ADAPTIVE INFORMATION PROCESSING IN A DOMAIN

ONTOLOGY USING RECURSIVE TRANSFER FUNCTIONS TO

DETERMINE THE NON-DETERMINISTIC VICINITY OF

INTELLIGENT AGENTS

Muthu Chithambara Jothi and Kabilan Giridharan

Sungard Software Solutions (India) Pvt.Ltd, Langford Road, Bangalore, India

Keywords: Ontology, semantic network, walk table, non-deterministic state automation, service entities, high cohesion

and low coupling, XMI representation of goals, association and generalization entities, non-deterministic

vicinity of agents, heuristic learning module of agents.

Abstract: An adaptive walk over a semantic network is possible by describing the domain information in the form of

ontology. Leveraging the relationships made between the domain entities by intelligent agents through non-

deterministic automation in an object model which is represented in the form of OWL or RDF resource, is

the theme of this paper. The successful adaptive walk over a semantic network is with the advent of

determining the vicinity of the intelligent agent based on analysing the current service it received, with the

prime goal it needs to achieve. The idea of recursive transfer functions is to make the agents travel in the

semantic network until the final goal is achieved. The recursive transfer functions takes the current service

received from an entity in the semantic network as its parameter and applies the light of the prime goal in

order to achieve the next set of possible states. The next set of possible states lie in the similar line of the

prime goal to be achieved. The adaptive walk over semantic network aids the agents to act as proxies for

human beings there by fulfilling the business needs for the human beings by travelling the vast network of

interconnected web resources.

1 INTRODUCTION

Adaptive information processing over a semantic

domain network will be a boon for a large number of

web information systems. With the advent of

Ontologies, and through the idea of leveraging the

object model relationships an adaptive walk over the

semantic network is discussed. The ability of the

intelligent agents to dwell upon the domain ontology

is discussed in the light of non-deterministic state

automation. The vicinity of next possible set of

states for an intelligent agent is decided by the

transfer function which analyses the current service

received from the semantic entity in the light of the

prime goal the agent needs to achieve.

2 KNOWLEDGE

REPRESENTATION OF A

DOMAIN

2.1 Business Process Execution in

Conventional Information Systems

Computerized applications which include both

enterprise information systems and web applications

assist the human beings in their respective business

process. A wealth management system or a hospital

management system is needed by the respective

business organizations to execute and manage their

business process at a faster rate and increase their

productivity. But the point to note here is it is still a

human being who sits in front of the system and

executes the business process. Using an intelligent

agent to perform this job is the scope of this paper.

315

Chithambara Jothi M. and Giridharan K. (2007).

ADAPTIVE INFORMATION PROCESSING IN A DOMAIN ONTOLOGY USING RECURSIVE TRANSFER FUNCTIONS TO DETERMINE THE

NON-DETERMINISTIC VICINITY OF INTELLIGENT AGENTS.

In Proceedings of the Ninth International Conference on Enterprise Information Systems - SAIC, pages 315-319

DOI: 10.5220/0002397603150319

 SciTePress

2.2 Intelligent Agent is not an

Aladdin’s Genie

Intelligent agents are nothing but software programs

which can dwell and feed on a domain network if

the semantics of the domain information is

understood. But, there exists a similarity between a

genie and an intelligent agent.

The high level goals are provided as inputs to an

intelligent agent and it becomes the responsibility of

the agent to travel through the network keeping the

goal in mind. So the onus now falls on the agents

when some surprises awaits while progressing the

network towards the goal. The important point to

note is that an intelligent agent does not write the

business rules in a domain, but it should be capable

of understanding the consequences of executing a

business rule in the light of the prime goal it needs to

fulfil for the human being.

2.3 Ontology based Representation of

Domain Information

In order for the agents to dwell upon a domain

network we need to express the domain information

in a way the agent can understand. But knowledge

representation had been a holy grail from the time of

trying to represent in terms of first order and second

order predicate calculus. But now the grail seems to

be over with the advent of ONTOLOGY. When

domain ontology is derived out of a loosely coupled

and highly cohesive object model then that

respective ontology can be used by the intelligent

agents to achieve the prime goals for the human

beings.

2.4 Semantics of Relationships in

Ontology

Ontology provides a way to express the kind of

relationship between the domain entities. Two

entities in a domain can be related through a

generalization relationship or an association

relationship or through an aggregation or even

through depends and a constraint relationship.

Relationships are not just associations between

domains entities which help the developer to decide

upon the cardinality and navigability. Throwing a

high volt light on the relationship we can understand

that the relationships actually explain the semantics

of the entities in the domain. For example consider

the Figure 1. (Note: The same information

represented in UML can be represented in OWL).

