Contents lists available at SciVerse ScienceDirect
Information Systems
Information Systems 38 (2013) 727–744
0306-43
http://d
n Corr
Comput
Tel.: þ6
E-m
a_kartik
journal homepage: www.elsevier.com/locate/infosys
An approach for sub-ontology evolution in a distributedhealth care enterprise
Anny Kartika Sari a,b,n, Wenny Rahayu a, Mehul Bhatt c
a Department of Computer Science and Computer Engineering, La Trobe University, Victoria 3086, Australiab Department of Computer Science and Electronics, Gadjah Mada University, Yogyakarta 55281, Indonesiac SFB/TR 8 Spatial Cognition, Informatics, University of Bremen, Germany
a r t i c l e i n f o
Available online 25 April 2012
Keywords:
Ontology evolution
Change propagation
Sub-ontology
Health
Semantics of change
79/$ - see front matter & 2012 Elsevier Ltd. A
x.doi.org/10.1016/j.is.2012.03.006
esponding author at: Department of Com
er Engineering, La Trobe University, Victo
1 420852501.
ail addresses: [email protected],
[email protected] (A.K. Sari).
a b s t r a c t
In response to the changing nature of health issues, standardized health ontologies such
as SNOMED CT and UMLS incline to change more frequently than most other domain
ontologies. Yet, semantic interoperability shared among institutions within a distrib-
uted health care enterprise relies heavily on the availability of a valid and up-to-date
standardized ontology. In this paper, we propose the creation and preservation of sub-
ontologies to deal with the frequent changes in health ontologies. Our approach focuses
on the nature and characteristics of standard health ontologies, however it can also be
applied to other domain ontologies with similar characteristics. Our sub-ontology
evolution approach defines ways to create valid sub-ontologies for each specific health
application, and to effectively develop a series of propagation mechanism when the
main ontology changes. Our approach will (i) isolate the required change propagation to
the relevant health applications that utilized the changing concepts only, and (ii)
optimize the propagation mechanism to include the minimum number of operations
only. Since a sub-ontology should be a valid ontology by itself, the change propagation
approach used in this process should contain the rules to assure the validity of the
produced sub-ontology while keeping the consistency of the sub-ontology to the
evolved base ontology. A change identification process, which considers the nature of
the health ontology change logs, is conducted to identify the semantics of the changes.
From the evaluation, it is shown that the content of the evolved sub-ontologies
produced using our approach is consistent to the evolved base ontology. Moreover,
the propagation process can be performed more efficiently because the number of
operations required for our change propagation method is lower than the number of
operations required for direct re-extraction from the evolved base ontology.
& 2012 Elsevier Ltd. All rights reserved.
1. Introduction
Health care providers range from hospital to specific carecenters such as rehabilitation centers. Patients can choose orbe recommended to different health care providers to achieve
ll rights reserved.
puter Science and
ria 3086, Australia.
their health goals. The increased mobility of people oftenresults in them receiving health treatment from caregiverswho are geographically separated. In these conditions, inter-operability among different health providers is vitally impor-tant so that the medical records for each patient can bepreserved and exchanged between providers. Brailer in [1]believes that the consumer can suffer from a lack of inter-operability and health information exchange because thehealth care enterprise hopes to gain a comparative advantageby imposing high costs when consumers change health careproviders. According to Dogac et al. [2], full ‘share-ability’ of
A.K. Sari et al. / Information Systems 38 (2013) 727–744728
data and information requires two levels of interoperability:semantic interoperability and functional (syntactic) intero-perability. Our focus is on semantic interoperability, which isdefined as the ability for information shared by systems to beunderstood at the level of formally defined domain conceptsso that the information is computer processable by thereceiving system [3].
It is believed that ontologies are one way to overcome thesemantic interoperability problem between different healthcare providers. Using ontology, the semantic meaning of eachhealth term can be uniformly interpreted. An example of theuse of ontologies for health care applications is the bindingbetween the archetype terms and the ontology concepts.Several health ontologies such as SNOMED CT (SystematizedNomenclature of Medicine—Clinical Terms), LOINC (LogicalObservation Identifiers Names and Codes) and UMLS (Uni-fied Medical Language System) can be referred to by arche-type terms. Archetype has been proposed by the openEHR1
Foundation as a model of specific domain knowledge. Thismodel has also been adopted by the CEN TC/2512 in itsHealth informatics—Electronic Health Record Communica-tion (EN 13606) European Standard. The binding betweenthe archetype terms and the ontology concepts is aimed atachieving semantic interoperability between different healthcare institutions which may use different electronic healthrecord standards. Health terminologies are also used in asimilar way in the HL7 (Health Level Seven) standard.Externally defined terms and codes such as SNOMED CTcan be utilized in HL7 as an Instance Identifier, which is usedto give a unique identity to people, persons, organizations,things and information objects.2
Since ontologies aim for standardization, their size isusually very large. Many ontologies in the health domainsuch as SNOMED CT, LOINC and especially UMLS havehundreds of thousands and even millions of concepts. For aspecific health application, the use of the whole ontology isnot appropriate since actually, only a small part of theontology is relevant to the application. For example, thehealth information system of a pharmacy, which usesSNOMED CT as the base ontology, requires only terms relatedto drugs, while there are many more terms in SNOMED CT,such as terms related to the examination of patients andmedical procedures, which obviously are not relevant to theapplication. For an application with such a specific focus, sub-ontologies can be utilized instead of the whole ontology.
A sub-ontology is a subset of an ontology derived fromthat base ontology using a specific extraction process. Acharacteristic which differentiates a sub-ontology from asubset is that a sub-ontology should be a valid ontology inits own right [4]. A sub-ontology refers to a particular part ofa base ontology which is appropriate to a specific context,user, specialty, etc. An example of the use of a specificcontext of knowledge in the clinical domain is the conceptof archetype, which has been mentioned previously. Anarchetype describes a complete clinical knowledge conceptsuch as ‘diagnosis’ or ‘test result’ [5]. An archetype maycontain clinical terms which refer to the terms in health
1 See: http://www.openehr.org/.2 See: http://www.hl7.org.au/HL7-V3-Resources.htm.
ontologies such as SNOMED CT and LOINC. Sari et al. proposethe use of archetype sub-ontology to represent the semanticcontent of an archetype [6]. The archetype sub-ontology isextracted from the health ontology. Similarly, Yu et al. [7] hasproposed a kind of sub-ontology referred to as the Termino-
logical Shadow, which is derived from SNOMED CT, torepresent the semantic content of an archetype. Thesestudies show the applicability of the use of sub-ontologiesin the health domain.
Ontology and sub-ontology should represent the currentknowledge in the domain. When the knowledge changes,they should be adjusted. This process is known as ontologyevolution and is one of the prominent issues in the use ofontologies in the health domain as it is one of the domains inwhich knowledge changes frequently. This is shown by thehigh frequency of health ontology changes. As an example, ineach version of SNOMED CT, which is released twice a year,the average number of changes is more than 50,000 whichconsist of additions (45.45%), status changes (30.87%), andminor changes (23.68%) [8].
When an ontology evolves, the sub-ontologies derivedfrom it should be adjusted as well so that they are consistentwith the base ontology. Re-extraction based on the evolvedontology can be the simplest method to maintain theconsistency of the sub-ontologies. However, this approachis not practical when the number of sub-ontologies is highand the changes in health ontologies occur frequently. In thiscase, it is more appropriate to change the sub-ontologiesaccording to the changes which have taken place in the baseontology. In the notion of ontology evolution, this process isknown as change propagation.
In addition to ensuring consistency to the base ontology,another important reason for the change propagation pro-cess to occur in sub-ontologies is that sub-ontologies shouldbe kept valid. In other words, it should be assured that theevolved sub-ontologies produced from the change propaga-tion process are the same as the ones extracted directly fromthe evolved base ontology. Rules are needed to determinewhich changes should be propagated to the sub-ontologiesto avoid the sub-ontologies from becoming too big or toosmall but, at the same time, keep its semantic content. Mostof the existing change propagation approaches [9–11] do notconsider this requirement because they have not beenapplied to sub-ontologies. Moreover, the approaches areusually based on the assumption that the semantics of thechanges are already known from the version log whichcontains the list of changes which occur from a previousversion of the ontology to the next one. This is not appro-priate for health ontologies as most of them do not provideversion logs containing a list of changes which are semanti-cally meaningful. For example, SNOMED CT provides a list ofbasic change operations such as additions and deletions ofconcepts and descriptions, while they actually permit morecomplex change operations such as the movement of con-cept. To capture such types of changes, an approach toidentify the semantics of change is needed which shouldbe based on the nature of health ontologies.
In this paper, we propose the use of sub-ontologies as away to simplify ontology evolution management in a dis-tributed health enterprise. The main issue addressed in thiswork is the change propagation mechanism from the base
A.K. Sari et al. / Information Systems 38 (2013) 727–744 729
ontology to the sub-ontologies derived from it which can bedistributed among different health institutions. The goal is tomaintain the consistency of the sub-ontologies with the baseontology while keeping the validity of the sub-ontologies. Tosimplify the propagation process, we also develop a changeidentification process which is based on the available changelogs of health ontologies. A case study is used to evaluate theapproach by comparing a sub-ontology produced from theproposed change propagation approach with the one directlyextracted from the evolved ontology in terms of the consis-tency of the content and the number of operations carried outto produce them. The efficiency of the sub-ontology evolutionprocess is also enhanced by isolating changes only to the sub-ontologies affected by the changes which occurred in the baseontology.
The rest of the paper is organized as follows. Section 2presents a motivational scenario which shows the benefitof our approach. Section 3 elaborates previous workrelated to ontology change propagation. The formalizationof a health ontology is presented in Section 4, followed bya discussion on change operations in Section 5 and adescription of the identification process of the semanticchange operations in Section 6. Section 7 discusses thechange propagation approach using some rules. The eva-luation of the performance of the approach is presented inSection 8, and finally, Section 9 concludes our paper.
