AN ONTOLOGY-BASED ARCHITECTURE FOR MULTI-AGENT

SYSTEM ENVIRONMENT

Roberto Paiano, Anna Lisa Guido and Enrico Pulimeno

Dipartimento Ingegneria dell’Innovazione, SalentoUniversity, Via per Arnesano, 73100 Lecce, Italy

Keywords: Multi agent systems, ontology reasoning, domain independent, domain dependent.

Abstract: The increasing interest towards e-business systems, and thus the need of the companies to communicate

with other companies in an efficient and flexible way, brings to a new way of thinking about information

systems that open itself towards distributed systems where the exchange of information may happen. The

problem is in the way that this exchange can be made both from the technology point of view and from the

conceptual point of view. If from a side a recent technology for information exchange (DDS- Data

Distribution Service) can help us from the other one it seem interesting the use of ontology as representation

tool of application domain that must open itself to the communication. In this paper we present an

architecture that is oriented to a multi-agent communication that uses ontologies and DDS as information

exchange protocol. The architecture here presented must be repeat without difficulties on an e-business

system assuming that each agent of the proposed architecture is a company in an e-business system.

1 INTRODUCTION

An e-business system is a system where involved

companies communicate among them exchanging

data in as much as possible efficient and flexible

way with the goal to satisfy the final user. Two are

the main problems that rise in this area:

 Conceptual problem related to the definition of

a semantics of interchange of the data in order

to assure the communication of the

companies in heterogeneous environments;

 Technological problem related to the use of a

suitable data interchange system useful to

assure the communication between

heterogeneous systems.

Both the problems are difficult to be faced and

surely can be of great help the formal ontologies to

solve the first problem (conceptual problem) and,

relatively to the technological problem, it can come

in help the DDS technology (Data Distribution

Service) (DDS, 2001) standard OMG and surely

useful not only for synchronous comminication

among companies but also for the asyncronous one.

Very often, in fact, the companies don’t need

only information in real time: we can think, for

example, to a business process that needs to have a

data coming from another information system. The

data requested from the business processes in order

to go on with the execution are not immediately

available. It is necessary a mechanism that allows to

ask for the data and wait until the data is made

available by some other information system that can

provide it.

A base architecture useful to think about an e-

business system based on a multi-agent has been

presented in a research project lead by SSI (Space

Software Italia) and the Department of Engineering

Innovation of the University of Salento. The project

was founded by Apulia region. Starting from a

multi-agent system already made up from SSI in a

demining system where each agent is represented by

a robot, the goal of the project has been to enable

the communication between agents through a level

of intelligence made up by semantic web

technologies, in order to define a run-time all the

parameters to publish/subscribe depending on the

operational context where agents work.

Naturally, when in the multi-agent system the

parameters to publish/subscribe has been identify , it

is necessary to avoid that they are manually

published/subscribed in the DDS; it would be,

therefore, useful that the classes (the parameters are

published/subscribed in the DDS through classes)

that materially define these parameters inside the

DDS, are automatically produced.

101

Paiano R., Lisa Guido A. and Pulimeno E. (2008).

AN ONTOLOGY-BASED ARCHITECTURE FOR MULTI-AGENT SYSTEM ENVIRONMENT.

In Proceedings of the International Conference on e-Business, pages 101-106

DOI: 10.5220/0001913301010106

 SciTePress

The "intelligent" level, that it is important to add,

it has to operate so that to allow "to understand",

depending on the context, the parameters to

publish/subscribe. To add this level, the formal

ontologies are particularly useful. Not only the

ontology but also the technologies of reasoning that

allow from a side to define the context where the

system operates and from the other to individualize

the rules of action that each agent can take in order

to answer to a well defined event.

The comparison with this kind of system and the

e-business systems is immediate: every agent

represents every information system involved in the

information interchange. The "intelligent" level

allows subsequently the different information

systems to publish/subscribe in the DDS the data of

interest to an appropriate system of reasoning that

allows to identify the correct information.

