Updating Ontology Alignment on the Relation Level based on

Ontology Evolution

Adrianna Kozierkiewicz

and Marcin Pietranik

Faculty of Computer Science and Management, Wroclaw University of Science and Technology,

Wybrzeze Wyspianskiego 27, 50-370 Wroclaw, Poland

Keywords:

Ontology Alignment, Ontology Evolution, Knowledge Management.

Abstract:

Ontologies are becoming a popular and convenient way for knowledge representation - they can store infor-

mation about objects and relations between them. However, nothing is constant and new information may

appear, therefore those alterations should be reﬂected both in an ontology as well as in alignment between two

ontologies if the knowledge about some domain is distributed in many sources. In the literature, it is possible

to ﬁnd approaches devoted to tracking changes in ontologies, but tools for updating ontology alignment are

limited, especially devoted to the level of relations. It became a motivation for this work, thus, the aim of this

paper is split into two parts. Firstly, we will present a criterion that tells us that modiﬁcation in an ontology

on the relation level is signiﬁcant, and how it inﬂuences the maintained alignment (also on the relation level).

Next, an algorithm for simple revalidation of existing mappings is proposed.

1 INTRODUCTION

Ontology alignment is a widely discussed and re-

searched topic. It addresses a seemingly simple prob-

lem of designating a mapping between two hetero-

geneous ontologies. Such mapping can be used to

”translate” content of one ontology to the content of

some other ontology. This possibility is invaluable

when some kind of communication of two indepen-

dently developed information systems is expected.

One cannot expect that such a system would uti-

lize a shared ontology as a backbone of their knowl-

edge bases. Different systems have different busi-

ness requirements and enforcing a common ontol-

ogy would eventually make it nearly impossible to

maintain ((Abbes and Gargouri, 2017), (Wang et al.,

2020)).

Designating a bridge between two ontologies (al-

though simple to understand) is a very difﬁcult task

in terms of both its semantic and computational com-

plexity. Informally speaking the task is to select parts

of two (or more) ontologies that express the same

parts of a modeled universe of discourse. In the lit-

erature, it is easy to ﬁnd a plethora of different pro-

cedures that address this problem ((Algergawy et al.,

https://orcid.org/0000-0001-8445-3979

https://orcid.org/0000-0003-4255-889X

2018)). However, in modern applications ((Kiour-

tis et al., 2019)) no one can expect that the underly-

ing ontologies will not change in time. Such a sit-

uation may potentially result in invalidation of the

designated alignment, which results in breaking the

communication between interacting information sys-

tems. This is especially visible on the level of rela-

tions, which is frequently omitted. None of the ana-

lyzed publications (which is presented in further parts

of the article) focus strictly on relations neither in the

context of their evolution, neither on the level of their

mappings.

For the better explanation of the considered in this

paper, problem let us follow Figure 1 which is an

easy example of evolving ontology and presents two

ontologies O

(n)

and O − 1 in a state m − 1 and m. On

the upper side of the picture between ontologies O

and O

the alignment on the relational level has been

designated. In both ontologies, the relation look after

appears and mapping between them is obvious. Re-

lation is mother is only less general than relation is

parent and between them also mapping has been de-

tected. On the bottom side of the picture ontology O

has been evolved. The relation look after has been

deleted which inﬂuences the current alignment (the

mapping can no longer exist). The new relation has

has appeared, however, it can not be connected with

any relation from ontology O

. The relation is parent

Kozierkiewicz, A. and Pietranik, M.

Updating Ontology Alignment on the Relation Level based on Ontology Evolution.

DOI: 10.5220/0009142002410248

In Proceedings of the 15th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2020), pages 241-248

ISBN: 978-989-758-421-3

241

has been changed by the concept modiﬁcation, how-

ever, it has not affected the alignment.

Figure 1: Examples of Ontologies Evaluation.

The obvious solution for the above problem is to re-

launch the accepted mapping procedure from scratch

for updated ontologies whenever they change. Such

an approach would give good results but entails bear-

ing the cost enforced by the mapping procedure. In

the article, we propose a different approach. We claim

that it is possible to update the possessed alignment

between two ontologies based solely on the analysis

of the applied changes. This task can be further de-

composed into two elements.