2. Motivational scenario
Consider a situation where a patient is suspected to sufferfrom diabetes. He goes to hospital to see the doctor. The
Archetype onlow-fat diet
Archetype onhipoproteinemia diagnosis
Archetype on bloodpressure measurement
LabAnalyst
Nutritionist
Doctor
Archetype on bloodglucose measurement
Archetype on diabetesdiagnosis
Archetype ondiabetic diet
Userinterface
semanticalrepresents
Laboratory
Hospital
Nutrition centre
Fig. 1. Sub-ontologies derived from a particular health ontology to be used
domain must be consistent with the evolved base ontology.
doctor advises him to go to the laboratory to have his bloodglucose checked. In the laboratory, a lab analyst analyzes hisblood glucose according to the standard procedure for bloodglucose measurement and then sends the results to thedoctor. The doctor will make a diagnosis based on hisknowledge of diabetes by examining the results of the bloodglucose test and other parameters. This diagnosis is also sentto a nutrition center so that the patient can discuss with thenutritionist the diet he must follow to prevent his diabetesfrom becoming worse. The nutritionist gives suggestionsbased on the treatment procedures for diabetic patients.
In the situation described above, three health providersare involved. Since they may come from different healthinstitutions, semantic interoperability between them mightbecome a problem. This problem can be avoided if the healthinformation systems used by these providers are based on astandardized health ontology where each has a copy of thehealth ontology used. However, since a standardized ontol-ogy tends to be general and hence often very large, thisapproach has disadvantages in terms of resources, whichincludes the storage and the time needed to access theontology. To address this drawback, we propose the struc-ture depicted in Fig. 1, where each health practitioner makesuse of the relevant application through a user interface. Inthis example, archetypes are used as the user’s front-endknowledge base. To minimize the resources needed in usinga large ontology, a sub-ontology is developed for each healthprovider, each of which is based on the same healthontology, in this case SNOMED CT. Each sub-ontology con-tains a particular part of the base ontology which is relevantto the domain knowledge content of the corresponding
Sub-ontology on
bloodanalysis
Sub-ontology onmetabolicdiseases
Sub-ontology on
diets
Healthontology
(SNOMEDCT)
Evolvedhealth
ontology(SNOMED
CT)
evolves
ly
derives
must beconsistent to
as the knowledge representation of different applications in the health
A.K. Sari et al. / Information Systems 38 (2013) 727–744730
health provider, including the content of all archetypesavailable in that particular health care provider. In terms ofthe resource requirements, an ontology view (a personalizedview of an ontology) can be an alternative to a sub-ontology.However, we consider two advantages in the use of sub-ontologies compared to the personalized views of ontology.Firstly, a sub-ontology is a valid ontology by itself, while anontology view may or may not be a valid ontology. Since thecontent is self explanatory, which means that the semanticmeaning of each main sub-ontology component is fullydescribed, the sub-ontology can be used independently bythe local system without the need to refer back to the baseontology. This is not the case for the ontology view. Secondly,a sub-ontology is always materialized because it is obtainedfrom an extraction process, while an ontology view may notbe materialized because it may be obtained and used onlyduring run time. For these two reasons, we believe that theuse of a sub-ontology in the information system of eachhealth care provider is more suitable than the use of anontology view.
The frequent changes which occur in health ontologiesmight affect the sub-ontologies. In Fig. 1, the changes areindicated by the right arrow from the health ontology to theevolved health ontology. The changes influence a sub-ontol-ogy if they happen to the components included in the sub-ontology. In this case, the changes should be propagated tothe sub-ontology so that the sub-ontology is consistent withthe base ontology and the current health knowledge. In thisupdating process, the use of sub-ontologies in the distributedhealth environment has several advantages as follows:
�
The updating of the ontology as the knowledge repre-sentation in each health provider can be done moreefficiently because the size of the ontology is small. � The updating process can be isolated only to the sub-ontologies affected by the changes. This also contri-butes to the efficiency of the process. For example, forthe sub-ontologies in Fig. 6, if the changes to the baseontology only apply to different kinds of food related todiets, only the sub-ontology on diets should beupdated, while the other two remain the same.
� With regard to the use of archetypes, the changes whichoccur to the sub-ontology corresponding to an archetypeindicate that the archetype should be modified so thatthey are also consistent with the current knowledge. Therelation between an archetype and the sub-ontology canbe seen in the binding between the archetype terms andthe sub-ontology concepts. The binding mechanism isavailable in the archetype specification. If a sub-ontologyconcept is changed, the archetype term bound to itshould be checked for the possibility of change. Thesystem can recommend to the archetype author whicharchetype terms should be checked. This mechanism canalso be applied to other different user front-end knowl-edge bases, as long as there is a mechanism for bindingbetween the terms used and the sub-ontology concepts.This issue will become part of our future work related tothe ontology change propagation mechanism.
In our previous work [8], change propagation is per-formed based on the delta of changes presented in each
release of a new (evolved) version of the health ontology.The delta of changes contains only changes of the basicoperations. The basic operations are based on the imple-mentation of the changes. The semantics of the changes,such as the validation that a deletion of a concept must bepreceded by the deletion of relationships related to theconcept, is not considered. Since the main goal of theapproach is simply to maintain the consistency of the sub-ontologies with the base ontology, the approach can beused as long as the produced sub-ontologies are consistentwith the base ontology. However, the approach cannotanswer queries between versions, such as which conceptshave been moved and which relationships have beenadded. We address this issue in this paper by identifyingthe semantics of the change operations using a systematicchange detection method.
3. Related work
Several frameworks of ontology evolution such asCHAO [12], KAON [13] and Evolva [14] include changepropagation as one of the phases in the ontology evolutionprocess. In [15], change propagation is the fifth phase of asix-phase ontology evolution process: change capturing,change representation, semantics of change, change imple-mentation, change propagation and change validation.According to [16], the aim of the change propagationphase is to bring all dependent artifacts in a consistentstate by propagating changes to all dependent artifacts.
Not all papers on ontology evolution framework discussthe strategy used in the change propagation phase indetail. Table 1 summarizes the work on change propaga-tion in terms of the dependent artifacts they are applied to,the knowledge of the semantics of change, the ontologiesused, and the source of change propagation. Our proposedapproach differs from this previous work in that we focuson sub-ontologies as dependent artifacts. As we mentionedpreviously, the change propagation process must producesub-ontologies which are the same as the ones directlyextracted from the evolved base ontology. Moreover, mostof the work in Table 1 is based on the assumption that thesemantics of the changes are known because they are notbuilt particularly for health ontologies which present onlybasic operations in their change log. Even though usingSNOMED CT as the base ontology, the work in [17] doesnot need knowledge of the semantics of change as thechange propagation is only applied to the subset whichdoes not have hierarchy. We address this issue by using anapproach to identify the semantics of change.
A large number of studies have been conducted on theevolution of health ontologies. In [22], a formal methodcalled RLR (represent, legitimate and reproduce the changesand their effects) is proposed for analyzing and supportingthe evolution and change management of biomedical ontol-ogies. The method is applied to the FungalWeb ontology. Yuand Cimino in [23] propose a formal representation of theConcept–Term relation as a special concept called Concept-TermRelation. The authors argue that the model is generalenough to handle each of the possible types of changesoriginally described in [24] which presents a taxonomy ofchanges in medical ontologies. A report on the changes and
Table 1Summary of change propagation works in ontology evolution.
Articles Dependent artifacts Semantics ofchange
Ontology Source of change propagation
[10,18] Other ontologies (in a
distributed ontologies
environment)
Known RDFS based ontologies Evolution log which contains instances of one of the sub-
concepts of the change concept
[9,19] Local health ontologies
(in a shared-local ontologies
environment)
Known Health ontology A change-documentation model for each change type
containing information on the changes
[17] Subset Known SNOMED CT SNOMED CT history table
[12] Other ontologies Known OWL based ontology Special change ontology called CHAO (Change and
Annotation Ontology) whose instances represent the
changes
[20,16] Not specified Known OWL based ontology A version log keeps track of all the different versions of all
concepts in the form of the Change Definition Language
[21] Knowledge management
(KM) system and the business
process (BP) systems
Known Not specified A registration server records every reference to the
elements in the ontology schema by the elements of
applications. It is used for checking and maintaining
consistency between the dependent applications and the
ontology schema
A.K. Sari et al. / Information Systems 38 (2013) 727–744 731
improvements of Sequence Ontology (SO), which providesthe terms and relations between terms to describe biologicalsequences, is presented in [25] in order to better definethe mereological, spatial and temporal aspects of biologicalsequence. While in this paper the change propagationmethod is the main issue, none of the work previouslymentioned discusses this in the approaches they propose.
4. Formalization of health ontology
Before formalizing the change operations of healthontologies, in this section we first formalize the definitionof a health ontology. Our definition is mainly based on twoprominent health ontologies, i.e. UMLS and SNOMED CT.Since UMLS contains several different health ontologieswhich can be represented uniformly, we consider that bybasing our definition on UMLS, we can accommodate therepresentation of other health ontologies. Some exampleswill be provided and taken from SNOMED CT to supportthe formal definition of health ontologies.
We need to give a specific definition of a health ontologybecause it has some characteristics which do not exist incommon ontologies. Firstly, health ontologies do not includeinstances. The instances of the concepts are the concreteobjects in real life. For example, in [26], it is stated thatSNOMED CT concepts are instantiated by three possibleentities: objects that exist independently of the clinicalcontext, artifacts as contained in an electronic patient recordand patients or clinical situations. An example provided inthe work is that the class Liver is instantiated by everyindividual liver. If these instances are included in theontology, the size of the ontology will become very large.Nevertheless, most formal ontology definitions, such as in[27,28], include instances. The inclusion of instances makesthe definitions not suitable to be applied to health ontologies.Hence, we build our own formal definition of health ontol-ogies. Secondly, it has a specific element which we refer to asdescription which explains a concept in natural language.This component is not available in a common ontology
definition, in which a description or label of a concept orclass is usually represented as a data type property. However,in health ontologies, the same description can be used bymore than one different concept. Thus, we need to have aspecial component to represent the description so that eachdescription has its own description ID. The terminologydescription used in this work is based on the term used inSNOMED CT, but such an element is also available in otherhealth ontologies such as UMLS with different terminology.Since a description can be used by more than one concept, anelement we referred to as description mapping which relatesa description to the concept it explains is also defined.
Elements of a health ontology consist of concepts,descriptions, description mappings and relationships. Thedefinition of health ontology is formalized by Definition 1.Note that the definition of health ontologies in this workcan also be used for ontologies in other domains as long asthey have similar characteristics to health ontology.