Each ticket needs to be paid a price. Price can be

paid either through cash or through credit card or

through debit card. Here the payment generalization

gives the semantics of the payment business process.

The payment can be either made through cash or

through credit or a debit card. That is if an intelligent

agent can understand that if the there is no money in

the debit then it can use the credit since it is also a

type of payment it is nothing but a program which

makes a decision making process based on the

current service it received.

Figure 1: UML representation of Payment module.

3 MODELING THE VICINITY OF

AN INTELLIGENT AGENT

3.1 Goal Description through XMI in a

Semantic Network

Any business domain can be modelled and

represented in the form of semantic network through

ontology. The requirements that are filtered out as

the primary use cases can be considered as primary

goals of the concerned domain. So each primary

goal has sequence of steps to be completed before

the final goal is achieved. The intermediate steps are

represented as steps of the use case during the

inception and analysis phase. When the same use

case is represented in the form of a XMI it stands as

a goal description repository for the agents.

Consider the semantic network described in the

figure below

Figure 2: Semantic Network of Online Ticket Purchase

System.

Ticket Price Pa

ment

Credit Debit

Tickets

Show

Movie

Complex

Theatre

Shops

ICEIS 2007 - International Conference on Enterprise Information Systems

316

The above semantic network is a derived out of a

online ticket booking system for a movie. Each

entity in the semantic network is capable of offering

a service in the domain of online ticket purchase.

The type of relationship existing between the entities

is described through the ontology written in the form

of OWL.

The primary use cases of the domain gets filtered

out into a finite set of goals over the respective

semantic network. Let us define,

F = {G1, G2, G3…Gn}

Where F is the set of finite goals possible in the

network

3.2 Meta Model of Human Mind in

Goal Progression

In this section we actually analyze the working of

the human mind before actually modelling the

intelligent agent.

Consider a human being who serves as an agent

for booking tickets in the above mentioned FIG.1

Booking tickets for the respective show and for the

respective movie is the primary goal of that guy. The

person has a map in his mind for achieving this goal.

He/she starts walking in the network as per the map

defines the successful ticket booking process.

The path taken by the guy gets disrupted

whenever he/she encounters a negative service form

one of the entities present in the path. For example

when the person receives a service from the entity

Ticket as “No tickets available for the requested

show” it is considered as a negative service in the

light of the prime goal to be achieved.

Whenever a negative service is encountered the

walk towards the goal gets altered in a non

deterministic manner. For example when the

negative service is received from the ticket entity the

mind brings alternative maps for “Another movie”,

or “Dating out with a girl friend” which actually lie

in the line of the prime goal which is to relax out the

mind.

3.3 Non Deterministic Vicinity of the

Intelligent Agent based on the

Current Service Received

Analogous to the human mind described above each

intelligent agent that dwells upon the semantic

network has a walk table. Each entry of the walk

table corresponds to the path that has to be taken for

each goal which is actually retrieved from the XMI

definition as described.

Figure 3: Goal path definition for a Semantic network.

For example the walk table of an intelligent

agent for the semantic network in FIG is as follows

Table 1: Walk Table for an agent goal repository.

Prime Goals Prime Walk Map

G1 A – B – D – G – H

G2 C – H – J – K – P

In case of an intelligent agent the vicinity of an

intelligent is determined in a non deterministic

manner with respect to the next set of semantic

entities. The intelligent agent travels in the semantic

network until the final goal is achieved.

The recursive transfer function that enables the

agent to travel in the semantic network until the final

goal is achieved can be described as follows,

T (f) = {Walk stops, Prime goal = achieved

SL [T (Q X ∑)], Prime Goal! = achieved

In the above recursive transfer function Q is the set

of all semantic entities capable of providing a

service in the semantic network.

Q = {q1, q2, q3 …. qn} where q1, q2 are the

semantic service entities present in the network.

∑ = {Type of service received, Light of prime goal}.

Q * ∑ is the transfer function which decides the

possible set of next states.

ADAPTIVE INFORMATION PROCESSING IN A DOMAIN ONTOLOGY USING RECURSIVE TRANSFER

FUNCTIONS TO DETERMINE THE NON-DETERMINISTIC VICINITY OF INTELLIGENT AGENTS

317

Figure 4: Non-deterministic vicinity of an agent based on

current service.

The transfer function T(Q * ∑ ) is actually used by

the heuristic module of the intelligent agent to filter

out the most possible successful set of semantic sets

based on the walk it had made on the semantic

network. This heuristic function module of the

intelligent agent is represented by SL (next set of

possible states) where

Next set of possible states = Q * ∑

3.4 Importance of the Heuristic

Learning Function in an Intelligent

Agent

The role of the heuristic function in an intelligent

agent can be compared to that of an experienced

person who had been working in that domain for a

larger period of time and knows the nooks and

corners of the domain.

