BUILDING BRIEF ONTOLOGIES

A Case Study for Floods Management

Juli

an Garrido, Ignacio Requena

Department of Computer Science and Artiﬁcial Intelligence, Universidad de Granada

C/ Daniel Saucedo Aranda, 18071 Granada, Spain

Stefano Mambretti

Wessex Institute of Technology, Ashurst, Southampton, U.K.

Keywords:

Hazards, Floods, Risk management, Knowledge representation, Brief ontology.

Abstract:

This paper introduces the generation of brief ontologies as a mechanism to obtain a reduced version of the

original ontology. The new ontology includes the relevant knowledge for a given context and thus reduces

reasoning time in applications. In order to do so, an automatic selection of the concepts that are included in

the brief ontology is done. A case of study for ﬂood management is also presented, creating a brief ontology

that contains only knowledge related to ﬂoods from a generic ontology of environmental assessment.

1 INTRODUCTION

There have been different attempts of summarizing

monolithic semantic networks and ontologies i.e. a

methodology for partitioning a vocabulary hierarchy

into trees (Gu et al., 1999). This methodology re-

ﬁnes a IS-A hierarchy of medical entities according

to prescribed rules in a process carried out by a user

in conjunction with the computer. In order to reach

more simplicity, the methodology aims to create a set

of very small trees where each concept has only one

parent. However, this simpliﬁcation makes the model

unrealistic. Moreover, they use human evaluation to

study how comprehensive are the resulting trees.

A comparison and description of pruning methods

for bio-ontologies is done in (Kim et al., 2007). They

describe whether a pruning method is more suitable

and whether they should be avoided by showing their

beneﬁts and drawbacks for the different cases. In gen-

eral, these methods include two phases: i) the selec-

tion phase identiﬁes relevant elements according to

the user’s goals. ii) The pruning phase uses the selec-

tion to remove irrelevant elements. In particular, they

describe for each method: the ontology base that the

method uses, whether it supports integrity constraints,

the level of automation, the type of selection strategy

and the size of the ﬁnal ontology. Although they plan

to use metrics to assess the effectiveness of the meth-

ods, they also assess the methods showing the results

to a group of experts.

Another different approach divides large ontolo-

gies into modules using partitioning based on struc-

tural properties (Stuckenschmidt and Schlicht, 2009).

Its criterion consists of building modules where the

semantic connection between concepts in a module is

maximized whereas the dependencies with concepts

belonging to other modules are minimized. Firstly, it

creates a weighted dependency graph, it does a parti-

tioning and ﬁnally it optimizes the modules by isolat-

ing concepts, merging or duplicating concepts (even-

tually).

According to (Noy and Musen, 2009), an ontol-

ogy view is a portion of an ontology that is speciﬁed

as a query in some ontology-query language (analo-

gously to databases). However, they extend this def-

inition to ontologies that are deﬁned by a traversal

speciﬁcation (concepts, relationships and the maxi-

mum distance to traverse along each relationship) or

by meta-information. They also present a tool able

to accomplish management tasks such as comparing

ontologies, merging them together, maintaining and

comparing different versions. However, they declare

as an open issue pruning the deﬁnitions of the con-

cepts.

The concept of brief ontology was ﬁrstly intro-

duced in (Delgado et al., 2005). They deﬁne a

Garrido J., Requena I. and Mambretti S..

BUILDING BRIEF ONTOLOGIES - A Case Study for Floods Management.

DOI: 10.5220/0003630900280036

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2011), pages 28-36

ISBN: 978-989-8425-80-5

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

brief ontology as the ontology which includes a small

amount of knowledge referring to concepts existing in

more generic ontologies. They introduce this concept

to provide relevant access to information in databases

for a web services-based and multi-agent architecture.

Nonetheless, a formal deﬁnition of brief ontology is

included in Section 2.

Methodologies like Methontology (Fern

andez

et al., 1999) point out the convenience of reusing other

existing ontologies whenever is possible. However,

the whole ontology has to be imported even if only

a small fraction is relevant for the problem. For this

reason, the size of the new ontology may grow with

useless conceptualizations from other ontology. This

problem worsens increasingly as more ontologies are

imported to the new one.