The ﬁrst of the aforementioned subproblems is

deciding whether or not the maintained alignment

should be checked for its validity. We claim that not

all alterations that are introduced to some ontology

during its evolution are equally signiﬁcant and not all

of them result in a situation in which some elements

of the alignment become stale or invalid. For exam-

ple, some small changes of some concept’s label are

not as inﬂuential as a major update of its attributes

and relations it is connected by with other concepts.

Since in this paper we are addressing only the level

of relations, formally, this task can be described as

follows: For a given ontology O in its two consecu-

tive states in time, denoted as O

(m)

and O

(m+1)

, one

should determine a function Ψ

representing the de-

gree of signiﬁcance to which relations within it have

been changed in time.

The second part is performing the actual revali-

dation. This task involves solving three problems:

(i) deleting stale mappings, (ii) revalidating and up-

dating existing mappings, and (iii) adding new map-

pings. Formally, it can be deﬁned as: For a given

source ontology O

(m)

in a moment in time denoted

as m, a target ontology O

(n)

in a moment in time de-

noted as n, and an alignment of their relations de-

noted as Align(O

(m)

(n)

), one should provide algo-

rithms which can update this alignment if the source

ontology signiﬁcantly evolves from the state O

(m)

the state O

(m+1)

according to applied changes.

The article is structured as follows. In Section 2

an overview of related works is given. Section 3 con-

tains basic notions and deﬁnitions utilized in the pa-

per. Section 4 provides the main contribution. It has

been split into two subparts. The ﬁrst contains a def-

inition of the developed function Ψ

that can be used

to indicate whether or not the alignment of relations

potentially needs revalidating. The second part pro-

vides algorithms for such revalidation. The devel-

oped procedures have been experimentally veriﬁed -

the obtained are described in Section 5. Our upcom-

ing research and a summary are given in Section 6.

2 RELATED WORKS

It is possible to ﬁnd many ontology alignment tools,

however, their efﬁciency may highly differ. The

choice of the best alignment tool is supported since

2004 by the Ontology Alignment Evaluation Initia-

tive (OAEI). OAEI coordinates an international ini-

tiative to evaluate, compare and improve the tools for

ontology mapping and alignment by providing stan-

dardized benchmark datasets.

Strong competitors of multiple OAEI editions

were tools like ALIN (Da Silva Jomar and Kate,

2018), AML (Faria et al., 2018), Kepler (Kachroudi

et al., 2018), LogMap (Jim

enez-Ruiz et al., 2018),

SANOM (Mohammadi et al., 2018). However, the

of the alignment system of object properties match-

ing, which can be understood as relations in terms of

OWL representation ((Hitzler et al., 2009)), lags sig-

niﬁcantly behind that on class and instance matching

(Cheatham et al., 2018). The average precision and

recall measures for class alignment obtained by the

top 2016 OAEI competitors on the Conference track

equal 0.82 and 0.65, respectively. For object prop-

erties alignment these measures are severely lower:

0.45 and 0.1, respectively.

The presented results in (Cheatham et al., 2018)

allow us to conclude that the accuracy of the veriﬁed

tools is almost three times worse for data and object

property in comparison with class alignment. Addi-

tionally, some of the presented systems do not gen-

erate any matches involving object properties at all.

This means that the alignment system’ designers fo-

cus their efforts mainly on class and instance level.

Therefore, the problem of determining reliable map-

ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering

242

pings for object properties is still open.

The systems mentioned above, allow the user to

determine a new alignment for given ontologies. This

paper is devoted to updating alignment in case of evo-

lution input one or more ontologies. It is rather ob-

vious, that we can use tools directly dedicated to an

ontology mapping and perform a process of determin-

ing the alignment from the beginning for the mod-

iﬁed ontologies. Such a solution is obviously cost-

and time-consuming. In the literature, it is possible to

ﬁnd some approaches which track ontology evolution

(i.e. (Abbes and Gargouri, 2017), (Noy et al., 2002)),

however, not many solutions are devoted to updating

the predesignated mappings.

Some of the tools for updating ontology align-

ments focus only on part of ontologies, especially on

concepts level. In (Dos Reis and Yamamoto, 2019),

(Dos Reis, 2018) authors proposed techniques to re-

ﬁne a set of established mappings based on the evolu-

tion of ontologies. Their algorithms process the con-

text of new concepts in both ontologies to ﬁnd new

matches between concepts. In other words, they sug-

gest new correspondences with the new version of the

ontology without applying a matching operation to

the whole ontologies.