The situation is analogous to an experienced

human agent who goes and tries out booking a ticket

in theatre which is at the city outskirts when the

tickets are filled for the theatre which is present

inside the city.

The intelligent agent also pays a similar role.

Over acting as proxy for a respective human being

over a respective semantic network the agent would

be able to know about the behaviour of human mind.

So whenever a negative service is received from a

Semantic entity like “Ticket” it would rather execute

the walk path for “Booking a table for dinner with

his girl friend” rather then trying for booking for

some tickets in some other movie.

3.5 Association and Generalization

Relationship in Determining the

Intelligent Agent’s Vicinity

The vicinity of the intelligent agent is determined

based on the type of the current service it received

from the semantic entity. When the type of service

received is positive in the light of the prime goal

then the probability of association entities is higher

than the probability of generalization entities.

In the above network when the service received

from the entity “Payment” is positive (i.e. cash has

been paid for ticket) then the next possible set of

entities includes “ShowTime”, “Parking Space”

which are having an association relationship with the

current entity “Ticket”.

Similarly when the service received from the

entity payment is negative (i.e. enough money is not

available in the debit) then the next possible set of

entities may include other modes of payments like

“Credit” and “Coupons” which are actually

belonging to the super type “Payment”.

The probability of non deterministic vicinity of

the agent can be summarized as follows,

Probability (association entities) > Probability

(generalization entities) When the current service

received is positive in the light of the Prime goal.

Probability (generalization entities) > Probability

(association entities) When the current service

received is negative in the light of the Prime goal.

3.6 Modelling of Services Offered by

Entities in a Semantic Network

Services offered by the entities in the semantic

network are highly cohesive with the responsibilities

the entities are entitled with. The core of the service

oriented architecture decouples the tight linkage

between the entities.

When the entities are loosely coupled but highly

cohesive the service offered by the entity becomes

obvious by their signature definition. Also

techniques like CRC cards can be used to decide

upon the high level responsibility of the entity from

a business point of view.

The basic idea is to create a loosely coupled and

highly cohesive object model which is capable

exposing its services and getting it leveraged by the

intelligent agents.

Non-deterministic vicinity based on

negative service

Non-deterministic vicinity based

on positive service

Non-deterministic vicinity based

on negative service

W E

ICEIS 2007 - International Conference on Enterprise Information Systems

318

4 CONCLUSIONS

The aim of this paper is to implement the vision of

adaptive information processing towards the

business goals with the advent of ontological

engineering. With the criteria of a service oriented

architecture, the type of service returned back by a

semantic entity is compared in the light of the prime

goal with which the intelligent agent is progressing

in the network. The recursive transfer functions gets

agent travelling in the network until the final goal is

achieved.

ACKNOWLEDGEMENTS

We would like to express our sincere thanks to our

colleague Shrilakshmi M. Shivaswamy for her

constant support in authoring this paper.

Also we would like to thank Anusha Bharathi for

helping out in producing the diagrams electronically.

REFERENCES

Grigoris Antoniou and Frank Van Harmelen.,2004 A

Semantic Web Primer, The MIT Press Cambridge,

Massachusetts.

Michael Sisper., 1997. Introduction to theory of

Computation, Thomson Brooks/Cole. Singapore 1st

reprint.

Larry Kerschberg, Mizan Chowdhury, Alberto Damiano,

Hanjo Jeong, Scott Mitchell, Jingwei Si, and Stephen

Smith, Knowledge Sifter: Agent-Based Ontology-

Driven Search over Heterogeneous Databases using

Semantic Web Services Paper.

Joakim Eriksson, Niclas Finne, and Sverker Janson in

Paper To Each and Everyone an Agent: Augmenting

Web-Based Commerce with Agents, Intelligent

Systems Laboratory, Swedish Institute of Computer

Science

Raghu Arunachalam, Norman Sadeh, Joakim Eriksson,

Niclas Finne and Sverker Janson., 2003 in paper The

Trading Agent Competition Supply Chain

Management Game, Institute for Software Research

International School of Computer Science, Carnegie

Mellon University

Joakim Eriksson, Niclas Finne, and Sverker Janson., 1998

in paper SICS MarketSpace – An Agent-Based Market

Infrastructure from Intelligent Systems Laboratory

Swedish Institute of Computer Science

Jeff Z. Pan and Ian Horrocks, in paper Web Ontology

Reasoning with Datatype Groups, Department of

Computer Science,University of Manchester, UK

ADAPTIVE INFORMATION PROCESSING IN A DOMAIN ONTOLOGY USING RECURSIVE TRANSFER

FUNCTIONS TO DETERMINE THE NON-DETERMINISTIC VICINITY OF INTELLIGENT AGENTS

319