In order to avoid this problem, brief ontologies

may be used to obtain reduced versions of the ontol-

ogy that the user wants to import. If these ontologies

contain only the portion of knowledge that the user

really needs then the size of the ontology will not in-

crease unnecessarily. By contrast, the objective might

be just isolating a portion of the knowledge modelled

on the ontology in order to use only the brief ontol-

ogy without unnecessary knowledge. Our approach

presents a traversal method to build brief ontologies

using not only concepts but also instances of concepts

(individuals) as starter point. Moreover, the method is

also fully compatible with pruning deﬁnitions of the

concepts.

As an example, a case of study for ﬂoods manage-

ment is presented because of the European normative

(Directive 2007/60/CE, 2004) that encourages to the

assessment of ﬂood risks in order to do an adequate

management of the problem. The starting point is

an ontology for environmental assessment and a brief

ontology for ﬂood management is created as a base of

a future knowledge-based system.

The paper is organized as follows, Section 2 in-

troduces the concept of brief ontology, Section 3 de-

scribes the procedure for generation of brief ontolo-

gies taking into account two different scenarios i.e.

the generation based on concepts and the generation

based on individuals, Section 4 describe the case of

study where a brief ontology for ﬂood management is

created. The ﬁnal sections give the conclusions and

list bibliography.

2 BRIEF ONTOLOGIES

According to (Baader et al., 2003), a typical DL (De-

scription Logic) knowledge base comprises the TBox

and the ABox. The TBox describes the intensional

knowledge (terminology) and it is represented with

declarations in order to describe general properties of

the concepts.

Operators of the DL knowledge base allow build-

ing the terminology and providing meaning to its dec-

larations. For instance, a concept may be deﬁned as

the intersection of other concepts. This type of sim-

ple deﬁnition allows deﬁning a concept in terms of

other previously deﬁned concepts. However, the set

of operators depends on the type of description logics

that the language implements (OWL-DL). This sub-

language implements the S H OI N (D) logic and it

has less expressivity than OWL in order to reduce the

computational complexity of reasoning and inferring

(Staab and Studer, 2009). It involves operators such

as: union, intersection, complement, one of, existen-

tial restriction, universal restriction and cardinality re-

striction.

The ABox contains the extensional knowledge

which is the knowledge that is speciﬁc to the indi-

viduals of the domain. It includes assertions about

individuals for example using properties or roles to

establish a relationship between individuals.

If an ontology is deﬁned as the union of its TBox

and ABox, a brief ontology is another ontology where

the extensional and intensional knowledge have been

restricted and modiﬁed in order to include only the

relevant knowledge for a given context. This is for-

mally described in the following deﬁnition where the

TBox is represented as O = (K

and the ABox is rep-

resented as K

). However, it is important to clarify

before that if two concepts have the same name in O

and O

, then they are referred as equivalent with inde-

pendence of their deﬁnitions. The concept of equiva-

lence has the same consideration for individuals and

roles.

Deﬁnition 1. For ontology O = (K

, K

), a brief

ontology is an ontology O

= (K

, K

) such that

v K

∧ K

v K

, and for every concept C ∈ K

and its equivalent C

∈ K

, the deﬁnition of C ex-

actly matches the deﬁnition of C

or the deﬁnition of

is a generalization of the deﬁnition of C. Analo-

gously, for every individual v ∈ K

and its equivalent

∈ K

, every assertion of v

exactly matches the as-

sertion of v or it is a generalization of the original as-

sertion of v.

In other words, the brief ontology is a pseudo-

copy of the original ontology that includes only a por-

tion of the knowledge base (a subset of the intensional

and extensional knowledge). It is referred as a portion

because not all the concepts, individuals and roles of

the original ontology are in the brief ontology. More-

over, it is considered a pseudo-copy because the deﬁ-

BUILDING BRIEF ONTOLOGIES - A Case Study for Floods Management

nition of the concepts may be modiﬁed or generalized,

and because some assertions of the individuals may

be also ignored or generalized (Garrido and Requena,

2011b).

The exclusion of elements of the original ontol-

ogy and the pseudo-copy are accomplished in order to

match with the restrictions for the brief ontology and

because of the brief ontology pretends to be a simpli-

ﬁed version of the original ontology.

3 PROCEDURE OF GENERATION

This section describes the extraction procedure of the

brief ontologies. However, some considerations must

be taken into account before describing the algorithm.

The brief ontology is built making a selective copy

of the original ontology. The goal is to obtain a

context-centered ontology where the context is con-

sidered the speciﬁcation of the user for the relevant

knowledge.

