EXPLOITATION OF ONTOLOGY-BASED RECOMMENDATION
SYSTEM WITH MULTI-AGENT SIMULATIONS
Martina Husáková
1
and Pavel Čech
2
1
Faculty of Informatics and Management, University of Hradec Králové, Hradec Králové, Czech Republic
2
Faculty of Military Health Sciences, University of Defense, Brno, Czech Republic
Keywords: Ontology, OWL, UML, SWRL, Multi-agent simulation.
Abstract: The paper investigates the usage of ontology engineering-based techniques for recommendation system
development. The potential of ontology containing SWRL rules is studied with relation to the generation of
scenarios (recommendations). These scenarios are the basis for the creation of multi-agent models and
simulations. The multi-agent systems offer dynamic view on implications of recommendations that are
suggested by the ontology-based recommendation system.
1 INTRODUCTION
The nature of response operation during the
biochemical incidents asks for a clear set of actions
that follow the generic goal of minimizing the
casualties and loss of protected assets. There are
strategic documents covering the response operation
in general but the specificity of agents, uniqueness,
and relative rareness of such incidents requires
response operation to be more tailored in detail
depending on the type of agent and its
characteristics. However, most of the knowledge
that needs to be processed is in the form of
declarative assertions and atoms of actions such as
simple rules resident in heads of experts. The aim of
the paper is to design automated support for decision
making during biochemical incidents. The
comprehensive framework for decision support
during emergency situation caused by biochemical
agent is proposed. This framework should cover the
whole decision making process and enable for
elicitation, translation, integration of knowledge and
imitation the decision making processes of human
expert. Traditionally, the expert system with
knowledge base and inference engine would be
deployed. However, we suggest to use ontology for
knowledge representation about biochemical
incidents and extend the expert system with
simulation engine to provide better justification for
recommendation. Thus, the recommended set of
actions will be assessed against the assumed impact
modelled in the simulation. The focus is on
technological aspects so that particular solution can
be designed and tested.
The structure of the paper consists of two parts.
First, the conceptual level of the framework is
outlined. Second, the technological and
implementation aspects of the solution are being
designed and discussed. The solutions will include
the integration of recommendation with simulation
for better justification of the recommended actions.
2 SOLUTION
The conceptual level is used to define basic
requirements and goals of the system. Also the
system architecture is being delineated and the
decomposition of the system into particular
subsystems together with interfaces is being
designed. The technological level deals with
possible technological platform. The particular
methods and techniques are determined and aligned
along the goals given in conceptual level. The
implementation level focuses on realizing the model
in a particular environment using particular
specification or language.
The conceptual level has already been described
in (Čech, 2011). The system was decomposed into
three subsystems:
subsystem modelling the incident,
subsystem for scenario generation,
433
Husáková M. and
ˇ
Cech P..
EXPLOITATION OF ONTOLOGY-BASED RECOMMENDATION SYSTEM WITH MULTI-AGENT SIMULATIONS.
DOI: 10.5220/0003659404330436
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2011), pages 433-436
ISBN: 978-989-8425-80-5
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
subsystem supporting the selection of action.
Subsystem modelling the incident is used for
capturing the domain knowledge and current data
about the incident. The input into this subsystem
consists of the domain knowledge that is
predominantly in the heads of expert on a particular
area being medicine, epidemiology, biology,
chemistry etc. There are also information in the form
of documents, books and similar resources that
contain already formalized knowledge, see “Fig.
1a”. The output of the subsystem that is at the same
time input into the subsystem for scenario generation
are the characteristics of agent together with relevant
characteristics of the environment that are important
relative to the agent.
Subsystem for scenario generation is based on an
inference engine and corresponding causality models
in form of rules, see “Fig. 1b”. Declarative domain
knowledge about the agent and current data about
the incident will be processed against the rules and
the recommended set of action or scenario of
response operations will be generated. It is necessary
to receive the feedback from the expert(s) for
ensuring the correctness of SWRL rules
representation, see “Fig. 1c”.
Figure 1: Ontology-based framework.
Subsystem supporting the selection of actions is
based on a simulation engine. The simulation takes
the characteristics of the agent and environment as
parameters and examines the impact of a particular
set of actions. The particular set of actions represents
alternatives that are to be considered. The resulting
estimations of casualties and impact on the protected
assets form the output of this subsystem, see “Fig.
1d”. Outputs of multi-agent simulations have to be
compared with the outputs presented the reality.
Expert is able to give us the information if the multi-
agent system behaves correctly or not, see “Fig. 1e”.