Definition 1. Health ontology definition.OH �/C,R,D,MS with C � fc1,c2, . . . ,cwg is the set of
concepts, Ca|, R� fr1,r2, . . . ,rxg is the set of relationships,D� fd1,d2, . . . ,dyg is the set of descriptions andM� fm1,m2, . . . ,mzg is the set of description mappings,where C, R, D, M are finite.
We define a health ontology as a tuple /C,R,D,MSconsisting of a finite, non-empty set of concepts C, a finiteset of relationships R, a finite set of descriptions D and afinite set of description mappings M. A concept is a mean-ing identified by a unique code. Each concept can have oneor more terms (strings in natural language) attached to it,each of which is called a description for that particularconcept. For example, the concept 75367002 in SNOMEDCT has the description blood pressure with description ID125176019. The assignment of a description to a particularconcept is formalized as a description mapping. Relation-
ships connect concepts to each other. The formal definitionof relationship and description mapping follows.
A.K. Sari et al. / Information Systems 38 (2013) 727–744732
Definition 2. Relationship definition.8r 2 R : r¼ ðc1,t,c2Þwith c1,c2 2 C, typeðrÞ ¼ t, t 2 T , where
T � ft1,t2, . . . ,tng is the set of relationship types. c1 is referredto as the first concept and c2 is referred to as the secondconcept of relationship R.
In this work, the form of relationship is regarded as athree tuple ðc1,t,c2Þ in which c1 and c2 are the concepts tobe connected and typeðrÞ ¼ t is the relationship type whichconnects the concepts. This form is to accommodate theOAV (Object Attribute Value) triplet form of the SNOMEDCT relationship. T can be adjusted to conform to therelationship types used in each health ontology. In thecase of SNOMED CT, t has to be one of the descendantconcepts of the top level concept Linkage Concept. Anexample of a relationship in SNOMED CT is the (75367002,IS-A, 252059006) triplet. In this example, 75367002 is thefirst concept, 252059006 is the second concept, and IS-A isthe relationship type. The main relationship type used inhealth ontology is the IS-A type, and thus, the IS-A relation-ship type must exist in T. The ontology will form a hierarchybased on the IS-A relationships. The hierarchy must have atleast one root concept croot.
Definition 3. Description mapping definition.8m 2 M : m¼ ðc,dÞ with c 2 C and d 2 D. m is referred to
as the description mapping of concept c.
A description mapping is the mapping of concept c
(element of C) onto description d (element of D). Thedescription d explains concept c in natural language. Theremust be at least one description mapping for a concept c.On the other hand, a description d can be included inmore than one description mapping of different concepts.The description mappings for concept 75367002 are(75367002, 125176019) and (75367002, 1495437014).The description ID 125176019 refers to blood pressure,while the description ID 1495437014 refers to BP—blood
pressure.A health ontology can be represented as an ontology
graph. The graph representation will be very useful tounderstand the change operations available in the ontol-ogy. The graph is defined in Definition 4.
Definition 4. Ontology graph for a health ontology.Given ontology OH �/C,R,D,MS.An ontology graph for OH is GH �/V ,ES with
V ¼ C, E¼ R and type(r), r 2 R, is the name of the edge e 2
E, representing r GH has no island.
The ontology graph consists of a set of vertices V and aset of edges E. Each vertex v (element V) represents anontology concept c (element C), while each edge e (ele-ment E) represents a relationship r (element R). Relation-ship type t in r¼ ðc1,t,c2Þ represents the name of the edge e
connecting c1 and c2. Since an ontology represents knowl-edge in a specific domain interest, the ontology graph of aspecific ontology must be connected, that is, no island ispermitted in the ontology graph. Most of the ontologygraphs of health ontologies will form hierarchies based onthe IS-A relationships.
5. Change operations
Ontology change operations can be observed from twopoints of view: user requirements and implementation.Changes based on these two points of view are different toeach other in their execution. For example, from the userrequirement view, a deletion operation of a concept fromthe hierarchy should also involve, other than the deletionof the concept itself, the deletion operations to the relation-ships connecting the concepts to other concepts and thedeletion operation to its description mappings. This is not thecase with the implementation-based change operation, inwhich a concept deletion operation simply consists of thedeletion of a concept. We discuss these two types of changeoperations in this section.
5.1. Basic change operations
In this paper, the implementation-based change opera-tions are referred to as basic change operations. They are theatomic operations used to perform the user requirement-based change operations. Most health ontologies present thechanges in their releases based on these basic changeoperations. For example, the changes in the SNOMED CTontology are classified into three types of basic operations:addition, status change, and minor change. We formalize thedefinitions of basic change operations in health ontologies inthis section. We begin the formalization with the notation ofthe changed ontology.
Notation 1. Changed ontology.Given ontology OH �/C,R,D,MS.O0H is the changed ontology of OH with O0H �/C0,D0,
R0,M0S, C0 is the changed set of concepts C, R0 is thechanged set of relationships R, D0 is the changed set ofdescriptions D and M0 is the changed set of descriptionmappings M.
The notation of the changed ontology of OH is O0H whichconsists of four tuples /C0,R0,D0,M0S, each represents thechanged set of concepts C, the changed set of relationshipsR, the changed set of descriptions D and the changed set ofdescription mappings M.
As defined in Definition 1, there are four elements of ahealth ontology: concepts, relationships, descriptions anddescription mappings. For each element, two basic opera-tions, i.e. addition and deletion, can be applied. For descrip-tions, one additional operation can be applied that is, thealter operation. Since the alter operation causes a minorchange to the description, it does not influence the structureof the ontology. The formal definition of each operation isshown in Table 2. In the table, for each operation, we presentan example which is taken from the January 2011 release ofSNOMED CT.
5.2. Semantic change operations
Changes based on user requirements suggest thesemantic meaning of the changes. It implies that changesare made by considering the other required operations to
Ta
ble
2T
he
form
al
de
fin
itio
no
fe
ach
ba
sic
cha
ng
eo
pe
rati
on
ap
pli
ed
toh
ea
lth
on
tolo
gy
.
Op
era
tio
nn
am
eS
NO
ME
DC
Tch
an
ge
ex
am
ple
sP
ara
me
ters
giv
en
De
fin
itio
n
Co
nce
pt
ad
dit
ion
Ad
dit
ion
of
con
cep
t4
47
00
50
06
On
tolo
gy
OH
,co
nce
pt
c new=2
CA
dd
Co
nðc
new
,OHÞ3
O0 H9C0 ’
C[fc
newg
Re
lati
on
ship
ad
dit
ion
Ad
dit
ion
of
rela
tio
nsh
ip(4
47
00
50
06
,IS
-A,
10
71
67
00
0)
On
tolo
gy
OH
,re
lati
on
ship
r new=2
R,
r new
con
form
sto
De
fin
itio
n2
Ad
dR
elðr
new
,OHÞ3
O0 H9R0 ’
R[fr
newg
De
scri
pti
on
ad
dit
ion
Ad
dit
ion
of
de
scri
pti
on
28
83
25
00
15
(Pa
roti
a)
On
tolo
gy
OH
,d
esc
rip
tio
nd
new=2
DA
dd
Desðd
new
,OHÞ3
O0 H9D0 ’
D[fd
newg
De
scri
pti
on
ma
pp
ing
ad
dit
ion
Ad
dit
ion
of
de
scri
pti
on
ma
pp
ing
(44
70
05
00
6,
28
83
25
00
15
)
On
tolo
gy
OH
,d
esc
rip
tio
nm
ap
pin
gm
new=2
M,
mn
ew
con
form
sto
De
fin
itio
n3
Ad
dD
esM
apðm
new
,OHÞ3
O0 H9M0 ’
M[fm
newg
Co
nce
pt
de
leti
on
De
leti
on
of
con
cep
t5
80
00
2O
nto
log
yO
H,
con
cep
tc d
el2
CD
elC
onðc
del
,OHÞ3
O0 H9C0 ’
C�fc
delg
Re
lati
on
ship
de
leti
on
De
leti
on
of
rela
tio
nsh
ip(5
80
00
2,
IS-A
,3
63
66
00
07
)O
nto
log
yO
H,
rela
tio
nsh
ipr d
el2
RD
elR
elðr
del
,OHÞ3
O0 H9R0 ’
R�fr
delg
De
scri
pti
on
de
leti
on
De
leti
on
of
de
scri
pti
on
79
66
85
01
2(w
ate
rtru
sh)
On
tolo
gy
OH
,d
esc
rip
tio
nd
del2
DD
elD
esðd
del
,OHÞ3
O0 H9D0 ’
D�fd
delg
De
scri
pti
on
ma
pp
ing
de
leti
on
De
leti
on
of
de
scri
pti
on
ma
pp
ing
(58
00
02
,7
96
68
50
12
)O
nto
log
yO
H,
de
scri
pti
on
ma
pp
ing
md
el2
MD
elD
esM
apðm
del
,OHÞ3
O0 H9M0 ’
M�fm
delg
De
scri
pti
on
alt
era
tio
nA
lte
rati
on
of
de
scri
pti
on
11
85
33
10
11
(ch
an
ge
of
cap
ita
liza
tio
n)
On
tolo
gy
OH
,d
esc
rip
tio
nd
alt2
DA
ltD
esðd
alt
,OHÞ3
D0 9
d0 a
lt’
da
lt
A.K. Sari et al. / Information Systems 38 (2013) 727–744 733
keep the validity of the ontology. For example, when a userneeds to add a concept, it should be assured that theconcept has at least a description mapping and a relation-ship which connects it to an existing concept in theontology; otherwise the concept is not meaningful in theontology. Hence, such a change operation can be a series ofbasic change operations as discussed in the previoussection. We refer to these kinds of user requirement-basedchange operations as semantic change operations.
Fig. 2 shows the taxonomy of the semantic changeoperations in a health ontology. Each operation is given areference number. Basically, each element in the ontologycan be added to or deleted from the ontology, thus, thereare addition and deletion operations for each element.Concept deletion and addition can be applied to a leaf ornon-leaf concept. The addition of a non-leaf concept isdivided into an addition of a non-leaf concept, in which norelationship needs to be deleted, and an insertion of aconcept, in which a concept is inserted into one or moreexisting relationships so that relationship deletion is needed.A concept can be moved to another branch in the ontologygraph such that it becomes a leaf concept (movement to leafconcept) or a non-leaf concept (movement to non-leafconcept). The addition and deletion operations to relation-ships and description mappings are the ones which are notpart of the operations to the concepts.