The extraction procedure is parametrized by this

user speciﬁcation because it is used as criteria to

spread a traversal exploration on the original ontol-

ogy and therefore it is used to decide whether an ele-

ment of the original ontology is relevant and must be

included in the brief ontology.

Traversal algorithms (Aho et al., 1983) are usu-

ally used in graphs theory to implement depth ﬁrst

or breadth ﬁrst searches. These algorithms start at

some node and then visit all the nodes that are reach-

able from the start node. If the graph is weighted

then the strength of the relationships between nodes

are usually deﬁned with matrix. Therefore, traversal

algorithms assist in the task of creating a sub-graph

because they establish an order to visit nodes in the

graph. Moreover, a threshold is useful to visit only

nodes that are strongly connected and thus restricting

the concept of reachable node.

Although an ontology is not a graph (Bizer and

Seaborne, 2004), a traversal exploration of an on-

tology implies analogously to consider two types of

nodes i.e. concepts and individuals. Moreover, these

elements are considered reachable whether there is

some kind of relationship between them. It may be

a parenthood relationship between concepts or a con-

cept and its individuals, relationships of a concept

with the concepts and individuals that are used in

its deﬁnition and relationships between individuals

that are represented in its assertions. Primitive val-

ues and datatype properties are not considered nodes

and connections between nodes, therefore, those data

are components of the node.

Whereas a threshold may be used to limit the

nodes that are going to be visited in a weighted graph,

other different mechanisms are used in ontologies. In

particular, a set of properties are speciﬁed to restrict

individuals and concepts that are visited during the

traversal exploration of the ontology (this set of prop-

erties is named relevant properties). If two nodes are

related with a property which is not a relevant prop-

erty then the second node will not be reachable by this

connection but it may be by another one.

If a property is relevant then the information that

it gives is interpreted as signiﬁcant for the purpose

of the user. For the same reason, if a property is not

in this set then all the information or semantic that it

provides must be ignored and not included in the brief

ontology.

The following subsections describe two different

methods to build the brief ontologies. The genera-

tion based on concepts should be considered if there

is more interest on extracting the taxonomy of con-

cepts rather than the individuals of the ontology. De-

pending on how the ontology is built, a brief ontology

where all the individuals have been rejected is pos-

sible. However, if there is special interest in these

individuals then the generation based on individuals

should be used.

3.1 Generation based on Concepts

Algorithm 1 describes the generic procedure to build

a brief ontology when the start set is compound of

concepts. Its inputs are the original ontology, the set

of main concepts (MC) where the traversal copy starts

and the set of relevant properties (RR) to restrict the

traversal exploration. The output is a new ontology

that contains the relevant knowledge of the original

ontology.

First of all, only the relevant properties (RR) will

be created in the brief ontology and the rest of prop-

erties of the original ontology are ignored. After this,

the traversal copy of concepts must be accomplished.

This task is done for every concept that belongs to the

set MC (ﬁrst loop).

The traversal exploration of concepts involves

spreading the algorithm to all the reachable nodes.

In this case, it spreads to concepts and individuals

by concept-concept, individual-individual, concept-

individual and individual-concept connections. How-

ever, only concepts are labeled with positive evalua-

tion to be created at this point. The reason is that the

complete taxonomy of concepts for the brief ontology

must be created before the creation of individuals, as-

sertions or concept deﬁnitions.

In the second loop, a traversal exploration of con-

cepts is started for each concept in MC. The next

KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development

nested loop starts a traversal exploration of individ-

uals for the ones that were reachable in the previous

exploration. The traversal exploration of individuals

is done following only individual-individual connec-

tions. All the individuals that are reached with the

set of relevant properties RR are created in the brief

ontology at this point.

All the concepts and individuals that have been in-

cluded in the brief ontology are visited in the third

loop. Firstly, a traversal exploration over the concepts

is done in order to create the deﬁnition of concepts.

Secondly, a traversal exploration over the individuals

allows deﬁning their assertions.