2.1 Ontology and Rules
The principles of ontological modelling are applied
in formalizing the captured knowledge. According to
Gruber (1993), ontology is regarded as “an explicit
specification of a conceptualization”. This definition
was slightly extended (Borst, 1997). Ontology is
“formal specification of shared conceptualization”.
Ontology should be formalized for machine-
processing and reflect consensus between people for
knowledge sharing.
There are several widely used languages for
ontology modelling. The OWL ontology, based on
the interviews with experts and relevant documents,
has been built during the course of the project. The
open-source Java-based environment Protégé 4.1.0.
has been used for the ontology modelling. Logic of
the ontology was verified with the aid of the open-
source Java-based Pellet reasoner in ver. 2.0.0.
Important classes with explanation of their meaning
are mentioned in “Tab. 1”.
In the course of the framework development also
the alternative languages were considered.
Especially, the attention was focused to UML
(Unified Modelling Language). UML is not
primarily oriented to modelling ontologies though;
there might be sound reasons to use UML also for
describing concepts and relations in a knowledge
base. UML is a de facto standard modelling
language in software industry. UML serves as
modelling tool for development of computer
information systems. The UML is designed to be
general enough so that it can serve also other
purposes than it was originally intended to. UML
offers specification together with graphical notation
for describing models.
Both OWL and UML are based on common
principles coming from object oriented thinking and
modelling. However, OWL conforms to the
principles of declarative programming paradigm, the
UML follows the imperative programming
paradigm. The declarative formalism is closer to the
nature of knowledge that was to be captured. The
imperative formalism was closer to what should be
recommended since the result of the
recommendation should be an imperative set of
actions. The major difference with respect to the
goal is that OWL enables to use existing inference
engines. In case of UML the inference engine has to
be implemented separately.
Currently, ontologies have been widely used in
semantic web applications in order to recommend
web pages or multimedia resources (Middleton,
2006); (Lops, 2011).
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
434
Table 1: Important classes of the ontology.
Classes Meaning
Agent This class represents a biochemical agent.
The agent is characterized by its
occurrence, reservoir, infectivity,
transmission, fatality, symptoms,
incubation period and prevention. The
particular agents are modelled as
individuals with specific values of given
characteristics represented as well as
individuals. The following individuals are
represented: Anthrax, Brucellosis,
Cholera, Glanders, Melioidosis, Bubonic
plague and Tularemia.
Environment This class describes the important
characteristics of the scene of the incident.
The environment is characterized by wind
speed, direction, temperature, humidity,
animal occurrence, density of population
in the area, and also number of infected
persons, number of infected animals,
number of dead persons, number of dead
animals, time from first symptom
observed, occurrence in public transport,
etc.
Response
Operation
This class describes particular response
operation mainly with its impact to
protected assets. The following individuals
are represented: Vaccination, Water
reservoir decontamination, Area
quarantine Animal kill off, Water supply,
Food supply, Insect repellent supply,
Protective mask supply, Army power
utilization, Soil reservoir decontamination,
Human quarantine, Animal quarantine,
Vaccine buying, Laboratory analysis of
sample, Air decontamination
Recommendation This class is going to be associated with
individuals of Response operation class.
The individuals associated will be inferred
based on the domain knowledge in form of
the rules.
Incident This class is used to associate Agent,
Environment and Recommendation class.
Individual belonging to this class would
represent a current incident and would be
associated with individuals of Agent class
that caused the incident, individual
describing the current environment setting
and would be linked to particular response
operations inferred as a recommendation
to tackle the incident.
Protected Asset This class represents the protected assets
that are threatened during the incident by
the agent. The protected assets are also
impact by recommended response
operations. Currently, there are three types
of subclasses and that is the tangible
property, intangible property or financial
assets of humans. Particular protected
assets will be represented as individuals
belonging to one of these subclasses.
The ontology describes user preferences and
particular items of interest and based on the
principles of content or collaborative filtering the
similarity was computed. In such cases the ontology
based inferences can be utilized since the description
of an item or user preferences can be enriched by
implicit classification based on the defined
properties and relations. However, there are some
limitations of ontology based reasoning. First, it
regards only classes and thus is not able to handle
individual. Modelling particular instances of certain
events and elements, however, better reflect the
reality. Second, it is not able to reason based on
expressed causality that is an evident part of
knowledge need during response operations. That is
why it was necessary to employ add into the
ontology another level of expressivity using rules.
OWL comes with extension including Horn-like
rules that is called SWRL (Semantic Web Rule
Language). There are six core classes in the paper,
see “Tab. 1”.
