communication among food consumers and experts in food quality/safety is carried out by means

of agents’ organizations. We specify two basic sets of elements: application agents, responsible for the logi-

cal information, and components that are in charge of persistency, interface, and communication among the

software artifacts.

1 INTRODUCTION

Most of information systems currently developed is

focused in Web applications. The Internet has

opened a vast field of distributed applications that

have changed definitely the enterprise environment,

allowing an intense level of interaction among play-

ers that no one has ever realized. The ubiquity and

the distributed and interconnected characteristics of

Internet represent a natural field for multi-agent sys-

tems (MAS). Multi-agent systems present properties

as autonomy and pro-activity that make them inter-

esting for Web applications. The importance of In-

ternet has led to efforts like MIX (Mediation of In-

formation using XML) (Baru et al., 1999), that uses

agents to integrate distributed information in dispa-

rate sources. The Web also can be seen as a big dis-

tributed database having XML (and its extensions or

modifications) as an underlying data model.

A class of problems that can strongly benefit from

these advances refers to the crosslingual communi-

cation, information retrieval, and computer-mediated

dialogue. These problems involve information ex-

change among communities with disparate language

levels and require solutions that preserve the accu-

racy of the exchanged information.

The crosslingual mediation problem is discussed by

Piwek (2006), focusing the banking domain that

requires an extremely high accuracy information

exchange among the players. The problem ap-

proached here also refers to the information ex-

change in a context demanding a high level accu-

racy, in which distributed players are looking for a

consensus regarding to a knowledge piece. The ideal

251

Antonio do Prado H., Haendchen Filho A., Sayão M. and do Nascimento Neto F. (2008).

A SOA-BASED MULTI-AGENT APPLICATION FOR THE CROSSLINGUAL COMMUNICATION PROBLEM.

In Proceedings of the Tenth International Conference on Enterprise Information Systems - SAIC, pages 251-254

DOI: 10.5220/0001716002510254

 SciTePress

solution for this problem is an automatic process to

translate one language to other, being able to cope

with the different semantic levels of each specific

language. A solution to this problem is much more

important considering the challenge to human-to-

human communication through a browser. However,

to build an effective automatic translator for this

scenery is a dream far to be realized. So, the neces-

sity of communication among communities is a real-

ity that requires, if not the best solution, at least one

viable solution. The solution we present uses the

agents’ technology to facilitate the negotiation

among the communities’ members.

We argue that it is a problem that can be naturally

approached with the MAS technology. Agents can

perform the roles required in mediation process. The

described case study refers to a consumer that is in

need for information regarding to a product. This

information must be supported by scientific evi-

dence and is generated by experts in the involved

domain.

2 CASE DESCRIPTION

Our case study has as context a situation in which a

food consumer needs to clarify a question regarding

to a product. This consumer can access the system in

any place (in home, in a supermarket, in a school,

etc). It may need information about food that is not

in the respective label or is unclear. We developed a

solution for this problem in which (a) the answer for

a consumer query may be available in a knowledge

base and can be directly accessed by the system or

(b) it is necessary to send the query to an expert do-

main network, able to respond it. The general con-

text for this system is shown in Figure 1.

Figure 1: Application context.

There is in this structure a set of consumers, a num-

ber of experts’ networks, a knowledge base and the

MAS system that is responsible for the action coor-

dination among the players. PA

, PA

, .., PA

repre-

sent the customers requiring specialized information

in non-technical language. E

, E

, …, E

represent

the experts that can provide technical answers to the

customers questions. The triangle represents a set of

communication specialists that are in charge of

translating the technical language from the experts to

the customer language. In order to enable its reuse,

the answers are stored in a knowledge base. Notice

that the information required by consumers are those

not supplied by the product labels. For instance, a

consumer may be interested in knowing about the

effects on human health of a cereal produced from a

genetically modified grain, or the difference between

organic to hydroponics lettuce. In this case, a media-

tor agent

will be responsible for building a consen-

sus among the specialists addressed to provide a

sound technical answer to the customer. This answer

will be translated by the communication team, sent

back to the consumer, and stored in the answers

server. The interrelation among the players and the

functionalities carried out are shown in Figure 2.

Figure 2: Interaction among the players.

The consumer question is generated during its inter-

action with the system. A query manager searches

for an answer in the knowledge base. If the answer is

found, the manager sends it to the consumer and

finishes the process. If the question is not in the

knowledge base, the manager starts an answering

cycle.

ICEIS 2008 - International Conference on Enterprise Information Systems

252

3 THE MAS DESIGN

The MAS design was made using MIDAS (Haend-

chen Filho et al 2007), a MAS development frame-

work specified under the service-oriented architec-

ture (SOA). The platform is based on the coexis-

tence of several containers, each one executing a

JVM (Java Virtual Machine). It provides a complete

execution environment in where agents can execute

concurrently in the same host and communicate via

web services with external heterogeneous platforms.

The mechanisms embedded in MIDAS

work into two abstraction levels: (1) in the generic

architectural level, providing infrastructure services,

such as message transport (send/receive,

pack/unpack, requests translate), the control over the

platform and agents’ lifecycle, tasks for services

handling (services registry, publication and discov-

ery), and (2) in the agents’ design level, providing

abstract classes that define the hot-spots from which

specific features of the concrete agents (or compo-

nents) can be implemented, wrapper components

encapsulating Web Services and a blackboard to

support the agents’ communication model. The

blackboard (Ferber 2000) is a software entity widely

used in symbolic cognitive multi-agent systems. Its

structure is defined to enable agents interact indi-

rectly by sharing data and perceiving environmental

changes.