------------------------------------------------

Algorithm 1

Input: Ontology O, main concepts MC,

relevant properties RR

Output: brief ontology OB Begin

Create RR properties

Foreach concept C in MC Begin

Traversal exploration of concepts (RR,C)

Create concepts with positive evaluation

End For

Foreach concept C in MC Begin

Traversal exploration of concepts (RR,C)

Foreach individual v in the exploration

Traversal exploration of individuals (RR,v)

Create individuals with positive evaluation

End For

Foreach concept C in MC Begin

Traversal exploration of concepts (RR,C)

Foreach concept with positive evaluation

Create relevant definitions

End For

Foreach individual v in the exploration

Traversal exploration of individuals (RR,v)

Create relevant assertions in individuals

with positive evaluation

End For

End

------------------------------------------------

Concepts and individuals may be reachable from

different concepts and individuals and it may im-

ply several traversal explorations in the same steps.

Moreover, cycles may appear depending on the orig-

inal ontology. In order to solve this problems, if a

concept or individual has been computed in a itera-

tion of the traversal algorithm then it does not have to

be computed a second time in subsequent iterations.

For this reason, the complexity of a traversal ex-

ploration is lineal O(n) if the number of relevant prop-

erties and main concepts are limited by constants.

Hence, the efﬁciency of the algorithm corresponds to

O(n*m), being n the number of concepts and m the

number of individuals.

3.2 Generation based on Individuals

Algorithm 2 describes the generic procedure to build

a brief ontology when the start set is compound of

individuals. Its inputs are the original ontology, the

set of individuals (MI) to start the traversal copy and

the relevant properties (RR). The output is the brief

ontology with the relevant knowledge.

This algorithm also starts creating the relevant

properties in the brief ontology. Then, it continues

with a traversal exploration of individuals for each in-

dividual in MI (ﬁrst loop). The individuals cannot be

created until the class they belong exists in the brief

ontology. For this reason, the class of every individ-

ual found in the exploration is created in a nested loop

immediately before its respective individual. At this

point all the direct connections between individuals

that start in the MI are created. The next logical step

is to spread the algorithm with a traversal exploration

of all the concepts that were classes of the individu-

als. Hence, the complete taxonomy of concepts is in

the brief ontology once ﬁnished this step.

Although the major part of individuals is already

in the brief ontology, the individuals that are con-

nected to concepts by its deﬁnition may not have been

included. This requires a second loop where all the

concepts that are classes of the individuals (which

were found in the ﬁrst loop of the algorithm) are again

the starting point of a traversal exploration. Conse-

quently, new individuals may be found in the con-

cepts deﬁnition as a result of this exploration of con-

cepts. These individuals are also starting point of a

new traversal exploration of individuals and the new

ones will be created in the brief ontology.

After ﬁnishing this second loop the complete tax-

onomy of concepts and individuals is in the brief on-

tology. However, individuals and classes are created

empty at ﬁrst attempt and it requires a second step

to include its deﬁnitions and assertions. As a gen-

eral rule, concepts must be created before individuals,

these before deﬁnitions or assertions, and deﬁnitions

before assertions.

In the third step, the algorithm consists of two

nested traversal explorations of individuals (starting

in the set MI) and its concepts. The deﬁnitions of

the concepts are created at this moment. Nonetheless,

it is important to remark that the original deﬁnition

may be modiﬁed according to (Garrido and Requena,

2011b) due to some of the concepts or individuals of

the deﬁnition may no longer exist in the brief ontol-

ogy.

In the last step, a traversal exploration starts for

BUILDING BRIEF ONTOLOGIES - A Case Study for Floods Management

every individual of MI and the assertions are created

for every individual that is found during the explo-

ration. Exploring the classes of these concepts, new

individuals may be found and the algorithm ends cre-

ating the assertions for these individuals.

Although this algorithm increases the order of

complexity compared to the case that is based on con-

cepts, it has still polynomial efﬁciency.

-----------------------------------------------

Algorithm 2

Input: Ontology O, main individuals MI,

relevant properties RR

Output: brief ontology OB

Begin

Create RR properties

Foreach individual v in MI Begin

Traversal exploration of individuals (RR,v)

Foreach individual p with positive evaluation

Create concept C that is class of p

Create individual p

Traversal exploration of concepts (RR,C)

Foreach concept with positive evaluation

create concept

End

Foreach individual v in MI Begin

Traversal exploration of individuals (RR,v)

Foreach individual p with positive evaluation

Select C that is class of p

Traversal exploration of concepts (RR,C)

Foreach individual u in the exploration

Traversal exploration of individual(RR,u)

Create individuals with positive evaluation

End

Foreach individual v in MI Begin

Traversal exploration of individuals (RR,v)