In this paper we present a high-level architecture

designed in the research project: the architecture is

useful in order to allow the communication among

the involved agents. As it will be clearer

subsequently, the base idea will be that to make the

application domain independent in comparison to

the definition of the key elements that make possible

the communication (DDS). Besides particularly

interesting is the sharing of an only one knowledge

base among several agents of the system.

After having introduced in the section 2 the state

of the art to the multi-agent systems, in the section 3

will be introduced three alternatives considered as it

regards the positioning of the knowledge base on the

various agents that constitute the system. In the

section 4 we present the architecture of the multi-

agent system, finally, in the section 5, we present the

conclusions of the paper.

2 BACKGROUND

The e-business systems have not been thought and

developed, until now, according to a multi-agent

logic. It is useful, however, to present the state of the

art related to the multi-agent systems in order to

understand the problem list related to the

development of a system of this type.

In a multi-agent system we may speak about 4

main aspects:

 decisional aspects related to the agent: what

actions an agent undertakes depending on the

external environment;

 outsourcing of the execution: possibility that

more agents collaborate together, performing

elementary actions to complete a complex

task;

 interactions among agents: information

interchange among agents in distributed

environment;

 Evolution toward component more and more

endowed with autonomy: agents that in full

autonomy reach their own goals.

The multi-agent systems need a study of the

nature of the interactions. It results therefore

important the notions of collaboration and

cooperation.

Interesting, in this paper, to underline that for the

realization of a multi-agent architecture it is

necessary to have a layer of communication and a

layer of conceptual modelling.

 Communication Layer: currently the most

qualified is standard seems to be FIPA-ACL

(FIPA, 2002) (Agent Communication

Language) created from the Foundation for

Intelligent Physical Agents (FIPA). This

standard is founded on the linguistic action

theory, elaborated by John Searle (Searle,

1969). An important implementation of the

FIPA standard is the framework JADE

(http://jade.cselt.it) an open source platform

for peer-to-peer agent based communication

developed by Telecom Italia Lab.

 Conceptual Model Layer: it is very important

to identify the domain where agents operate

and it is important the dynamics of

interactions between agents.

Currently, particularly interesting within the

modelling of the multi agent systems it results:

 The BDI Model (Beliefs-Desires-Intentions)

(Chang-Hyun, Guobin et. Al, 1969): it

considers the agent environment belief, which

is the result of its knowledge and perceptions,

and a set of Desires. Intersecting these two

sets, we obtain a new set of intentions, which

can become actions.

 Tropos

(Bresciani, Perini et. Al, 2004) a

software development methodology founded

on concepts used to model early requirements.

In particular, the proposal adopts Eric Yu's

modelling framework, which offers the

notions of actor, goal and dependency, and

uses these as a foundation to model early and

late requirements, architectural and detailed

design. The language used in Tropos for the

conceptual modelling is formalized in a meta-

model described with a set of UML class

diagram.

In literature there are several papers that examine

the software engineering paradigm applied to the

multi-agents system, among them an article of Pratik

K. Biswas (Patrik, Biswas, 2007) describes

ICE-B 2008 - International Conference on e-Business

102

extensively the multi-agent system elements and

their correlations, confronting them with UML

paradigm that it reuses for their modelling.

Very interesting in a multi-agent system is the

ontological approach very useful to provide an

explicit and formal representation of the domain.

This representation is simple to realize and it is

simple to exchange between agents thanks the

ontological languages such OWL (W3C, 2004).

3 USE OF KNOWLEDGE BASE IN

MULTI-AGENT SYSTEM

An e-business system thought in the multi-agent

terms it will have, for the complexity of the

scenarios in which it will operate, an elevated degree

of hardiness and modularity. To reach this goal it is

important to make more independent possible the

various agents (nodes) of the system, that they need

a continuous interchange of data in order to reach

their own goal.