Similarly in (Thenmozhi and Vivekanandan,

2012), the authors proposed a new approach which

is a semi-automatic way update mapping. The ontol-

ogy was constantly monitoring for any changes and

updating in the log, however, only information con-

cerning concepts are stored. Next, the proposed so-

lution removes staleness from the alignment based on

predeﬁned rules.

It is possible to ﬁnd complex tools for updating

ontology alignments like (Khattak et al., 2015). The

authors proposed an approach that employs the analy-

sis of ontology change history. These logged changes

are later used with extensions to the existing map-

ping systems during the reconciliation of the map-

ping process. The three change types have been dis-

tinguished like create (such as ClassAddition, Prop-

ertyAddition, and IndividualAddition), update (such

as ClassRenaming, PropertyRenaming, and Individu-

alRenaming) and delete (such as ClassDeletion, Prop-

ertyDeletion, and IndividualDeletion).

In (Hartung et al., 2013) COnto-Diff tool was de-

scribed. Although COnto-Difff determined an expres-

sive and invertible diff evolution mapping between

two versions of the same ontology, authors noticed,

that it can also be used to semi-automatically adapt

annotations and ontology mappings after ontology

modiﬁcations. The solution took into account the in-

sertion and deletion of relationships.

In (Dos Reis et al., 2013) a set of mapping adapta-

tion actions to maintain mappings up-to-date based on

ontology change operations of different natures have

been proposed. Two types of ontology change opera-

tions: atomic and complex have been deﬁned. Based

on them an overall adaptation of mappings accord-

ing to the revision of knowledge in ontology was pro-

posed.

Noteworthy is the fact, that all of the described

methods concentrate on the concept level. This has

two implications. First, the level of relations is fre-

quently omitted. None of the analyzed publications

focus strictly on relations neither in the context of

their evolution, neither on the level of their map-

pings. Secondly, overviewed solutions of ontology

mapping evolution look inside of the content of evolv-

ing source concepts, and not holistically analyze on-

tologies. Ontologies have dynamic nature, which

directly impacts mappings established between con-

cepts, relations, and instances from different ontolo-

gies. Although many pieces of research are aware of

the need to track the changes in ontologies, the prob-

lem of maintaining the actual mappings between in-

put ontologies is still open. These remarks became

a motivation behind this paper, which addresses the

problem of updating ontology alignment strictly on

the relation level.

3 BASIC NOTIONS

The solution proposed in this paper for updating on-

tology alignment is based on our formal model of an

ontology deﬁned as a quintuple:

O = (C,H,R

,I,R

) (1)

where: C is a set of concepts; H is a concepts’ hi-

erarchy; R

is a set of relations between concepts

= {r

,...,r

}, n ∈ N (natural numbers), such

that r

∈ R

(i ∈ [1, n]) is a subset of C ×C; I is a set

of instances’ identiﬁers; R

= {r

,..., r

} is a set of

relations between concepts’ instances.

In the following article we focus on the relation

level, therefore, provided deﬁnitions concern only el-

ements related to this level. By D

we deﬁne a set

containing atomic descriptions of relations. Subse-

quently, we deﬁne a sub-language of the sentence cal-

culus built from elements of D

and logic operators

of conjunction, disjunction and negation. We denote

it as L

. It is used within a function that assigns se-

mantics to relations from the set R

. This function

has the following signature:

: R

→ L

(2)

As a consequence, we can deﬁne formal criteria for

relationships between relations:

Updating Ontology Alignment on the Relation Level based on Ontology Evolution

243

• equivalency between relations r and r’ (denoted as

r ≡ r

) occurs only if a sentence S

(r) ⇐⇒ S

)

is a tautology

• a relation r’ is more general than the relation r

(denoted as r

← r) if a sentence S

(r) =⇒ S

)

is a tautology

• contradiction between relations r and r’ (denoted

as r ∼ r

) occurs only if a sentence ¬(S

(r) ∧

)) is a tautology

The formal deﬁnition of an ontology on the rela-

tion level allows us to track changes in the ontol-

ogy. The ontology evolution is based on a notion

of a timeline, which can be treated as an ordered set

of discrete moments in time. It can be deﬁned as

T L = {t

|n ∈ N}. T L(O) is a subset of this time-

line, containing only those moments from T L dur-

ing which the ontology O has been changed. By

(m)

= (C

(m)

C(m)

(m)

I(m)

) we denote the

ontology O in a given moment in time t

∈ T L(O).