Foreach individual p with positive evaluation

Select C that is class of p

Traversal exploration of concepts (RR,C)

Foreach concept with positive evaluation

Create definition

End

Foreach individual v in MI Begin

Traversal exploration of individuals (RR,v)

Foreach individual p with positive evaluation

Create assertions of p

Select C that is class of p

Traversal exploration of concepts (RR,C)

Foreach individual v in the exploration

Create assertions of v

End

-----------------------------------------------

4 CASE OF STUDY: FLOOD

MANAGEMENT

First of all, building a brief ontology from a detailed

one according to our needs is done by the genera-

tion process described in Section 3. Nonetheless, the

complete semi-automatic procedure to obtain a brief

model is detailed below.

1. Establish the aim and scope for the brief ontology.

2. Selection of a detailed ontology with knowledge

about the aim and scope.

3. Analysis and study of the detailed ontology.

4. Selection of the best type of extraction algorithm

for this ontology.

5. Selection of the starter point and relevant proper-

ties.

6. Generation of the brief ontology.

7. Evaluation of the resulting model.

4.1 Aim and Scope

Flood is a body of water which overﬂows its usual

boundaries over a land area with other land use, re-

sulting in adverse impacts. The socioeconomic devel-

opment in the ﬂoodplains and the reduction of the nat-

ural water retention by the land use increase the con-

sequences of ﬂoods. For this reason, a European Di-

rective (Directive 2007/60/CE, 2004) encourages the

ﬂood management and risk assessment. This manage-

ment requires in general prevention, protection and

mitigation actions (De Wrachien et al., 2011).

The main goal for this case of study is to build a

model for ﬂood management.

4.2 Detailed Ontology

The detailed ontology will be the environmental im-

pact assessment ontology that is originally described

in (Garrido and Requena, 2011a). This ontology was

built with two purposes. Firstly, in order to provide

and establish the conceptual framework of environ-

mental assessment (EA) and secondly, to facilitate the

development of methodologies and applications (Gar-

rido and Requena, 2010). Indeed, the ontology was

also born to be the knowledge base of an EA system.

KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development

Table 1: Set of relevant object properties.

ByMeansOf characterizeRainfall

characterizeRainfall dischargeAffectedBy

dischargeProducedBy ﬂoodProducedBy

hasCharacterizingIndicator hasDataSource

hasMitigatingAction hasPreventiveAction

hasRecoveryAction isCharacterizingIndicatorOf

isDataSourceOf isObteinedWith

isParameterOf isParametrizedBy

manageedBy produce

produceDischarge produceFlood

rainfallCharacterizedBy use&Need

4.3 Analysis of the Ontology

The EA ontology describes in essence the relation-

ships between industrial activities and environmen-

tal impacts considering for instance the environmental

indicators that should be controlled for every impact.

Although the ontology is focused on industrial ac-

tivities and human actions, it is also taking under con-

sideration natural processes and natural events as im-

pacting actions. A natural process or event is consid-

ered an impacting action whether they interact with

human activities and this interaction implies an incre-

ment of its environmental impact.

As a result of the inclusion of natural events, the

EA ontology contains knowledge about ﬂoods and it

allows using this detailed ontology as a base to extract

the relevant knowledge about our case of study.

A deeper description of the EA ontology is found

in (Garrido and Requena, 2011a). It contains a de-

scription for the taxonomy of concepts, the properties

and its justiﬁcation.

4.4 Selection of the Algorithm

The selection of the algorithm may depend on the type

of ontology and the portion of the ontology that the

user is interested in.

Some ontologies are built only as semantic mod-

els where no instances of concepts are stored. In this

case, the application of the generation based on con-

cepts is mandatory.

By contrast, other ontologies are used as knowl-

edge base with a high number of instances. In this

case, the user may be interested only in the seman-

tic model represented by the taxonomy of concepts

or the factual knowledge represented by the instances

of concepts. The ﬁrst case implies the utilization of

the generation based on concepts whereas the second

case requires the generation of brief ontologies based

on individuals.

In our case of study, the generation based on con-

cepts is chosen because of there is no special interest

on individuals and we are interested only in the se-

mantic model (taxonomy of concepts with their for-

mal deﬁnitions).

4.5 Parameters of the Algorithm

Because of the traversal algorithm based on concepts

has been chosen (Section 3.1), the parameters are a set

of starter concepts and a set of properties to traverse.