The addition and movement operations can also beapplied to a fragment. A fragment is a group of connectedconcepts, usually in the form of a parent concept and all itssuccessors. The difference between a fragment and a sub-ontology is that a fragment tends to be used for a partition,while a sub-ontology is a subset of an ontology of specificinterest. The components in a sub-ontology may comefrom several different places/branches in the ontology, butthey must have some commonalities with regard to therequirement of the sub-ontology. On the other hand, afragment usually comprises some ontology componentsfrom the same branch(es). Furthermore, a sub-ontologyshould be a valid ontology by itself and can be usedindependently, while a fragment may not be a validontology by itself. In this work, we restrict our definitionof a fragment to a rooted fragment. Definition 5 formalizesthe definition of a fragment and is based on the definitionof an ontology graph.
Definition 5. Ontology fragment definition.Given ontology OH �/C,R,D,MS.An ontology fragment is FH �/Cf ,Rf ,Df ,MfS with Cf � C,
Rf � R, Df � D, Mf � M. Ontology graph of F is Gf with Gf
has no island, Gf has one root concept cfroot , and � ð(r 2
R9r¼ ðc1, IS�A, cf Þ3r¼ ðcf , IS�A, c1Þ3r¼ ðc1, IS�A,cfrootÞ,cf 2
Cf , cfacfroot , c1=2Cf Þ.
In the above definition, a fragment consists of a set ofconcepts Cf which is also a subset of C, a set of relationshipsRf which is also a subset of R, a set of descriptions Df which isalso a subset of D and a set of description mappings Mf whichis also a subset of M. The ontology graph of F must beconnected and has exactly one root, cfroot. The connection ofthe fragment to the ontology is only established through theroot concept cfroot in which cfroot must be a child concept. This
Table 3Description of semantic change operations.
Operation Description Basic operations involved
AddLeafCon(cnew,cp) Adds new leaf concept cnew which becomes the child of the existing concept cp AddCon, AddDesMap, AddRel
InsCon(cnew, (cp, cc)) Adds new concept cnew by inserting it to an existing relationship (cp,IS-A,cc) so that cp and
cc, respectively, become the parent and the child of cnew
AddCon, AddDesMap, AddRel,
DelRel
AddNonLeafCon(cnew, cp, cc) Adds new concept cnew in which it becomes a non-leaf concept by adding a relationship in
which it becomes the child concept of cp and another relationship in which it becomes the
parent concept of cc
AddCon, AddDesMap, AddRel
AddFrag(FH, cp) Adds new fragment FH in which the root concept cfroot becomes the child of a (previously)
leaf concept cp
AddCon, AddDesMap, AddRel
DelLeafCon(cdel ,cp) Deletes an existing leaf concept cdel which previously has cp as its parent DelDesMap, DelRel, DelCon
DelNonLeafCon(cdel, cp, cc) Deletes an existing non-leaf concept cdel which has cp and cc as its parent and child
concepts, respectively
DelDesMap, AddRel, DelRel,
DelCon
MovToLeaf(cmov, cp) Moves an existing concept cmov (can be a leaf or non-leaf concept) to another place
(branch) in which it becomes a leaf concept and cp is its new parent concept
DelRel, AddRel
MovToNonLeaf(cmov, cp, cc) Moves an existing concept cmov (can be a leaf or non-leaf concept) to another place
(branch) in which it becomes a non-leaf concept and cp and cc, respectively, become its
new parent and child concepts
DelRel, AddRel
MovFragðFH ,cfroot ,cpÞ Moves a fragment FH to another place (branch) such that the root concept cfroot becomes
the child of a (previous) leaf concept cp
DelRel, AddRel
AddRel(rnew, cc, cp) Adds a new relationship in which cc becomes the first concept and cp becomes the second
concept, not part of a concept operation
AddRel
DelRel (rdel, cc, cp) Deletes the relationship rdel which previously connects cc and cp, not part of a concept
operation
DelRel
AddDes(dnew) Adds new description dnew AddDes
DelDes(ddel) Deletes the description ddel DelDes
AltDes(dalt) Edits the description dalt AltDes
AddDesMap(mnew ,c,d) Adds new description mapping mnew in which description d describes concept c, not part
of a concept operation
DelDesMap
DelDesMap(mdel) Deletes the description mapping mdel, not part of a concept operation DelDesMap
Semantic changeoperations
Operations toconcepts
Addition
Deletion
Operations torelationships
Operations todescriptions
Operations todescription mappings
Movement
2.2 Delete relationship
4.1 Add description mapping4.2 Delete description mapping
2.1 Add relationship
1.1 Add leaf concept1.2 Insert concept
1.5 Delete leaf concept1.6 Delete non-leaf concept1.7 Move to leaf concept1.8 Move to non-leaf concept1.9 Move fragment
3.1 Add description3.2 Delete description3.3 Alter description
1.3 Add non-leaf concept1.4 Add fragment
Fig. 2. Taxonomy of the semantic change operations in a health ontology.
A.K. Sari et al. / Information Systems 38 (2013) 727–744734
last requirement is to simplify the operation of fragmentmovement in this paper.
A fragment can be added to the ontology, moved toanother branch in the ontology graph or deleted from theontology. The deletion of a fragment equals the repetitionof a deletion operation of a single concept, and thus, thisoperation is not discussed here. Actually, the addition of afragment can be regarded as several additions of a leafconcept, but we consider that it will be more difficult to
identify them in terms of ordering. In this fragmentaddition, it is assumed that the root of the fragmentbecomes the child of an existing leaf concept. This meansthat the fragment cannot be inserted to an existingrelationship. The movement of a fragment does not equalseveral movements of a single concept because it can besimplified to the movement of the only root concept.
Operations to concepts and relationships will alter thestructure of the ontology hierarchy. Those operations can
A
B C
F G
D
IE H
A
B C
F
G
D
I
E
HQ RJ
J
S
T U
Fig. 3. Examples of the semantic change operations in health ontology: (a) the original ontology; (b) the changed ontology.
1.1 Add leaf concept1.2 Add non-leafconcept
1.3 Delete leaf concept1.4 Delete non-leafconcept
1.5 Move to leaf concept1.6 Move to non-leafconcept1.7 Move fragment
2.1 Add relationship2.2 Delete relationship
4.1 Add descriptionmapping4.2 Delete descriptionmapping
conceptaddition
3.1 Add description3.2 Delete description3.3 Alter description
conceptdeletion
relationshipaddition
relationshipdeletion
descriptionaddition
descriptiondeletion
descriptionmappingaddition
descriptionmappingdeletion
descriptionalteration
Semantic based (user requirement) change operations
Basic (implementation based) change operations
Fig. 4. Identification of semantic change operations based on basic change operations.
A.K. Sari et al. / Information Systems 38 (2013) 727–744 735
be identified through the ontology structures of differentreleases of ontologies. Although operations to descriptionsand description mappings will not alter the structure ofthe ontology, some of them must be checked for theirvalidity in order to meet the constraints applied to theontology. Hence, we include them in the taxonomy.
To explain the semantics of each operation, we presentTable 3 which contains the description of each changeoperation and the basic operations involved to implementit. Fig. 3 shows some examples of the semantic changeoperations in an ontology. The left part of the figure is theoriginal ontology, while the right part is the changedontology. We assume that all the relationship types inthe ontology graph are IS-A. The semantic change opera-tions found in the changed ontology are the insertion ofthe new concept Q into the relationship ðE,IS�A,BÞ (opera-tion InsConðQ ,ðB,EÞÞ), the addition of a non-leaf concept R
(operation AddNonLeaf ðR,B,FÞ), the movement to a non-leafof concept G (operation MovToNonLeaf ðG,D,JÞ), the move-ment to a leaf of concept H (operation MovToLeaf ðH,DÞ)and the addition of fragment FSTU (operation AddFrag
ðFSTU ,AÞ) to the ontology.All of the semantic change operations applied to the
original ontology will change the ontology into its evolvedform. In this work, the set of all semantic change opera-tions which transform an ontology to its evolved form isreferred to as the transformer operations. When all of thetransformer operations are executed, the ontology ischanged to the evolved ontology. Below is the formaldefinition of the transformer operations, which is precededby the notation of Apply operation which is used in theformalization of the transformer operations definition.
Notation 2. Apply operation.Given ontology OH and the semantic change operations
Op1,Op2,y,Opn.ApplyððOp1,Op2, . . . ,OpnÞ,OHÞ is the notation of opera-
tion to apply the change operations Op1,Op2, . . . ,Opn toontology OH .
Definition 6. Transformer operations.Given ontology OH.TO � fOp1,Op2, . . . ,Opng is the set of semantic opera-
tions such that if ApplyððOp1,Op2, . . . ,OpnÞ,OHÞ is executedthen O0H’OH .
6. Identifying the semantic change operations
As previously mentioned, most health ontologies presenttheir change logs based on the basic change operations. Fromthe change logs as well as the comparison of the versions ofthe ontologies, we can derive the following lists of basicchanges: list of concept additions, list of concept deletions,list of relationship additions, list of relationship deletions, listof description additions, list of description deletions, list ofdescription alterations, list of description mapping additionsand list of description mapping deletions. Each operation inthe lists conforms to the corresponding basic operationdiscussed in Section 5. From these lists, the semantic changeoperations can be identified.
In Fig. 4, we present the relationships between thesemantic change operations and the basic operations usedto identify them. We group the semantic operations based onthe similarity of the basic operations used in the identifica-tion process. A line from a basic operation to a group of
Table 4Examples of the lists of basic change operations which are used to
identify the semantic change operations.
List of conceptaddition
List of relationshipaddition
List of relationshipdeletion
X (X, IS-A, A) (B, IS-A, A)
(B, IS-A, X)
Y (Y, IS-A, C)
A.K. Sari et al. / Information Systems 38 (2013) 727–744736
semantic operations shows that the basic operation is usedto identify the operations in the group. A discussion of theidentification process of each group is as follows:
1.