These classes with their individuals are basis for
the implementation of SWRL-based rules. These
rules represent knowledge of experts that were
elicited during interviews. SWRL editor of Protégé
4.1.0 tool was used. SWRL-based rules are the
inputs for the inference engine. We use the open-
Pellet reasoner in ver. 2.0.0. It is able to infer new
relations between classes and individuals or between
individuals only (Sirin, 2007).
2.2 Simulation
The subsystem for modelling the incident is linked
to the subsystem responsible for simulation. The
main goal of the simulation is to estimate the impact
of these actions to people and protected assets as the
time develops. The recommended set of actions
together with the description of the environment
represents the input into the simulation subsystem.
The simulation model is based on the domain
knowledge gained from experts and other resources
such as papers, reports, etc. In particular, data from
Committee on Toxicology (1997) and U. S.
Department of the Army (1990) were used for the
compiling of the document with chemical agent
characteristics (NBC, 2011). This document was
used in our ontology development.
Simulation is based on multi-agent technology.
Multi-agent simulation appropriately reflects the
emergency situation during biochemical incident in
which there are many heterogeneous elements
characterized by given properties and with its own
behaviour. Agents represent infected persons
(individuals), dangerous object (virus, bacteria, etc.)
as well as protected assets. Currently the model
simulating the spread of Anthrax in an environment
EXPLOITATION OF ONTOLOGY-BASED RECOMMENDATION SYSTEM WITH MULTI-AGENT SIMULATIONS
435
is prepared in order to test the technology and the
possibility of integration with the subsystem for
modelling the incident. The model reflects the
characteristics of the Anthrax and simulates the
spread in an environment based on the wind speed
and wind direction and the corresponding effect of
the contaminated cloud on population.
The subsystem for simulation is implemented in
the open source environment called NetLogo.
NetLogo enables to study systems that change in
time. The connection to subsystem modelling the
incident is realized using the NetLogo API
(Application Programming Interface).
Using the simulation the person responsible for
commanding the response operations can better
predict the outcomes of performed activities. The
simulations are useful also in situations in which
alternative set of actions is being recommended.
3 CONCLUSIONS
The response operation during emergency situation
caused by biochemical agent requires to process
huge amount of domain knowledge from various
areas. Ontology represents a possibility how to
capture domain knowledge and enable
recommendation inferred based on the rules
imitating expert decision making. Presented
ontology together with rule based reasoning and
simulation is a part of a proposed decision support
framework. The paper points to technical and
implementation aspects of the framework
development.
ACKNOWLEDGEMENTS
This paper was created with the support of the
research proposal Information support of crises
management in health care No. MO0FVZ0000604,
and the GAČR project SMEW, project num.
403/10/1310.
REFERENCES
Borst, W. N., 1997. Construction of Engineering
Ontologies for Knowledge Sharing and Reuse: Ph.D.
Dissertation, University of Twente.
Committee on Toxicology, National Research Council.
1997 Review of Acute Human-Toxicity Estimates for
Selected Chemical Warfare Agents. Washington,
D.C.: Nationa Academi Press.
Čech, P. et al., 2011. Ontological models and expert
systems in decision support of emergency situations in
Military Medical. Science. Letters, vol. 80, 2011, p.
21-27.
Gruber, T., 1993. A translation approach to portable
ontologies. Knowledge Acquisition, vol. 5, issue 2, pp.
199-220, 1993.
Lops, P., M. de Gemmis and G. Semeraro, 2011. Content-
based Recommender Systems: State of the Art and
Trends in Recommender Systems Handbook, Part 1,
73-105, 2011.
Middleton, S. E., D. De Roure and N. R. Shadbolt, 2009.
Ontology-Based Recommender Systems in Handbook
on Ontologies International Handbooks on
Information Systems, 2009, Part 6, 779-796.
NBC Biological and Chemical Agent Characteristics,
NBC Product and Service Handbook [Internet]
http://www.approvedgasmasks.com/BioChemAgentC
haracteristics.pdf [accessed 22.1.2011].
Sirin, E, 2007. Pellet: A Practical OWL-DL Reasoner.
Web Semantics: Science, Services and Agents on the
World Wide Web Vol. 5, Issue 2, June 2007, pp. 51-
53.
U.S., Department of the Army, Potential Military
Chemical/Biological Agnes and Compounds, U.S.
Army Field Manual 3-9, (NAVFAC P-467, AFR 355-
'7), 12 December 1990. Washington, D.C. Government
Printing Office.
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
436