(m−1)

≺ O

(m)

means that O

(m)

is a subsequent ver-

sion of O than O

(m−1)

In order to compare two states of a single on-

tology O in our previous publication (Kozierkiewicz

and Pietranik, 2019) we introduced a function di f f

which, when fed with two successive states O

(m−1)

and O

(m)

of one single ontology (such that O

(m−1)

≺

(m)

), generates three sets containing relations added,

deleted and altered. Formally, these sets are deﬁned

below:

1. new

C(m−1)

C(m)

) =



r|r ∈ R

C(m)

∧ r /∈ R

C(m−1)



2. del

C(m−1)

C(m)

) =



r|r ∈ R

C(m−1)

∧ r /∈ R

C(m)



3. alt

C(m−1)

C(m)

) =



(m−1)

(m)

)|r

(m−1)

∈ R

C(m−1)

∧ r

(m)

∈ R

C(m)

∧

(m−1)

⊕ r

(m)

6= φ ∨ S

(m−1)

) 6= S

(m)

))



The ﬁrst two descriptors are self-explanatory. The

last one represents a situation where some pairs of

concepts within such relation have been added or re-

moved or some alterations to a semantic of relations

expressed using the deﬁned function S

(⊕ represent

the exclusive or operator). For a detailed descrip-

tion of the holistic approach to managing ontologies

in time please refer to (Kozierkiewicz and Pietranik,

2019).

Having two ontologies O

= (C

)

and O

= (C

) a relations alignment

between them is a set:

) = {(r

,λ

), ˜r)|

∈ R

∧ r

∈ R

∧ λ

) ≥ T

}

(3)

where: λ

is a degree to which relation r

can be

aligned to relation r

, ˜r represents the type of map-

ping (equivalency, generalization etc.) and T

is some

assumed threshold. For short we write (r

) ∈

) to indicate that a relation r

can be

aligned to relation r

to some degree higher that the

assumed threshold T

by equivalency relationship.

The set from Equation 3 fulﬁlls the requirement

for alignment completeness which states that if some

relation r

taken from the ﬁrst ontology O

is more

general than some other relation r

, then they can be

both mapped to a concept r

from the second ontology

. Formally it can be deﬁned as follows:

¬∃(r

) ∈ R

× R

: λ

) ≥ T

∧(r

,λ

), ˜r) /∈ AL

)

(4)

4 UPDATING ONTOLOGY

ALIGNMENT ON A RELATION

LEVEL

4.1 The Degree of Change Signiﬁcance

on the Ontology Relation Level

Having an ontology O in its two subsequent states

(m−1)

and O

(m)

, such that O

(m−1)

≺ O

(m)

, and a re-

lations difference function di f f

we can deﬁne the

degree of signiﬁcancy to which relations within a

given ontology have been changed in time. This

function utilizes a function d

which takes as an in-

put two logical sentences and returns a distance be-

tween them based on the analysis of their elements

taken from the set D

deﬁned in the previous section.

Formally, it can be deﬁned as follows:

: R

C(m−1)

× R

C(m)

→ [0,1] (5)

C(m−1)

C(m)

) =

|new

C(m−1)

C(m)

)| + |del

C(m−1)

C(m)

| + |del

C(m−1)

C(m)

∑

)∈alt

C(m−1)

C(m)

)

),S

))

C(m)

| + |del

C(m−1)

C(m)

(6)

The function above meets the following two postu-

lates:

ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering

244

• P1. Ψ

C(m−1)

C(m)

) = 0 ⇐⇒

di f f

C(m−1)

× R

C(m)

) =



φ,φ, φ



• P2. Ψ

C(m−1)

C(m)

) = 1 ⇐⇒

del

C(m−1)

C(m)

) = R

C(m−1)

∧

∧ alt

C(m−1)

C(m)

) = φ

P1 addresses a situation in which no alterations on the

relation level have been applied - no new relations ap-

peared, no relation was removed, no relation changed.