Identification of concept or fragment addition. A conceptaddition operation can be recognized from a row in thelist of concept additions. However, it must be decidedwhether it is an addition of a leaf or non-leaf concept,an insertion of a concept to an existing relationship, oreven an addition of a fragment. This can be identifiedfrom the list of relationship additions and deletions.Table 4 shows an example of three lists of basic changeoperations which are used to identify the semanticchange operations. In the first two rows, the additionof concept X, which is accompanied by the addition ofrelationships (X, IS-A, A) and (B, IS-A, X) and the deletionof relationship (B, IS-A, A) indicates the insertion operationof concept X to relationship (B, IS-A, A). Hence, operationInsConðX,A,BÞ is found. In the third row, the addition ofconcept Y, which is followed by the addition of relation-ship (Y, IS-A, C), indicates a leaf concept addition opera-tion, as no other relationship operations related to Y arefound. In this case, operation AddLeaf ðY ,CÞ is found.Similar identification should be made for other types ofaddition operations.2.
Identification of concept deletion. A concept deletionoperation can be recognized from a row in the list ofconcept deletions. Similar to concept additions, it mustbe decided whether it is a deletion of a leaf or non-leafconcept. This can be identified from the list of relation-ship deletions.3.
Identification of concept or fragment movement. Since amovement operation does not include the addition ordeletion of concepts, description, or description mapping,only the lists of relationship additions and relationshipdeletions are used. The key in the identification process isthe relationship deletion operations, and thus, this type ofidentification must be performed after the identificationof concept deletions so that all the rows in the list ofconcept deletions which are involved in the conceptdeletion operations have been deleted from the list. Thetype of movement operation, e.g. whether it is a move-ment to a leaf concept, a movement to a non-leaf concept,or a movement of a fragment, can be identified from theconcepts involved in the relationship deletion operations.4.
Identification of relationship addition/deletion. The identifi-cation of relationship operations is very simple. Relation-ship additions can be identified from a row in the list ofrelationship additions, while relationship deletions can beidentified from the list of relationship deletions. However,this identification must be performed after the identifica-tion of the concept operations (additions, deletions, andmovements) are finished so that the relationship opera-tions which are part of the concept operations have beendeleted from the list of relationship additions/deletions.
5.
Identification of description addition/deletion. The identi-fication process is similar to the one for the relationshipoperations. Description operations are never part ofconcept operations. However, sometimes when a newconcept is being added, a new description is needed tomap its description but the description needed beincluded in the ontology first. For this reason, the identi-fication of the description addition operation must bedone before the identification of the concept addition. Onthe other hand, the identification of description deletionmust be performed after the identification of concept anddescription mapping operations to avoid a reference to adeleted description.6.
Identification of description mapping addition/deletion.Similar to the identification of relationship addition/dele-tion, the identification of description mapping addition/deletion can be performed after all concept operationshave been performed.From the above description, we determine the order ofthe identification process as follows: description additions,description alterations, concept and fragment additions,concept deletions, concept and fragment movements,relationship additions, relationship deletions, descriptionmapping additions, description mapping deletions anddescription deletions.
7. Propagation of changes to sub-ontologies
The changes which occurred to the main ontology mustbe propagated to the sub-ontologies to ensure that the sub-ontologies are kept updated. Since the changes are reflectedby the change operations, they are the ones which must bepropagated to the sub-ontology. However, there are twoissues which must be considered in the propagation process.Firstly, the change propagation must not cause the unneces-sary growth of the number of concepts in the sub-ontology.Only changes which affect the semantic meaning of theconcepts in the sub-ontology should be propagated. Sec-ondly, the validity and the semantic content of the sub-ontology must be maintained. The validity of a sub-ontologyis related to the structure of the sub-ontology. We presentthe definition of a valid sub-ontology in the next sub-section.The semantic content of an evolved sub-ontology is main-tained when it has the same content as a sub-ontologydirectly extracted from the evolved base ontology. In thissection, we present some rules for the propagation process toaddress these two issues.
7.1. Definition of a valid sub-ontology
In [6], the complete algorithm to extract a sub-ontologyhas been presented and evaluated. Some rules have beenestablished to ensure that the produced sub-ontology is avalid ontology. The method has been applied to SNOMEDCT ontology. In this paper, we extend the definition of a
A.K. Sari et al. / Information Systems 38 (2013) 727–744 737
sub-ontology in [6] such that it can be applied to generalhealth ontologies presented in Section 4. We assume thatthe sub-ontologies discussed in this work conform to thedefinition of valid sub-ontologies. The formal definition ofour sub-ontology is presented in Definition 7.
Definition 7. Sub-ontology definition.SH �/CSH ,RSH ,DSH ,MSHS is a sub-ontology of OH � /C,
R, D, MS3ðCSHaf, CSH � C98c 2 CSH : (m 2 MSH9m¼ ðc,dÞ,d 2 DSHÞ4ðRSH � R98r 2 RSH : (r ¼ ðc1,t,c2Þ 2 RSHÞ4ðDSH � D98d 2 DSH : (m 2 MSH9m¼ ðc,dÞ, c 2 CSHÞ4ðMSH � M98m 2MSH : m¼ ðc,dÞ, c 2 CSH , d 2 DSHÞ4ð( ontology graph GSH
for SHÞ4ð8 c 2 CSH : XðcÞ ¼ l, l 2 fselected, affectedgÞ.
A sub-ontology of health ontology OH is a four-tuplecontaining CSH, RSH, DSH and MSH. CSH is a subset of C in OH
and CSH must not be null, RSH is a subset of R, DSH is a subsetof D and MSH is a subset of M. Each concept in CSH must haveat least one description mapping in MSH. RSH, DSH and MSH
each is used to explain concepts included in CSH by includingonly members which are related to the concepts in CSH. Forinstance, a relationship r in RSH must connect two concepts inCSH. This constraint will avoid the inclusion of the unneces-sary membership of a component to the sub-ontology.Moreover, the sub-ontology will be self-described since eachcomponent in the sub-ontology is sufficiently explained byother components in the sub-ontology itself. Hence, back-reference to the base ontology will not be needed. The sub-ontology must have an ontology graph, which means that allthe concepts in CSH must be connected to each other usingthe relationships in RSH. The relationships can be any type ofrelationships and are not restricted to IS-A.
The most important part of our definition of sub-ontologyis the labeling XðcÞ ¼ l, l 2 fselected, affectedg for eachincluded concept. Originally, as stated in our previous work[6], all elements (concepts and relationships) in the sub-ontology should be labeled with selected, if the element isoriginally chosen to be included in the sub-ontology, oraffected, if the element is included as the side effect of therule application during the extraction process. In this work,we only consider labeling to concepts because it is adequatefor the application of change propagation rules.
To simplify the change propagation process, we add anattribute to each concept which includes the information inthe list of sub-ontologies in which it is involved. The attribute
Sub-ontology
Base ontology
directlyextracted from
evolves
changes
List of semchange o
Fig. 5. The propagation of change from the base ontology to the
can be a list of numbers, each of which refers to a sub-ontology. The attribute is referred to as sub-onto attribute.When a change involves a specific concept, it can be imme-diately seen to which sub-ontologies the changes must bepropagated by checking the sub-onto attribute of the concept.
7.2. Change propagation process
Fig. 5 shows the change propagation process. Asub-ontology extracted from the base ontology must bechanged when the base ontology changes. The lists ofsemantic change operations guide the changes applied tothe sub-ontologies. At the end, it is expected that thechanged sub-ontologies are still consistent with the chan-ged base ontology.
Changes to the base ontology can have an impact on thesub-ontology derived from it if the components changedare also included in the sub-ontology or related to thecomponents in the sub-ontology. As previously mentioned,we use the sub-onto attribute of a concept to determinewhich sub-ontologies are affected by the changes. Thechanges might influence the size as well as the semanticcontent of an affected sub-ontology. Since not all changeoperations in the base ontology need to be propagated tothe sub-ontologies, we develop some rules to guide thisprocess. The goals of the rules are
1.
gu
antipera
de
sub
To maintain the structural validity and the semanticcontent of the sub-ontologies such that the requirementsof a valid sub-ontology, e.g. the inclusion of all inheritancerelationships of the selected concepts and the non-exis-tence of islands, as guided in [6] are met. Hence, the sub-ontologies produced from the propagation process will beconsistent with the ones directly extracted from theevolved base ontology.
2.
To produce the optimized sub-ontologies so that theunnecessary growth of the sub-ontologies can be pre-vented. This can be achieved by limiting the successorconcept level of and the non-IS-A relationship levelconnected to the selected concepts. These rules havealso been established in [6].The rules are classified based on the different groups ofthe semantic change operations discussed in Section 5.2.
ChangedSub-ontology
Changed baseontology
consistent to
ides
c basedtions
rives
-ontology using the list of semantic change operations.
A.K. Sari et al. / Information Systems 38 (2013) 727–744738
For each group, we provide a table which contains thedetail of the actions for each change operation and possibleconditions which influence the propagation. The details ofthe rules follow:
1. Rules for propagating the addition operations. Foraddition operations, the factors to be considered in propa-gating the changes are (1) whether the parent and child ofthe new concept (or root concept, in the case of AddFragoperation) included in the sub-ontology, and; (2) the labelsof the parent and child concept in the sub-ontology. If theparent is included and labeled selected, the new concept(or root concept, in the case of AddFrag operation) shouldbe included in the sub-ontology as well because itbecomes the direct child of the selected concept. However,the sub-ontology should be kept optimized by limiting thelevel of child concept into 1, as suggested in [6]. Whenboth the parent and the child concepts are in the sub-ontology, the same change operation must be applied tothe sub-ontology. When only the child concept is included,the new concept should be included to complete thesemantic definition of the concepts in the sub-ontology.Hence, all ancestors of the new concept must be includedas well. Details of the actions for each operation andcondition are presented in Table 5, while the general defini-tions of the rules are formalized as follows. In the rules,Opaddition refers to the addition operation, parent(c) refers tothe parent of concept c and child(c) refers to the child ofconcept c. cnew is replaced by croot in AddFrag operation.
Rule 1.1: 8 Opaddition 2 TO : parentðcnewÞ 2 CSH4Xðparent
ðcnewÞÞ ¼ selected) Apply ðOpaddition,SHÞ4keep the sub�ontology
structure optimized.Rule 1.2: 8 Opaddition 2 TO : parentð cnewÞ, childðcnewÞ 2
CSH , child ðcnewÞ not a leaf ) ApplyðOpaddition,SHÞ.
Table 5The propagation rules for each addition operation.