At the design level, the following application

agents must be developed:

Publisher: responsible for formatting and publishing

answers to the consumers, identifying the different

GUI devices to display the information.

1. QueryMgt: plays the roles related to requests

receiving, proceeding to the matching between

the user question and the expert profiles, look-

ing for the best set of experts able to answer the

question.

2. DomainExpert: created for each expert involved

in the question answer, this agent is responsible

to support the human expert in managing her/is

answering process, maintaining all information

relevant (e.g., the timetable to generate an an-

swer and the experts´ names working together

to find a consensus) to the process.

3. Consensus: its main role is to analyze the an-

swers from the DomainExpert and control a co-

ordination cycle until finding a consensus

among the domain experts.

Beyond application agents, some components

must be developed: The AnswerSearch component

encapsulates the web services getAnswerInKB that

retrieve information about knowledge base and

searchAnswerer that retrieves a researcher able to

reply the consumer question. Other components,

such as data access objects and graphical user inter-

face devices must also be implemented.

4 EXPERT/CUSTOMER

MEDIATION

The application developed is named BeyondLabel

(BL). The agents in the presentation and logical

layer (Publisher, QueryMgt, DomainExpert, Con-

sensus) interact with MIDAS (SOM, Blackboard).

The sequence diagram in Figure 3 shows the actions

carried out when a service request is sent from a user

to obtain any product information. The players

weakly shadowed represent application agents,

while the darker ones are components of MIDAS.

The sequence of actions in BL starts when a

service request is sent from a user. This action is

performed in a GUI device and directed to the SOM

agent. Following a typical SOA model, the agents do

not communicate directly among them. The Que-

ryMgt agent receives the service request via SOM,

verifying first if an answer already exists in the KB.

If so, the agent requests the responseDirect service

that is forwarded by SOM to the Publisher agent and

sent to publication. If there is no answer in the KB,

the QueryMgt agent requests the queryDirect service

that is forwarded by SOM to the DomainExpert

agent that will proceed a matching between the user

question and the expert profiles, looking for the best

set of experts able to answer the question.

One DomainExpert agent is instantiated for

each expert involved in the question answer. After

receiving the technical response from the expert that

may be a human or a software agent, the Domain-

Expert analyses the responses. If no divergence re-

mains regarding to the answers, it is driven to SOM

and forwarded to publication. If the DomainExpert

agent detects any divergence, the message is written

in the Blackboard.

The Consensus agent monitors the Blackboard

and, when a message to him is there, it is recovered

and the agent processes this message until finding a

consensus. In case a consensus is not possible, the

conflict information is sent to the administrator that

will forward the problem to a set of experts in that

domain that will discuss deeper the question. From

this point, all interactions among the experts fall out

of the system.

A SOA-BASED MULTI-AGENT APPLICATION FOR THE CROSSLINGUAL COMMUNICATION PROBLEM

253

Figure 3: The sequence diagram for the expert/customer mediation.

5 FINAL REMARKS

In this paper the mediation among communities with

disparate levels of language was approached by

means of a SOA-based multi-agent application. Us-

ing a low coupling structure, the development be-

comes easier, since the application agents can be

developed without concerns with other specific

agents. Furthermore, the location transparence pro-

vided by SOA simplifies the development, making

easier to perform local and remote services invoca-

tion. The application benefits from multi-agent ap-

proach since the human behavior of searching for

consent on polemic issues can be easily mapped into

agents. Considering that, even for simple domains,

the automatic translation of different jargons is a far

to be achieved reality, the solution involving human

actors sounds to be an interesting one. So, we have

modeled the solution by applying the MAS ap-

proach, controlling automatically the process while

assuring the answer accuracy.

The ongoing work includes to scale the application

to a broader set of consumers, involving some su-

permarkets networks.

REFERENCES

Baru, C.; Gupta, A.; Ludaescher, B.; Marciano, R..; Papa-

konstantinou, Y.; Velikhov, P. XML-based Informa-

tion Medidation with MIX. Proceedings of the Exhibi-

tions Program of ACM SIGMOD, 1999, 25-34.

Ferber, J. Multi-Agent Systems: An Introduction to Dis-

tributed Artificial Intelligence. Addison Wesley,

USA, February 2000.

Haendchen Filho, A. A Middleware Framework for Buid-

ing Multi-Agent Systems in the Internet. Ph.D. Thesis,

Pontifícia Universidade Católica do Rio de Janeiro

PUC-Rio, Rio de Janeiro, RJ, Brasil, 2005.

Klusch, M. (Ed.). Preface of “Intelligent Information

Agents.” Springer Verlag, Berlin, 1999.

Piwek, P.; Power, R. CROCODIAL: Crosslingual Com-

puter-mediated Dialogue. In Proceedings of the 8

In-

ternational Conference on Enterprise Information Sys-

tems, Paphos, Cyprus, 2006, 75-84.

ICEIS 2008 - International Conference on Enterprise Information Systems

254