Then, the signiﬁcance of a change is minimal. P2

describes an opposite situation, in which the change

signiﬁcance is maximal. It occurs when the ontology

has been completely modiﬁed on the relation level -

every relation from the earlier state has been deleted

and every relation in a later state is new.

Having the function deﬁned in Equation 6 it is

straightforward to confront it with some assumed

threshold. It allows deciding whether or not the align-

ment designated between the tracked ontology and

some other ontology needs revalidation. If such a ne-

cessity appears, algorithms presented in the next sec-

tion should be utilized.

4.2 Procedures for Updating Ontology

Alignment on Relation Level

Having two ontologies O

(m−1)

) and

(n)

= (C

(n)

), a rela-

tions alignment between them given as a set

(m−1)

(n)

), a description of updates applied

to O

given as a function di f f (O

(m−1)

(m)

), an al-

gorithm that updates the alignment AL

(m−1)

(n)

)

to its new state AL

(m)

(n)

) can be split into three

separate scenarios:

1. deleting stale mappings

2. revalidating and updating existing mappings

3. adding new mappings

The ﬁrst procedure presented on Algorithm 1 is very

simple. It addresses the most basic situation when

some relations have been removed from the source

ontology. In this case, all related mappings (des-

ignated in Line 2) should also be discarded (which

is done in Line 3), because they connect something

which doesn’t exist anymore.

The last two scenarios both end with searching

for new mappings (due to adding new relations to the

source ontology or modifying existing relations in

such a way that new mappings may be added). Due to

Algorithm 1: Removing Stale Mappings of Deleted

Relations from the Existing Alignment.

Input : AL

(m−1)

(n)

),di f f

(m−1)

(m)

)

Output: AL

(m)

(n)

)

1 begin

del :=



,λ

), ˜r)|(r

,λ

), ˜r) ∈

(m−1)

(n)

) ∧ r

∈

del

(m−1)

(m)

)



;

3 AL

(m)

(n)

) :=

(m−1)

(n)

) \

del;

4 return AL

(m)

(n)

);

5 end

this, both can be implemented within the same proce-

dure presented on Algorithm 2. It starts with accept-

ing the unmodiﬁed alignment of ontologies in their

earlier states (m-1 and n) as its initial state (Line 2).

Then, (Line 3) it designates a set of alignments that

refer to relations modiﬁed in the later state (m). El-

ements of this set are later confronted (Line 4) with

some assumed threshold - if it is not exceeded then

such element is removed from the alignment. Sub-

sequently, in Line 5, the algorithm generates a set of

altered relations from the source ontology that was

not mapped into any relation in the target ontology.

This set is then combined with a set of relations newly

added to the source ontology (Line 6).

Eventually, the algorithm enters (Line 7) a sub-

procedure that ﬁnds new potential mappings. At

this point it is possible to use any kind of available

alignment procedure. The algorithm is fully agnos-

tic and this step is easily swappable. However, for

the demonstration purposes, in this paper we use an

alignment procedure developed in our previous pub-

lications ((Pietranik and Nguyen, 2014)) that is based

on comparing relation semantics from Equation 2 us-

ing a d

distance function that is also used in Equa-

tion 6. Basically, the algorithm traverses through the

set of relation available in the target ontology (Line

8) and checks if the degree to which a relation from

the source ontology can be aligned to the relation

from the target ontology exceeds the assumed thresh-

old (Line 9). If yes, then the new mapping is added

to the ﬁnal alignment (Line 10). The sub-procedure

ends with complementing the relation alignment to

fulﬁll the completeness postulate from Equation 4. In

Line 13 the set of relations that are less or more gen-

eral than the current relation is designated. Then, in

Line 14, the procedure checks if the current relation

has any mappings. If this is the case, then appropri-

Updating Ontology Alignment on the Relation Level based on Ontology Evolution

245

ate mappings are added for related relations (Line 15).

This step is done to avoid searching for mappings for

relations that are connected with a generalization re-

lationship with the current one and to automatically

add them to the resulting set of mappings. Therefore,

omitting unnecessary calculations.

The presented procedure is quite complex in terms

of a number of iterations. However, ontology align-

ments usually do not contain many mappings con-

necting relations (as shown in Section 2). These sets

are frequently very limited in terms of their size. It is

not uncommon that they contain only a few mappings

(or even none), therefore, the presented procedure for

small alterations of the source ontology can become

very useful.