Operation Condition of theparent concept
Condition of thchild concept
AddLeafConðcnew ,cpÞ cp 2 CSH , XðcpÞ ¼ selected N/A
InsCon(cnew, (cp, cc)) cp 2 CSH , XðcpÞ ¼ selected N/A
AddNonLeafCon(cnew, (cp, cc)) cp 2 CSH , XðcpÞ ¼ selected N/A
AddFragðFH ,cpÞ cp 2 CSH , XðcpÞ ¼ selected N/A
InsCon(cnew, (cp, cc)) cp 2 CSH cc 2 CSH , cc not a
AddNonLeafCon(cnew, (cp, cc)) cp 2 CSH cc 2 CSH , cc not a
AddNonLeafCon(cnew, (cp, cc)) cp=2CSH cc 2 CSH , cc not a
Table 6The propagation rules for each deletion operation.
Operation Status of thedeleted concept
Condition of thand child conc
DelLeafCon ðcdel ,cpÞ XðcdelÞ ¼ selected N/A
DelNonLeafCon ðcdel ,cp ,ccÞ XðcdelÞ ¼ selected N/A
DelLeafCon ðcdel ,cpÞ XðcdelÞ ¼ affected cp 2 CSH , XðcpÞ ¼
DelNonLeafCon ðcdel ,cp ,ccÞ XðcdelÞ ¼ affected cp 2 CSH , XðcpÞ ¼
DelNonLeafCon ðcdel ,cp ,ccÞ XðcdelÞ ¼ affected cp ,cc 2 CSH , cc no
Rule 1.3: 8 Opaddition 2 TO : parentðcnewÞ=2CSH4child ðcnewÞ 2
CSH ) ApplyðAddConðcnew,SHÞ,SHÞ48 c ancestors of cnew:
Apply ðAddCon ðc,SHÞ, SHÞ.
2. Rules for propagating the deletion operations. A deletionoperation is propagated to the sub-ontology only if thedeleted concept is included in the sub-ontology. However,there are some factors to be considered in propagating thedeletion operations, which are slightly different from thefactors to be considered in propagating the addition opera-tions. They are (1) the label of the deleted concept in the sub-ontology; (2) the labels of its parent and child concepts in thesub-ontology. If the deleted concept is included in the sub-ontology with selected label, it must be deleted from the sub-ontology together with all of its descendants and ancestorswhich are included only to complete the semantic definitionof the deleted concept. If the parent of the deleted concept islabeled selected, the concept should be deleted from the sub-ontology, but the semantic definition of the parent must bekept complete by including its 1-level child concepts, if it isnecessary. If both the parent and the child concepts of thedeleted concept are included the same operation should beapplied to the sub-ontology. The general definitions of therules are formalized as follows, while the details of theactions for each operation and condition are presented inTable 6. In the formalization below, Opdeletion refers to thedeletion operation.
Rule 2.1: 8 Opdeletion 2 TO : cdel 2 CSH4XðcdelÞ ¼ selected)
ApplyðDelConðcdel,SHÞ,SHÞ48c ancestors and descendants of
cdel : ApplyðDelConðc,SHÞ,SHÞ.Rule 2.2: 8 Opdeletion 2 TO : cdel,parentðcdelÞ 2 CSH4X
ðparentðcdelÞÞ ¼ selected ) ApplyðDelLeafConðcdelÞ, SHÞ4keep
the sub-ontology structure optimized.
e Actions
ApplyðAddLeafConðcnew ,cpÞ,SHÞ
ApplyðAddLeafConðcnew ,cpÞ,SHÞ4ApplyðDelLeafConðccÞ,SHÞ
ApplyðAddLeafConðcnew ,cpÞ,SHÞ
ApplyðAddLeafConðcfroot ,cpÞ,SHÞ
leaf concept Apply(InsCon(cnew, (cp, cc)),SH)
leaf concept Apply(AddNonLeafCon(cnew, (cp, cc)),SH)
leaf concept Apply ðAddConðcnew ,SHÞÞ,SHÞ4ApplyðAddConðc,SHÞÞ,SHÞ
for all c ancestors of cnew
e parentepts
Actions
ApplyðDelConðcÞ,SHÞ48c ancestors and
descendants of cdel : ApplyðDelConðcÞ,SHÞ
ApplyðDelConðcdel ,SHÞ,SHÞ48c ancestors and
descendants of cdel : ApplyðDelConðc,SHÞ,SHÞ
selected ApplyðDelLeafConðcdel ,cpÞ,SHÞ
selected ApplyðDelLeafConðcdel ,cpÞ, AddLeafConðcc ,cpÞ,SHÞ
t a leaf concept ApplyðDelLeafConðcdel ,cpÞ,SHÞ
A.K. Sari et al. / Information Systems 38 (2013) 727–744 739
Rule 2.3: 8 Opdeletion 2 TO : parentðcdelÞ, childðcdelÞ 2 CSH ,childðcdelÞ not a leaf ) ApplyðOpdeletion,SHÞ.
3. Rules for propagating the movement operations. Thefactors to be considered in propagating the changes are (1)whether the moved concept (or root concept, in the case ofMovFrag operation) is included in the sub-ontology; (2)the label of the moved concept, if it is included the sub-ontology; (3) whether the new parent and child of themoved concept included in the sub-ontology; and (4) thelabels of the parent and child concept in the sub-ontology.A moved concept can be moved to another branch in thesub-ontology if it is labeled selected, has a parent conceptand a child concept included in the sub-ontology, or has aparent concept with selected label. However, the structuralvalidity and the optimization of the sub-ontology must bemaintained. If the moved concept moves to a new parentwhich is not included in the sub-ontology, it must bedeleted from the sub-ontology. Details of the actions foreach operation and condition are presented in Table 7. Thegeneral rules for movement operations follow. In the rules,Opmovement refers to the movement operation, and cmov isreplaced by croot in the MovFrag operation.
Table 7The propagation rules for each movement operation.
Operation Condition of the moved concept Condition of thparent and chiconcepts
MovToLeaf ðcmov ,cpÞ XðcmovÞ ¼ selected N/A
MovToNonLeaf
ðcmov ,cp ,ccÞ
XðcmovÞ ¼ selected N/A
MovFrag ðFH ,cfroot ,cpÞ XðcfrootÞ ¼ selected or there is c 2 CFH
which is labeled selected
N/A
MovToLeaf ðcmov ,cpÞ XðcmovÞ ¼ affected cp 2 CSH , cp not
concept
MovToNonLeaf
ðcmov ,cp ,ccÞ
XðcmovÞ ¼ affected cp ,cc 2 CSH , cc n
concept
MovToLeaf ðcmov ,cpÞ,
MovToNonLeaf
ðcmov ,cp ,ccÞ
XðcmovÞ=2CSH XðcpÞ ¼ selected
MovFrag ðFH ,cfroot ,cpÞ XðcfrootÞ=2CSH XðcpÞ ¼ selected
MovToLeaf ðcmov ,cpÞ,
MovToNonLeaf
ðcmov ,cp ,ccÞ
XðcmovÞ ¼ affected, cmov was a leaf
concept, c1 was its parent
cp=2CSH
MovToLeaf ðcmov ,cpÞ,
MovToNonLeaf
ðcmov ,cp ,ccÞ
XðcmovÞ ¼ affected, c1 and c2 were its
parent and child concept,
respectively
cp=2CSH
Table 8The propagation rules for each relationship operation.
Operation Relationshiptype
Condition of the first and secondconcept
Act
AddRelðrnew ,cc ,ccÞ IS-A cc ,cp 2 CSH App
DelRelðrdel ,cc ,cpÞ Anything cc ,cp 2 CSH App
AddRelðrnew ,cc ,cpÞ IS-A cc 2 CSH , cc not leaf, cp=2CSH (8 c
Add
AddRelðrnew ,cc ,cpÞ Other than
IS-Acc 2 CSH , XðccÞ ¼ selected, cp=2CSH App
Rule 3.1: 8 Opmovement 2 TO : cmov 2 CSH4XðcmovÞ ¼ selected
) ApplyðOpmovement , SHÞ4keep the structural validity of the
sub-ontology.Rule 3.2: 8 Opmovement 2 TO : parentðcmovÞ, childðcmovÞ 2
CSH , if ðcmovÞ has a child) ApplyðOpmovement ,SHÞ.Rule 3.3: 8 Opmovement 2 TO : cmov=2CSH4parentðcmovÞ 2
CSH4 XðparentðcmovÞÞ ¼ selected) ApplyðAddLeafCon ðcmov,parentðcmovÞÞ, SHÞ.
Rule 3.4: 8 Opmovement 2 TO : cmov 2 CSH4XðcmovÞ ¼
affected4parentðcmovÞ=2 CSH ) ApplyðDelConðcmovÞ,SHÞ.
4. Rules for propagating the relationship operations. Foroperations related to relationships, the factors to be con-sidered in propagating the changes are (1) the type of therelationship; (2) the condition of the first and second
concepts of the relationship, i.e. whether they are includedin the sub-ontology and their status. If both the first andsecond concepts of the relationship are included in the sub-ontology, or if the first concept is included in the sub-ontology while the second concept is not, the same opera-tion must be applied to the sub-ontology, while at the sametime the structural validity should be kept. Details of theactions for each operation and condition are presented in
e newld
Actions
(8c ancestors and descendants of cmov : ApplyðAddConðcÞ,SHÞ if
they are not included in SHÞ4ApplyðMovToLeaf ðcmov ,cpÞ,SHÞ
(8c ancestors and descendants of cmov : ApplyðAddConðcÞ,SHÞ if
they are not included in
SHÞ4ApplyðMovToNonLeaf ðcmov ,cp ,ccÞ,SHÞ
(8c ancestors and descendants of cfroot : ApplyðAddConðcÞ,SHÞ if
they are not included in SHÞ4ApplyðMovFragðFH ,cfroot ,cpÞ,SHÞ
a leaf ApplyðMovToLeaf ðcmov ,cpÞ,SHÞ
ot a leaf ApplyðMovToNonLeaf ðcmov,cp ,ccÞ,SHÞ
ApplyðAddLeafConðcmov ,cpÞ,SHÞ
ApplyðAddLeafConðcfroot ,cpÞ,SHÞ
ApplyðDelLeafConðcmov ,cÞ,SHÞ
ApplyðDelNonLeafConðcmov ,c1 ,c2Þ,SHÞ
ions
lyðAddRelðrnew ,cc ,cpÞ,SHÞ
lyðDelRelðrdel ,cc ,cpÞ,SHÞ4delete if there is an island
ancestors of cp=2CSH : ApplyðAddConðc,SHÞ,SHÞÞ4ApplyðAddConðcp ,SHÞ,
Relðrnew ,cc ,cpÞ,SHÞ
lyðAddConðcp ,SHÞ, AddRelðrnew ,cc ,cpÞ,SHÞ
A.K. Sari et al. / Information Systems 38 (2013) 727–744740
Table 8. The rule can be formalized in single statement asfollows. In the rule, Oprelationship refers to the operationrelated to relationship, cfirst refers to the first concept inthe relationship, and csecond refers to the second concept.