The first problem to face is that to add to a

multi-agent system a layer of "intelligence" that is

able to provide a good level of autonomy and facility

of updating the system, besides it provides the

possibility to describe the several events that are

verified inside the system and to which the agents

must answer. This level of intelligence is constituted

by an ontology (opportunely supported by a system

of rules) that it results particularly useful to

guarantee a complete understanding of the domain to

all the interested agents, and accordingly one swifter

change of his in case of changes or the insertion and

management of unexpected events.

In the use of ontology in the e-business

systems based in multi-agents it is had to analyze

with attention as it must be defined and above all if

this ontology has to be positioned only on a node

inside the system or must be distributes on different

nodes. We describe 3 hypotheses of work

individualized defining among them the most

suitable to the context in which we work.

3.1 Centralized Knowledge Base

The centralized system foresees that all the agents

that cooperate make reference to a knowledge base

centralized on only one agent. This system involves

that, all the information push through the central

node which stores them in the knowledge base and

elaborates them.

The advantages of this approach are:

 the system is simple to manage;

 presence of a supervision node from which it is

possible to access all the information and to

provide precise commands to the other agents;

The disadvantages are:

 The supervisor node results a critical Point of

Failure, in fact if the supervisor node had to

come less for some reason the knowledge base

would result unreachable from the other nodes

and, therefore, the overall system go down.

 Agents have little decision autonomy.

3.2 Knowledge Base Total Distribute

In a structure that introduces an high degree of

distribution, all the agents work in independent way

and the exchange of data it has the goal to make

possible the coordination among the agents. All the

agents are in communication exploiting a system of

connection reliable (the DDS) that limit the Point of

failure. At ontology level, for this scenario two

different solutions can be identified, one that

foresees to repeat the whole knowledge base of

domain on every agent and the other is to foresees

the distribution of the domain ontology so that to put

on every agent only the part of knowledge base that

interests the specific agent for the carrying out of

his/her own role in the system.

3.2.1 Knowledge Base Replied on the Agents

In this structure, in which knowledge base is repeat

on every agent every node of the system, knows the

whole domain.

The advantages of this approach are:

 Non-existence of a Point of Failure;

 The autonomy degree of the agent enhance

 All the agents have peer decisional ability

exploiting the knowledge base;

The disadvantages are:

 Every change on the knowledge base has to be

repeat on every agent;

 The system needs a good structure of

coordination;

 The complexity of each agent enhances.

3.2.2 Knowledge Base Distribute on the

Agents

Another solution foresees the distribution on the

several agents of the description of the domain that

must be decomposed in modules depending on the

specific criterions of competence. In other words, it

deals with decomposing in several functionalities the

ontology and to implement them on the several

agents based on the demands of these.

AN ONTOLOGY-BASED ARCHITECTURE FOR MULTI-AGENT SYSTEM ENVIRONMENT

103

The advantages of this approach are:

 Least redundancies of the data, every agent

has only the information of which it has need;

 Absence of a single Point of Failure, and

therefore greater independence of the robots;

 Possible changes to be brought to the

ontological structure of the data base would be

alone on the single agent and not on

everybody.

The disadvantages are:

 Planning is complex because each agent is

complex.

 Limited decisional ability for the agents.

3.3 The Selected Solution

The selected solution is to have a distributed system

with a knowledge base repeat on every agent of the

operational context. In this way the several nodes

that operate in the system will be fully autonomous

and able to develop his/her own task without

depending in some way on the other participants to

the mission, also communicating and exchanging

data, in continuous way, with the other agents.

4 LOGICAL ARCHITECTURE

OVERVIEW

Before introducing the logical architecture

conceived it is fundamental to clarify shortly the

operation of the middleware of communication

selected.

4.1 Use of the DDS as Tool for the

Communication among Agents

The DDS is a useful tool proposed by the OMG that

enable the communication within data-centric

systems. The system of communication is

asynchronous and it is based on the

publishing/subscribing protocol: when agent that

operates in the system has the necessity of a data, it

makes a subscribing of it pointing out the data of

which it has need; when some other agent of the

system makes the data available, it effects a

publishing of it. The agent that has previously

effected the subscribing is able, to this point, to get

the data of interest. The data publishing/subscribing

is encapsulated in a class and send to the DDS. Since

information may change depending on the particular

operational context, also the relative

publishing/subscribing class can vary depending on

the particular data to send to the DDS.