5 EXPERIMENTAL

VERIFICATION

Our procedure for updating ontology alignment on

the relation level has been experimentally veriﬁed.

For this task benchmark datasets provided by On-

tology Alignment Evaluation Initiative (OAEI) have

been used. From all available datasets ”the Confer-

ence Track” has been chosen which describes the do-

main of organizing conferences.

Updating alignment on the relation level has

been compared with mapping determined by using

LogMap (Jim

enez-Ruiz and Grau, 2011). LogMap

is an ontology alignment and alignment repair sys-

tem which earned high positions in subsequent OAEI

campaigns. However, it is worthy of notice, that

the accuracy of current ontology mapping systems on

property (in our work called as the relation) align-

ment is not very high. As it was mentioned before,

in (Cheatham et al., 2018) authors show that the aver-

age precision and recall measures for properties align-

ment equal 0.45 and 0.17, respectively. For LogMap

these measures are correspondingly higher: 0.62 and

0.28. However, the presented results demonstrate that

the mappings tool on the relation level are not efﬁcient

and they are required to be improved.

In the ﬁrst part of our experiment, we have cho-

sen from the datasets a source ontology (called CMT )

and a target ontology (called MyReview). The source

ontology has been modiﬁed randomly. For the com-

parison of both tested methods, we have used an accu-

racy measure. It is calculated as the number of com-

mon links (relations’ mappings) between two ontolo-

gies divided by the number of all connections found

by both methods:

Algorithm 2: Revalidating the Existing Alignment

and Adding New Mappings.

Input : AL

(m−1)

(n)

),di f f

(m−1)

(m)

)

Output: AL

(m)

(n)

)

1 begin

2 AL

(m)

(n)

) := AL

(m−1)

(n)

);

alt =



,λ

), ˜r)|(r

,λ

), ˜r) ∈

(m−1)

(n)

) ∧ r ∈

alt

(m−1)

(m)

)



4 AL

(m)

(n)

) := AL

(m)

(n)

) \



,λ

), ˜r)|(r

,λ

), ˜r) ∈

alt ∧ λ

) < T



;

alt



r|r ∈ alt

(m−1)

(m)

) ∧

¬∃(r,r

,λ

(r,r

), ˜r) ∈

(m−1)

(n)

)



new =

alt

∪ new

(m−1)

(m)

)

7 for r ∈

new do

8 for r

∈ R

(n)

9 if λ

(r,r

) ≥ T

then

10 AL

(m)

(n)

) :=

(m)

(n)

) ∪



(r,r

,λ

(r,r

), ˜r)



11 end

12 end

13 if ∃r

∈ R

(m)

: ((r

← r) ∨ (r ←

)) ∧ (r 6= r

) then

14 if ∃(r,r

,λ

(r,r

), ˜r) ∈

(m−1)

(n)

) then

15 AL

(m)

(n)

) :=

(m)

(n)

) ∪



,λ

), ˜r)



16 end

17 end

18 end

19 return AL

(m)

(n)

);

20 end

(

|Align

LogMap

(m+1)

(n)

) ∩ Align(O

(m+1)

(n)

|Align

LogMap

(m+1)

(n)

) ∪ Align(O

(m+1)

(n)

(7)

where Align

LogMap

is a set of mappings created by

LogMap, and Align is a set of mappings determined

by our updating procedure

Our approach described in Section 4.2 distin-

guishes three ontology evolution on the relational

ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering

246

level which can inﬂuence existing mappings time-

liness: adding new relations, removing existing re-

lations and modiﬁcation of existing relations. By

modiﬁcation of relations, we understand a process

of adding or removing concepts connected by such

relations or a process of changing the semantics of

those relations. In our work, we have focused only on

how adding new relations in a base ontology inﬂuence

changes of alignments. The removing and modiﬁca-

tion of existing relations and their impact on updating

alignment have been omitted, because the LogMap

tool determines the mapping from the beginning. If

any relation does not exist anymore it is obvious that

any mapping does not appear. Thus, the efﬁciency of

updating alignment in case of removing or modiﬁca-

tion of relations doing by the LogMap tool and our

approach is perfect.