Rule 4.1: 8 Oprelationship 2 TO : ðcfirst ,csecond 2 CSHÞ 3ðcfirst
2 CSH 4cfirst not leaf 4csecond=2CSHÞ ) ApplyðOprelationship,SHÞ
4keep the structural validity of SH .
5. Rules for propagating the description and description
mapping operations. A description or description mappingchange operation must be propagated to the sub-ontologyif the corresponding description or description mapping isincluded or referred by a concept in the sub-ontology.Since those two operations do not influence the structureof the ontology graph, there is no need to check thestructure of the sub-ontology upon performing thoseoperations. Details of the actions for each operation andcondition are presented in Table 9, while the general ruleis presented as follows. In the rule, Opdesc is the operationrelated to description or description mapping, while argu-ment is the parameter(s) of the operation.
Rule 5.1: 8 Opdesc 2 TO : if argument 2 SH 3argument is
referred by c 2 CSH ) ApplyðOpdesc , SHÞ.
8. Evaluation and discussion
To evaluate our proposed method of change propagation,we conduct an experiment using a case study. We performthe evolution of some sub-ontologies using both our methodand direct re-extraction. Then, we compare the sub-ontolo-gies produced using both methods in terms of the contentand the number of operations executed in both methods todetermine the advantage of our approach. The following sub-section is the description of the experimental design. It isfollowed by the discussion on the issues which arose duringthe implementation of the approach.
Table 9The propagation rules for each description and des
Operation Conditions
AddDesðdnewÞ (AddDesMapðmnew ,c,dnewÞ
DelDesðddelÞ ddel 2 DSH4:(m¼ ðc,dÞ 2 M
AltDesðdaltÞ dalt 2 DSH
AddDesMapðmnew ,c,dÞ c 2 CSH , d 2 DSH
Table 10The sub-ontologies built for the evaluation and the types of the semantic cha
Sub-ontology
Number ofconcepts
Number ofrelationships
Types of semantic ch
A 66 65 1 add leaf concept, 1 a
B 87 136 1 add non-leaf concep
C 88 97 3 add leaf concept, 1 a
D 127 135 30 add leaf concept, 2
concept, 1 move fragm
E 100 127 1 add leaf concept, 1 i
relationship
8.1. Experimental design
In this evaluation, we used SNOMED CT, particularly itsInternational Release, as the base ontology because weconsider that it has high number of changes and varioustypes of changes as classified in Section 5. The July 2010Release of SNOMED CT is considered as the original version,while the next release, i.e. the January 2011 version, becamethe evolved base ontology. We built five different sub-ontologies based on the July 2010 release, and these areconsidered as the original sub-ontologies to evolve.
To identify the semantic change operations applied to theoriginal version to become the evolved version of SNOMEDCT, we deduced the lists of the basic change operations fromthe Component History Table of the January 2011 Releaseand the Relationship Table of July 2010 and January 2011releases. Note that in SNOMED CT, a description cannot beadded independently from the description mappings inwhich it is involved. Thus, no lists of description changesare needed in this process. There are six lists of basic changeoperation produced: list of concept additions, list of conceptdeletions, list of relationship additions, list of relationshipdeletions, list of description mapping additions and list ofdescription mapping deletions. From these lists, we identi-fied the semantic change operations, especially the opera-tions which involve the concepts included in our sub-ontologies. Since we consider that the description mappingoperations do not affect the hierarchy of the sub-ontology,we did not include them in the evaluation.
Table 10 summarizes the sub-ontologies used in thiswork in terms of the number of concepts and relationshipsand the types of the semantic change operations which affecteach sub-ontology. The sub-ontologies were built such thatthey can accommodate all types of semantic change opera-tions discussed in Section 5 in their evolved versions. Onlythe move to non-leaf concept change operation is not includedin the sub-ontologies as we could not find this type ofoperation in the changes of SNOMED CT from the July
cription mapping operation.
Actions
2 TO ApplyðAddDesðdnewÞ,SHÞ
SH9d¼ ddel ApplyðDelDesðddelÞ,SHÞ
ApplyðAltDesðdaltÞ,SHÞ
ApplyðAddDesMapðmnew ,c,dÞ,SHÞ
nge operations affecting them.
ange operations affecting the sub-ontology
dd non-leaf concept, 3 add fragment, 9 del leaf concept
t, 2 add relationship
dd non-leaf concept, 2 del leaf concept, 1 del non-leaf concept
insert concept, 2 add fragment, 28 del leaf concept, 3 move to leaf
ent
nsert concept, 1 add non-leaf concept, 13 add relationship, 5 del
A.K. Sari et al. / Information Systems 38 (2013) 727–744 741
2010 Release to the January 2011 Release. Furthermore, thesub-ontologies represent seven different types of top levelSNOMED CT concepts, i.e. clinical finding, body structure,procedure, substance, organism, qualifier value and linkageconcept. Thus, we conclude that the sub-ontologies used inthis case study can sufficiently represent the types of changeoperations as well as the different concepts of SNOMED CT.
Based on the list of semantic change operations, weapplied our change propagation approach to the definedsub-ontologies to produce the evolved sub-ontologies. Therules defined in Section 7 are applied to guide the evolutionof each sub-ontology. The result would be the evolved sub-ontology produced using the change propagation method.In addition to this, we also did a direct re-extraction of thesub-ontologies based on the evolved ontology, i.e. theJanuary 2011 Release of SNOMED CT. Each pair of sub-ontologies produced both by change propagation methodand by re-extraction was then compared in relation to twoissues (1) the content of the sub-ontologies and (2) thenumber of operations performed to produce the evolvedsub-ontologies. The aim of the first comparison is to see thecorrectness of the sub-ontologies produced by changepropagation, while the second one is aimed at comparingthe efficiency of both approaches.
8.2. Results and discussion
Our rules in change propagation are built with the aim ofmaintaining the validity of the sub-ontology while propagat-ing the changes to the sub-ontologies. They are specificallydesigned to fit in the required structure of a valid sub-ontology. Hence, it is not surprising that the content of theevolved sub-ontology produced using change propagation is
Fig. 6. Comparison of the number of basic change operations as w
exactly the same as the corresponding sub-ontologyextracted directly from the evolved ontology. This meansthat our change propagation method can reserve the validityof the sub-ontology as well as keep the consistency of thesub-ontology to the evolved base ontology.
To compare the efficiency of both approaches in termsof the number of operations, it first must be noted that anextraction only includes concept and relationship additionoperations, both of which are basic operations. Thus, ourcalculation on the operations using the change propaga-tion method was based on those basic operations as well.In our method of sub-ontology extraction presented in ourprevious work [6], concept deletion and relationship dele-tion operations are not used because each time the mainontology changes (e.g. updated, deleted, etc.), the wholesub-ontology is re-extracted. The rules in the labelingprocess only include the change of labeling from void,which means that the element is not included in theontology, to either selected or affected, which means thatthe element is added to the sub-ontology. These changes oflabeling imply that there are additions of elements (con-cepts or relationships) to the sub-ontology. Labeling fromselected or affected to void is never done, which means thatthe concept or relationship deletion operation is not used.However, the two deletion operations are clearly used inthe application of rules in our change propagation process,and hence, we included them in calculating the number ofoperations.
Fig. 6 presents the comparison of the two methods on thenumber of operations. The vertical axis shows the number ofoperations, while the horizontal axis shows the sub-ontolo-gies (sub-ontology A to sub-ontology E). For concept additionand relationship addition operations, it can be seen that the
ell as the total number of operations of the two approaches.
Fig. 7. The ratio of the number of semantic change operations to the number of concepts in the sub-ontology influences the efficiency of the change
propagation approach in terms of the number of operations.
A.K. Sari et al. / Information Systems 38 (2013) 727–744742
number of operations using change propagation is lowerthan the number of operations using re-extraction for eachsub-ontology. Even though re-extraction does not use con-cept deletion and relationship deletion operations, for eachsub-ontology, the total number of operations using changepropagation is still lower than the total number of operationsusing the re-extraction method. This result shows that ourchange propagation approach is more efficient than thedirect re-extraction method in terms of the number ofoperations performed.
It is also interesting to know the influence of the numberof concepts or the number of semantic operations on theefficiency of the approach. We calculated the efficiency bytaking the ratio of the number of operations by changepropagation to the number of operations by re-extraction,referred to as the operation ratio. The higher the operationratio, the lower the efficiency of the approach compared tothe re-extraction method. Fig. 7 shows the relation betweenthe number of concepts in a sub-ontology, the number ofsemantic change operations affecting that sub-ontology, theratio of the number of semantic change operations to thenumber of concepts and the operation ratio. From the chart,it can be implied that the number of concepts does notindependently influence the efficiency of the approach. Theefficiency of the change propagation approach applied insub-ontology A, which has less concepts than sub-ontology Band C, apparently is lower (shown by the higher operationratio) than the application to sub-ontology B and C. Thenumber of semantic change operations has a higher influ-ence on the efficiency of the approach. For sub-ontologieswith a high difference in the number of semantic changeoperations, such as sub-ontology E, it can be clearly seen thatthe higher the number of semantic change operationsaffecting a sub-ontology, the higher the operation ratio.Nevertheless, this relation cannot be clearly seen in sub-ontologies with a small difference in the number of semanticchange operations such as sub-ontologies A and B. We thenuse the ratio of the number of semantic change operations tothe number of concepts and this is referred to as the change
ratio. From the chart, we can see that the change ratio has astronger influence on efficiency than the number of semanticchange operations, i.e. the higher the change ratio, the higherthe operation ratio. This implies that the cost of changepropagation is becoming closer to the cost of re-extractionwhen the number of semantic change operations is very high.However, we believe that, in general, our change propagationapproach has better efficiency than direct re-extractionbecause, in our definition of a valid sub-ontology, it isnecessary that all the ancestors of a selected concept areincluded in the sub-ontology for completeness of the seman-tic definition. The concepts to be changed in the baseontology are most likely at the low level of the ontologygraph, and thus are mostly located at the low level of the sub-ontologies graph. For example, our study from the list ofchanges of SNOMED CT shows that most of the additions ordeletions are applied to leaf concepts. Since it is likely that thenumber of leaf concepts is lower than the number of all of itsintermediate level concepts, the change ratio is low. A lowchange ratio will imply a low operation ratio, which meansthat the efficiency is high. Hence, it can be generalized thattaking into consideration the nature of changes in a well-known health ontology such as SNOMED CT, the efficiency ofour proposed change propagation method will outperformthe re-extraction method in such a changing pattern.