4.2 Goal in the Realization of the

Architecture

The idea that is at the base of the conceived

architecture is born from two fundamental requisite:

 Decouple the decisional aspect related to the

publishing/subscribing of well defined

information, from the technological aspect

tied up to the necessity to produce on the fly

the useful classes to publishing/subscribing

the parameters in the DDS

 To provide an high flexibility level to the

system in this way to allow an adaptation of

the same to the different operational scenarios

where the architecture could operate, also

without denying the consequential

potentialities from the existence of a

middleware able to guarantee the

communication among the various agents.

Having decided to use a semantic base replied on

every agent of the environment, we decided to

decouple the component of the knowledge base of

domain from that related to the management and

representation of the concepts that describe the

middleware of communication (DDS).

The proposed decoupling makes the system

independent from the particular operational context,

providing a flexible and easily adaptable structure.

The proposed architecture is in fig. 1.

Figure 1: Logical architecture.

First of all we observe the presence of two layers

one called "domain independent" and one called

"domain dependent". Within these two layers,

ICE-B 2008 - International Conference on e-Business

104

interconnected through a system of rules as will be

detailed after, the presence of two knowledge bases

is observed:

 Knowledge Base Domain Dependent: this

module represents the KB specific of a special

operational context.

 Knowledge Base Domain Independent: this

module represents several elements that

constitute the DDS.

We can observes, within the layer "domain

dependent" the presence of an "agent". It represents

a generic agent that will contain both the knowledge

base domain dependent and the knowledge base

domain independent and that will make operation of

publishing/subscribing of the information on the

DDS.

4.3 Ontology Domain Dependent

Module

The knowledge of the context, in which the agents

work, and of its dynamics, it allows to be able to

manage in the best way the events that can be

verified in it. The module has the goal to represent

the application domain in which the agents operate

and the possible semantics relationships existing

among the information that the various agents could

exchange. In this module it will be present,

therefore, this domain ontology that must have

realized from the expert of domain that, better of

everybody, it is able to define every aspect of it. In

the knowledge base it is possible to find the specific

information about the domain and their semantic

relationships as an example if the domain is an

information system that manage data about

environment, the concept in the knowledge base will

be the sensor used to obtain data or the area where

data will be obtained and these two concepts will be

related each other in order to provide the semantic

link. For each sensor it will be possible to define the

period of time when the data will be obtained and

the threshold values defined in the specific context.

These information will constitute a small part of the

overall domain ontology that describe the specific

information system.

4.4 Ontology Domain Independent

Module

In this module there is the ontology that represents

the specifications of the substratum of

communication chosen for this architecture. This

knowledge base reproduces, of fact, the OMG

specification for the DDS: this knowledge base will

be completed with the individuals, depending on the

specific necessities, from the information obtained

following inference, from the KB domain

dependent. For example, if it is important, for the

information system to publish/subscribe a well

specific data obtained by the sensor, for example the

temperature data, the information useful to make a

topic for the temperature will be moved from the

knowledge base domain dependent to the knowledge

base domain dependent.

4.5 Jena Reasoner Module

The Jena reasoner module have the task to realize,

through the application of ad-hoc SWRL rules

written by the business expert, the deductions on the

knowledge base that allow to intercept events that

must happen and that they require, eventually, of a

publishing/subscribing of information on the DDS

with the goal to complete one determined activity.

With an opportune analogy, every information

system within an e-business system will deduce

from the domain ontology what information to

publish/subscribe in order to communicate with the

other systems.

4.6 Rule Engine Module

The rule engine module will provide, through the

opportune rules defined in the SPARQL language, to

extract from the KB domain dependent the

information to publish/subscribe and will add

individuals to the ontological classes that represent,

instead, the DDS (KB domain independent).