In the case of adding new relations, a divergence

between the results of the mentioned approach is big-

ger. In the ﬁrst part of our experiment, we verify how

changes in base ontologies inﬂuence changes of align-

ments. At the beginning only a 10% new relations

have been added and in the end, the number of rela-

tions in the source ontology has been doubled. The

results clearly showed that the number of introduced

modiﬁcations didn’t have any signiﬁcant impact on

the value of accuracy - for each iteration we obtained

a value between 75% and 83%.

The conducted experiment allows us to conclude

that our approach and LogMap give similar align-

ments (in terms of the accuracy measure deﬁned in

Equation 7) of two ontologies. However, our ap-

proach found more correct connections between re-

lations. As it was mentioned before, the efﬁciency of

LogMap on the relational level is not sufﬁcient.

Thus, the smaller number of mappings generated

by LogMap in comparison with our solution con-

ﬁrmed the correctness of our assumptions and the

results of the experiment and demonstrated the efﬁ-

ciency of our approach. Furthermore, updating an ex-

isting alignment is less expensive than building new

mappings from the beginning which combined with

the low computational complexity of our algorithm

makes it a more attractive solution than existing so

far.

In the second part of our experiment, we have not

changed a target ontology. However, we have applied

the same amount of modiﬁcations in different ontolo-

gies from the OAEI dataset which has been treated

as a source ontology. Thus, we would like to verify

how the same changes inﬂuence on updating align-

ment. The obtained results are presented in Table 1.

As in previous research, our approach found more

correct links between relations than the LogMap tool.

The accuracy of results obtained by both methods is

usually very high (around 80 %) which proves the

utility and correctness of our approach.

6 FUTURE WORKS AND

SUMMARY

The paper is devoted to the problem of updating ontol-

ogy alignment in the case of ontology evolution. This

is still an open problem and not well investigated by

other researches. So far, it is possible to ﬁnd some

tools dedicated to the mentioned problem. However,

many solutions focus only on concept or instances

level, and push the relation level to the background,

treating it as less important.

In this paper, we deﬁned the function which re-

ﬂects the degree of signiﬁcance to which relations

within a given ontology have been changed in time.

This function served as the criterion for deciding

whether or not the designated alignment (between

the tracked and some other ontology) needs updating.

The main result of our work is an algorithm that reval-

idated the existing alignment by removing stale and

adding new mappings.

Our method has been experimentally veriﬁed in

comparison with the well-known tool called LogMap.

The mappings updated by our method have been jux-

taposed with mappings designated from the beginning

by LogMap. The experiments show us that our ap-

proach returned a similar result like LogMap. How-

ever, our procedure found more correct link between

relation than LogMap. It proves the correctness and

efﬁciency of our approach.

In the nearest future work, we want to elaborate

on the ontology alignment updating procedure for the

instance level. If all levels of an ontology will be cov-

ered by proper alignment revalidating procedures then

we will plan to implement and prepared the complex

tool for ontology storing, tracking changes, knowl-

edge integration, mappings determination and actu-

alization in case of ontology evolution.

ACKNOWLEDGEMENTS

This research project was supported by grant No.

2017/26/D/ST6/00251 from the National Science

Centre, Poland.

Updating Ontology Alignment on the Relation Level based on Ontology Evolution

247

Table 1: Different Source Ontologies.

Source

ontology

Number of

relations in

the source ontology

Number of maps

found by

the proposed approach

Number of maps

found by LogMap

Accuracy

CMT 49 7 5 0.714

Cocus 35 6 4 0.667

Conﬁous 52 7 6 0.857

ConfTool 13 5 4 0.8

Crs 15 9 8 0.89

Edas 30 8 8 1

Ekaw 33 5 4 0.8

Iasted 38 5 4 0.8

Linklings 31 5 4 0.8

Micro 17 7 7 1

OpenConf 24 9 8 0.89

PCS 24 11 5 0.455

Sigkdd 17 6 5 0.833

Sofsem 64 9 8 0.889

REFERENCES

Abbes, H. and Gargouri, F. (2017). Managing modular on-

tology evolution under big data integration. In Euro-

pean, Mediterranean, and Middle Eastern Conference

on Information Systems, pages 17–28. Springer.

Algergawy, A. et al. (2018). Results of the ontology align-

ment evaluation initiative 2018. In 13th International

Workshop on Ontology Matching co-located with the

17th ISWC (OM 2018), pages 76–116.