The selection of the starting point and the set of

relevant properties of the EA ontology require the

study and the analysis of the existing properties that

are used in the concept deﬁnitions, i.e. studying how

the concepts are related by these properties.

A property will be relevant in our domain depend-

ing on its semantics and its meaning. There are two

different cases: i) The property is speciﬁc for the tar-

geted domain. ii) The property has general use in dif-

ferent domains but it is used in concept deﬁnitions of

the targeted domain.

If the knowledge that the user have about the de-

tailed ontology is not enough to choose the set of rele-

vant properties, a heuristic for the selection of relevant

properties consist of the following steps. First, the set

of speciﬁc properties for the domain (ﬂoods) has to

be identiﬁed. Among them, a group of relevant prop-

erties is selected by studying its informal description,

domain or range in order to understand its semantics

and decide if it is relevant. Then, a temporary brief

ontology may be built with this set of relevant prop-

erties. The resulting ontology is studied to identify

new properties that are not speciﬁc in our domain but

they are considered also relevant for the concepts of

the brief ontology. Finally these properties are added

to the set of relevant properties.

For example, the property ﬂoodProducedBy has

its domain in the concept Flood and it allows deﬁn-

ing the causes of a ﬂood. This property is considered

relevant because it represents knowledge that we want

in our brief ontology for ﬂoods. The table 1 includes

the ﬁnal set of relevant object properties for our case

of study.

Regarding to the starting points or main concepts,

the user should try to ﬁnd some representative con-

cepts in the targeted domain that are not connected by

a traversal path with relevant properties. In our case

study, the selection of the concept FreshWaterFlood is

enough because of it is the best concept to represent

our targeted domain.

4.6 Generation of the Brief Ontology

The construction of the brief ontology for ﬂoods is

automatically carried out with the traversal algorithm

BUILDING BRIEF ONTOLOGIES - A Case Study for Floods Management

Figure 1: Schema of the brief ontology.

once it has the starting point for the algorithm and the

relevant properties.

As an example, the concept DDF

is subclass

of the concept RainfallStatisticalAnalysis and it has

three existential restrictions over the property is-

ParametrizedBy. When the concept DDF is added

to the brief ontology, its relationship with other con-

cepts and individuals is analysed. These connections

are represented in the deﬁnition below.

DDF

v RainfallStatisticalAnalysis

∃ isParametrizedBy.RainfallDepth

Rainfall statistical analysis whose acronym stands for

Depth, Duration and Frequency.

∃ isParametrizedBy.RainfallDuration

∃ isParametrizedBy.RainfallFrequency

According to the traversal algorithm, because of

the property isParametrizedBy is included in the set of

relevant properties, the concepts RainfallDepth, Rain-

fallDuration and RainfallFrequency will be added to

the brief ontology and the traversal algorithm will

continue through these concepts.

Figure 1 depicts a schema of the resulting model

for ﬂood assessment and management. The schema

shows the main concepts and how they are related by

the properties.

KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development

4.7 Evaluation

The interpretation of the model is that ﬂoods are pro-

duced by a high level of water discharge or isolated

events like bridge occlusion and embankment breaks.

The discharge may be produced as well by rainfall, ar-

tiﬁcial water like canals, superﬁcial waters like rivers

or bad water regulation but it also is affected by the

catchment area. The rainfall, which is usually the

main cause of high discharge, is usually character-

ized by statistical analysis and design hyetograph. Fi-

nally, ﬂood management is the union of the preven-

tive, mitigating and recovery actions that must be ac-

complished. However, the management also involves

some processes like forecasting, economic evalua-

tion, etc. (different agents like the municipality are

in charge for each process).

The detailed ontology contains 2054 named

classes and this number has been reduced to 91 in

the brief ontology. Therefore, the brief ontology for

ﬂoods only includes the relevant knowledge for this

case of study.

As (Stuckenschmidt and Schlicht, 2009) says,

there is not golden standard to compare the results

with and the goodness of the brief ontology depends

on the application that will use the ontology. For this

reason, the resulting brief ontology has been posi-

tively evaluated by several experts in the targeted do-

main (ﬂoods). Nonetheless, the quality of the brief

ontology depends totally on the quality of the detailed

ontology.