4.7 Java Class Generator Module

Since the system of publish/subscribe of the DDS

founds him, of fact, on a system of publish/subscribe

of classes, the presence of this module is

fundamental because it able to produce, beginning

from the individuals of the KB domain independent,

the useful classes to make the operation of

publish/subscribe. The class draws the information

that the domain expert associates to every topics

(“topics” are the elements that able the

communication in the DDS) this information,

through the rule engine module, is repeat in the KB

domain dependent and from here used for the

creation of the relative classes.

4.8 Description of the Information

Flow

The information flow in the architecture can be so

defined:

AN ONTOLOGY-BASED ARCHITECTURE FOR MULTI-AGENT SYSTEM ENVIRONMENT

105

 The agent captures the information stored in

the knowledge base domain dependent;

 The Jena reasoner module makes the

appropriate deduction through the SWRL rule

engine;

 The RULE ENGINE module defines the

mapping rules between the two knowledge

base (domain dependent and domain

independent).

 The java classes generator generates the .java

file and the corresponding .class file;

 The .class file will be used to send information

through the DDS.

It is important to observe that the separation

between knowledge base domain dependent and

knowledge base domain dependent increases the

flexibility level of the system. Changing the context

of business in which the system operates mean to

define the knowledge base domain dependent and

the system of reasoning (module rule engine and

Jena reasoner) leaving the dynamics of

communication made up through the DDS.

5 CONCLUSIONS AND FUTURE

WORKS

In this paper we propose an architecture conceived

for the communication of a data centric multi-agent

systems that they use as middleware of

communication the DDS recently proposed by

OMG.

Particularly interesting is the analogy among the

data-centric systems and the multi-agent systems

applied to an e-business context. The analogy can be

done thinking about every information system

involved in the e-business environment as a single

agent of the architecture here presented. In this way

it will be possible to bring on a e-business system

the whole efficiency and the flexibility that the

proposed architecture introduces: the great

advantage obtainable is the separation between the

KB domain independent and the knowledge base

domain dependent, another advantage is the

possibility to produce in automatic the useful classes

to make the data publishing/subscribing of the data.

Naturally, repeating the architecture proposed on

an e-business system there are many aspects to

clarify that brings to several research ideas.

The first problem is surely tied to the semantic

interoperability: it is necessary to build a knowledge

base that describes the whole domain in which every

information system participates but as it is possible

to effect the transfer of the information proper of

every informative system (which those present in the

database) in format compatible with those described

in the ontology domain dependent? And what mean,

practically, to describe a domain?

Parallels to these problems, the problem of

reasoning is still open: few has been done in

international scientific community in this sense so

much that doesn't exist, until now, a standard

language universally recognized for realizing the

reasoning. In this architecture SWRL is used but this

doesn't exclude, in a next future, the use of a

language more efficient.

In every case, is interesting the idea at the base

of the present paper that consists of using in an e-

business system the formal ontologies and the

mechanisms of reasoning.

ACKNOWLEDGEMENTS

We would thank SSI (Space Software Italia)

company and Antonella Falzone for the tangible

support.

REFERENCES

Bresciani, P., Perini, A., Giorgini P., Giunchiglia, F.,

Mylopoulos, J. 2004 Tropos: An Agent-Oriented

Software Development Methodology. In Autonomous

Agents and Multi-Agent Sytems, 8, 203–236, 2004

Chang-Hyun, J., Guobin, C., James, C. (2004) A New

Approach to the BDI Agent-Based Modelling In 2004

ACM Symposium on Applied Computing

Data Distribution Service for Real-time Systems Version

1.2 OMG Available Specification formal/07-01-01

FIPA ACL Message Structure Specification 1996-2002

Pratik K. Biswas, Towards Agent-Oriented

Conceptualization and Implementation. 2007, IGI

Global,

Searle, J. 1969 Speech Acts: An Essay in the Philosophy

of Language, Cambridge University Press, 1969 ISBN

052109626X

W3C (February, 10 2004) “OWL Web Ontology language

Reference” W3C

ICE-B 2008 - International Conference on e-Business

106