Our proposed sub-ontology-based concept representa-tion has two major benefits. Firstly, it can be used toisolate changes such that only affected sub-ontologies arechanged. In this way, the change propagation method canbe implemented more efficiently compared to the re-extraction method to all the sub-ontologies in terms ofthe number of operations performed. Secondly, it canreduce the effort for semantic change operations identifi-cation in the situation where not many sub-ontologies arepresent. In this situation, identification of all semanticchange operations might not be needed since only a few ofthe operations affect the small number of the sub-ontol-ogies. Using concept representation, the semantic changeoperations which must be identified can be chosen.
A.K. Sari et al. / Information Systems 38 (2013) 727–744 743
9. Conclusion
This paper proposed a new approach to manage theevolution of standardized ontologies used in health enter-prises. The use of distributed sub-ontologies in such anenvironment substantially improves the efficiency ofupdate propagation to interconnected ontologies within adistributed enterprise.
A formal definition of a health ontology, as well as avalid sub-ontology, is presented to give a theoreticalfoundation to the approach. Based on this, several ruleshave been built in propagating the changes of the baseontology to the sub-ontologies. The evaluation of theapplication of the approach in SNOMED CT shows thateach of the produced sub-ontologies is identical to the onedirectly extracted from the evolved ontology. Hence,semantic consistency and validity are assured. In termsof efficiency, our approach is proven to be more efficientthan direct extraction as the case study shows that thenumber of operations performed using our approach islower than the number of operations carried out usingdirect extraction.
There are two main differences between our approachand the existing work. Firstly, our method focuses on thechange propagation process to sub-ontologies, each ofwhich is a valid ontology itself. The rules ensure that thesub-ontologies will be kept valid after changes haveoccurred in the base ontology. The validity of the sub-ontologies is important to guarantee that the sub-ontolo-gies can always be used independently from the baseontology. Secondly, we built our change propagationmethod based on the assumption that the changes inhealth ontologies are usually presented as basic changeoperations. Other approaches do not consider this fact, andhence, they might need to be adjusted in order to apply themethods to the change process in health ontologies.
In our future work, we will focus on three areas related tothe change propagation of health ontologies presented in thispaper. Firstly, we plan to further validate the efficiency of ourpropagation change mechanism to a distributed healthenterprise which utilizes a merging of different healthontologies. This issue is interesting since the ontology mer-ging process is highly probable in health care applications.For example, an archetype can be linked to different healthontologies, which means that the sub-ontology it refers toshould contain the concepts from those ontologies. Secondly,we will develop an approach to propagate the changes fromthe sub-ontologies to the applications, especially to find anindication that an application needs to be checked for itscurrentness due to the changes. This area has not been muchexplored since most of the existing works on the ontologychange propagation focus on the propagation betweenontologies. Finally, we will extend our initial work in [29]on the implementation of ontology evolution in OWL tocover the ontology update operations presented in this work.
Acknowledgment
This work is partially supported by the Ministry ofNational Education of the Republic of Indonesia through
the scholarship granted to the first author. We also wouldlike to express our gratitude to any anonymous reviewersof our manuscript for their valuable suggestions toimprove this paper.
References
[1] D.J. Brailer, Interoperability: the key to the future health caresystem, Health Affairs Web Exclusive (2005). W5-19–W5-21.
[2] A. Dogac, G.B. Laleci, Y. Kabak, S. Unal, S. Heard, T. Beale, P.L. Elkin,F. Najmi, C. Mattocks, D. Webber, M. Kernberg, Exploiting ebXMLregistry semantic constructs for handling archetype metadata inhealthcare informatics, International Journal of Metadata 1 (1)(2005) 21–35.
[3] V. Bicer, G.B. Laleci, A. Dogac, Y. Kabak, Providing semanticinteroperability in the healthcare domain through ontologymapping, in: Proceedings of eChallenges 2005, Ljubljana, Slovenia,2005.
[4] C. Wouters, A Formalization and Application of Ontology Extraction,Ph.D. Thesis, La Trobe University, 2005.
[5] H. Leslie, S. Heard, Archetypes 101, in: J. Westbrook, J. Callen (Eds.),HIC, Health Informatics Society of Australia Ltd. (HISA), Sydney,2006.
[6] A.K. Sari, W. Rahayu, M. Bhatt, Archetype sub-ontology: improvingconstraint-based clinical knowledge model in electronic healthrecords, Knowledge Based Systems 26 (2012) 75–85.
[7] S. Yu, D. Berry, J. Bisbalz, An investigation of semantic links toarchetypes in an external clinical terminology through the construc-tion of terminological ‘‘shadows’’, in: IADIS Multi Conference onComputer Science and Information Systems, Freiburg, Germany,2010.
[8] A.K. Sari, W. Rahayu, A methodology for change propagation inhealth ontology, in: 15th Pacific Asia Conference on InformationSystems, Brisbane, 2011.
[9] D.E. Oliver, Y. Shahar, E.H. Shortliffe, M.A. Musen, Representation ofchange in controlled medical terminologies, Artificial Intelligence inMedicine 15 (1999) 53–76.
[10] A. Maedche, B. Motik, L. Stojanovic, R. Studer, R. Volz, Aninfrastructure for searching, reusing and evolving distributedontologies, in: Proceedings of WWW 2003, ACM Press, 2003,pp. 439–448.
[11] R. Palma, O. Corcho, A. Gmez-Prez, P. Haase, A holistic approach tocollaborative ontology development based on change management,Web Semantics: Science, Services and Agents on the World WideWeb. 9 (3) (2011).
[12] N.F. Noy, A. Chugh, W. Liu, M.A. Musen, A framework for ontologyevolution in collaborative environments, in: Proceedings of 5thInternational Semantic Web Conference, Lecture Notes in ComputerScience, Springer, 2006, pp. 544–558.
[13] T. Gabel, Y. Sure, J. Voelker, Kaon-Ontology Management Infrastruc-ture, SEKT Informal Deliverable, vol. 3, 2004.
[14] F. Zablith, Dynamic ontology evolution, in: ISWC Doctoral Consor-tium, 2008.
[15] A. Maedche, B. Motik, L. Stojanovic, N. Stojanovic, User-Driven Ontol-ogy Evolution Management, Springer-Verlag, 2002, pp. 285–300.
[16] P. Plessers, O.D. Troyer, Ontology change detection using a versionlog, in: Proceedings of the 4th International Semantic Web Con-ference, Springer, 2005, pp. 578–592.
[17] J. Mapoles, C. Smith, J. Cook, B. Levy, Strategies for updatingterminology mappings and subsets using SNOMED CT, in:R. Cornet, K. Spackman (Eds.), Proceedings of the 3rd InternationalConference on Knowledge Representation in Medicine, 2008.
[18] A. Maedche, B. Motik, L. Stojanovic, R. Studer, R. Volz, Managingmultiple ontologies and ontology evolution in ontologging, in:Ontologging. Intelligent Information Processing, Kluwer, 2002,pp. 51–63.
[19] D.E. Oliver, Y. Shahar, Change management of shared and localhealth-care terminologies, Methods of Information Medicine 39(2000) 278–290.
[20] P. Plessers, O.D. Troyer, S. Casteleyn, Understanding ontology evolu-tion: a change detection approach, Web Semantics: Science, Servicesand Agents on the World Wide Web 5 (1) (2007) 39–49.
[21] Y. Wang, X. Liu, R. Ye, Ontology evolution issues in adaptableinformation management systems, in: Proceedings of the 2008 IEEEInternational Conference on e-Business Engineering, IEEE ComputerSociety, Washington, DC, USA, 2008, pp. 753–758. http://dx.doi.org/10.1109/ICEBE.2008.69.
A.K. Sari et al. / Information Systems 38 (2013) 727–744744
[22] A. Shaban-Nejad, V. Haarslev, Bio-medical ontologies maintenanceand change management, in: A. Sidhu, T. Dillon (Eds.), BiomedicalData and Applications. Studies in Computational Intelligence, 2009.
[23] A.C. Yu, J.J. Cimino, A comparison of two methods for retrieving icd-9-cm data: the effect of using an ontology-based method forhandling terminology changes, Journal of Biomedical Informatics44 (2011) 289–298.
[24] J.J. Cimino, Formal descriptions and adaptive mechanisms forchanges in controlled medical vocabularies, Methods of Informationin Medicine 35 (1996) 202–210.
[25] C.J. Mungall, C. Batchelor, K. Eilbeck, Evolution of the sequenceontology terms and relationship, Journal of Biomedical Informatics44 (2011) 87–93.
[26] S. Schulz, R. Cornet, SNOMED CTS ontological commitment, in: B.Smith (Ed.), ICBO: International Conference on Biomedical Ontology,Lecture Notes in Computer Science, National Center for OntologicalResearch, Buffalo, New York, 2009, pp. 55–58.
[27] J. Davis, R. Studer, P. Warren, Semantic Web Technologies Trendsand Research in Ontology-Based Systems, John Wiley & Sons, Ltd.,West Sussex, 2006.
[28] M. Ehrig, Ontology Alignment: Bridging the Semantic Gap, SemanticWeb and Beyond Computing for Human Experience, Springer, 2007.
[29] M. Bhatt, W. Rahayu, S.P. Soni, C. Wouters, Ontology driven semanticprofiling and retrieval in medical information systems, Web Seman-tics: Science, Services and Agents on the World Wide Web 7 (6)(2009) 317–331.