5 CONCLUSIONS

In general, brief ontologies have a wide range of ad-

vantages when, for some reason, the user or applica-

tion does not wish to deal with the whole original on-

tology. Sometimes, the user is no interested in using

all the information or the application is not capable of

dealing with such a huge resources.

Moreover, reusing a large ontology when only a

small portion is useful and relevant for our applica-

tions may involve unfavourable consequences i.e. the

reasoning time increases with the size of the knowl-

edge base and this issue may be essential in real-time

applications. For this reason, the efﬁciency of our

knowledge base is improved by isolating portions of

knowledge from large ontologies in form of brief on-

tologies.

As an example, a case of study in ﬂood manage-

ment has been presented. A brief ontology is created

specifying the initiator concept (ﬂood) for the traver-

sal algorithm and the set of relevant properties to de-

cide which concepts on the ontology are relevant. The

result has been an ontology where the number of con-

cepts has been dramatically reduced and thus it con-

tains only concepts related to ﬂood.

As future work, it is planned to develop metrics to

compare the detailed and brief ontologies. For exam-

ple, the abstraction degree of equivalent concepts in

both ontologies or the representativeness of the brief

ontology.

ACKNOWLEDGEMENTS

This work has been partially supported by re-

search projects (CICE) P07-TIC-02913 and P08-

RNM-03584 funded by the Andalusian Regional

Governments.

REFERENCES

Aho, A. V., Ullman, J. D., and Hopcroft, J. E. (1983). Data

Structures and Algorithms. Addison Wesley.

Baader, F., Calvanese, D., McGuiness, D., Nardi, D., and

Patel-Schneider, P. (2003). The Description Logic

Handbook: Theory, Implementation and Applications.

Cambridge University Press.

Bizer, C. and Seaborne, A. (2004). D2rq - treating non-

rdf databases as virtual rdf graphs (poster). In The

Semantic Web-ISWC.

De Wrachien, D., Mambretti, S., and Schultz, B. (2011).

Flood management and risk assessment in ﬂood-

prone areas: Measures and solutions. Irrigation and

Drainage, 60(2):229–240.

Delgado, M., P

erez-P

erez, R., and Requena, I. (2005).

Knowledge mobilization through re-addressable on-

tologies. In EUSFLAT Conf., pages 154–158.

Directive 2007/60/CE (2004). Directive 2007/60/CE of the

European Parliament and of the Council of 23 Octo-

ber 2007 on the assessment and management of ﬂood

risks (OJ L 288, 6.11.2007, p. 2734).

Fern

andez, M., G

omez, A., Pazos, J., and Pazos, A. (1999).

Ontology of tasks and methods. IEEE Intelligent Sys-

tems and Their Applications, 14(1):37–46.

Garrido, J. and Requena, I. (2010). Knowledge mobiliza-

tion to support environmental impact assessment. a

model and an application. In Proceedings - inter-

national Conference on Knowledge Engineering and

Ontology Development, KEOD, pages 193–199.

Garrido, J. and Requena, I. (2011a). Proposal of ontology

for environmental impact assessment. an application

with knowledge mobilization. Expert System with Ap-

plications, 38(3):2462–2472.

Garrido, J. and Requena, I. (2011b). Towards summaris-

ing knowledge: Brief ontologies. Submited to Expert

System with Applications.

BUILDING BRIEF ONTOLOGIES - A Case Study for Floods Management

Gu, H., Perl, Y., Geller, J., Halper, M., and Singh, M.

(1999). A methodology for partitioning a vocabu-

lary hierarchy into trees. Artiﬁcial Intelligence in

Medicine, 15(1):77–98.

Kim, J., Caralt, J., and Hilliard, J. (2007). Pruning bio-

ontologies. In Proceedings of the Annual Hawaii In-

ternational Conference on System Sciences.

Noy, N. and Musen, M. (2009). Traversing ontologies to

extract views. Lecture Notes in Computer Science

(including subseries Lecture Notes in Artiﬁcial Intel-

ligence and Lecture Notes in Bioinformatics), 5445

LNCS:245–260.

Staab, S. and Studer, R., editors (2009). Handbook on

Ontologies (International Handbooks on Information

Systems). Springer.

Stuckenschmidt, H. and Schlicht, A. (2009). Structure-

based partitioning of large ontologies. Lecture Notes

in Computer Science (including subseries Lecture

Notes in Artiﬁcial Intelligence and Lecture Notes in

Bioinformatics), 5445 LNCS:187–210.

KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development