Cheatham, M., Pesquita, C., Oliveira, D., and McCurdy,

H. B. (2018). The properties of property alignment

on the semantic web. IJMSO, 13(1):42–56.

Da Silva Jomar, nd Baiao, F. A. and Kate, R. (2018). Alin

results for oaei 2018. In 13th International Workshop

on Ontology Matching co-located with the 17th ISWC

(OM 2018), pages 117–124.

Dos Reis, J. C. (2018). Mapping reﬁnement based on on-

tology evolution: A context-matching approach. In

ONTOBRAS, pages 251–256.

Dos Reis, J. C., Dinh, D., Pruski, C., Da Silveira, M., and

Reynaud-Dela

ıtre, C. (2013). Mapping adaptation ac-

tions for the automatic reconciliation of dynamic on-

tologies. In Proceedings of the 22nd ACM interna-

tional conference on Information & Knowledge Man-

agement, pages 599–608. ACM.

Dos Reis, J. C. and Yamamoto, V. E. (2019). Updat-

ing ontology alignments based on new concepts and

their context. In Revista dos Trabalhos de Iniciac¸

Cient

ıﬁca da UNICAMP, Campinas, SP, volume 27.

Faria, D., Pesquita, C., Balasubramani, B. S., Tervo, T.,

Carric¸o, D., Garrilha, R., Couto, F. M., and Cruz, I. F.

(2018). Results of aml participation in oaei 2018.

In 13th International Workshop on Ontology Match-

ing co-located with the 17th ISWC (OM 2018), pages

125–131.

Hartung, M., Groß, A., and Rahm, E. (2013). Conto–diff:

generation of complex evolution mappings for life sci-

ence ontologies. Journal of biomedical informatics,

46(1):15–32.

Hitzler, P., Kr

otzsch, M., Parsia, B., Patel-Schneider, P. F.,

Rudolph, S., et al. (2009). Owl 2 web ontology lan-

guage primer. W3C recommendation, 27(1):123.

Jim

enez-Ruiz, E. and Grau, B. C. (2011). Logmap: Logic-

based and scalable ontology matching. In Inter-

national Semantic Web Conference, pages 273–288.

Springer.

Jim

enez-Ruiz, E., Grau, B. C., and Cross, V. (2018).

Logmap family participation in the oaei 2018. In 13th

International Workshop on Ontology Matching co-

located with the 17th ISWC (OM 2018), pages 173–

178.

Kachroudi, M., Diallo, G., and Yahia, S. B. (2018). Kepler

at oaei 2018. In 13th International Workshop on On-

tology Matching co-located with the 17th ISWC (OM

2018), pages 173–178.

Khattak, A. M., Pervez, Z., Khan, W. A., Khan, A. M.,

Latif, K., and Lee, S. (2015). Mapping evolution

of dynamic web ontologies. Information Sciences,

303:101–119.

Kiourtis, A., Mavrogiorgou, A., Menychtas, A., Maglogian-

nis, I., and Kyriazis, D. (2019). Structurally mapping

healthcare data to hl7 fhir through ontology align-

ment. Journal of medical systems, 43(3):62.

Kozierkiewicz, A. and Pietranik, M. (2019). A formal

framework for the ontology evolution. In Asian Con-

ference on Intelligent Information and Database Sys-

tems, pages 16–27. Springer.

Mohammadi, M., Hofman, W., and Tan, Y.-H. (2018).

Sanom results for oaei 2018. In 13th International

Workshop on Ontology Matching co-located with the

17th ISWC (OM 2018), pages 125–131.

Noy, N. F., Musen, M. A., et al. (2002). Promptdiff:

A ﬁxed-point algorithm for comparing ontology ver-

sions. AAAI/IAAI, 2002:744–750.

Pietranik, M. and Nguyen, N. T. (2014). A multi-attribute

based framework for ontology aligning. Neurocom-

puting, 146:276–290.

Thenmozhi, M. and Vivekanandan, K. (2012). A semi-

automatic approach to update mapping for ontology

evolution. In Int. Conf. on Computational Intelligence

and Information Technology, CIIT, pages 319–324.

Wang, T., Gu, H., Wu, Z., and Gao, J. (2020). Multi-source

knowledge integration based on machine learning al-

gorithms for domain ontology. Neural Computing and

Applications, 32(1):235–245.

ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering

248