OMNIMOD: Automating Ontology Modularization for Digital Library

Data Using CIDOC-CRM as Use Case

Giulia Biagioni

Unit ISP, Netherlands Organisation for Applied Scientiﬁc Research (TNO), New Babylon,

Anna van Buerenplein 41a, Den Haag, The Netherlands

Keywords:

Ontology Modularization Method, Evaluation Method for Ontology Modularization, Cognitive Load Theory.

Abstract:

This paper introduces OMNIMOD, a new method designed to modularize ontologies, RDF-based structures

that organize knowledge within specialized domains. By simplifying complex information into manageable

components, OMNIMOD enhances the analysis, understandability, and navigation of large ontological frame-

works while also extending its functionality to include the modularization of associated data records, known

as instance data. The method has been developed based on theoretical insights gathered from Cognitive Load

Theory (CLT) and has been successfully tested and applied to CIDOC-CRM (Conceptual Reference Model of

the International Committee for Documentation), the ISO standard for describing data related to cultural her-

itage materials. The accompanying Python functions, developed for OMNIMOD and provided in the corpus

of the text, empower readers to adapt and utilize OMNIMOD according to their speciﬁc needs.

1 INTRODUCTION

Ontology modularization, which involves breaking

down an ontology into smaller, more manageable

parts, is particularly valuable for large ontologies

(Parent and Spaccapietra, 2009). This technique im-

proves reasoning capabilities (Gatens et al., 2014),

simpliﬁes ontology maintenance (Sarkar and Dong,

2011), enhances alignment with external ontologies

(Ghafourian et al., 2013), assists in the structural ex-

amination of the taxonomy and axioms (Vescovo

et al., 2011a), and allows for selective access to spe-

ciﬁc sections of the ontology while concealing oth-

ers (Konev et al., 2009). As repositories of knowl-

edge that enable meaningful communication between

agents, ontologies are naturally designed to grow and

incorporate an ever-expanding body of knowledge.

This expansion is further motivated by European ini-

tiatives focused on establishing data spaces and dig-

ital libraries

grounded in ontologies, aiming to or-

ganize and utilize data more efﬁciently (Corcho and

Simper, 2022; Commission, 2022).

Despite its recognized importance within the sci-

entiﬁc community, the ﬁeld of ontology modulariza-

https://orcid.org/0000-0002-9005-7945

An example of European digital library for cultural

heritage material is provided by https://www.europeana.eu/

tion faces signiﬁcant hurdles, as outlined by LeClair

and colleagues (LeClair et al., 2023). These chal-

lenges span from the absence of uniform guidelines

for conducting the modularization process to the need

for more intuitive methods for examining modular-

ized entities (d’Aquin et al., 2009; LeClair et al.,

2023). Additionally, there is also a lack of techniques

that take into account the data records structured using

the ontology in the modularization process, as well

as standardized methods to evaluate the modulariza-

tion process. Such obstacles markedly hinder the pro-

cesses of navigation, analysis, and the assessment of

intricate knowledge structures.

To address these issues, this study presents a new

methodology for ontology modularization referred to

here as OMNIMOD. The approach centers around

CIDOC-CRM for the description of cultural heritage

assets as primary use case. By doing this, the cur-

rent initiative pays a double contribution. On the one

hand, it aims to reﬁne modularization techniques and

enhance the management, understanding, and practi-

cal application of extensive ontologies. On the other

hand, it seeks to provide cultural heritage profession-

als and users of extensive ontological conceptualiza-

tions with speciﬁcally designed tools for automated

ontology modularization, enhancing their ability to

access, comprehend, and work within large ontolog-

102

Biagioni, G.

OMNIMOD: Automating Ontology Modularization for Digital Library Data Using CIDOC-CRM as Use Case.

DOI: 10.5220/0013011200003825

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 20th International Conference on Web Information Systems and Technologies (WEBIST 2024), pages 102-111

ISBN: 978-989-758-718-4; ISSN: 2184-3252

ical frameworks.

The methodology behind the

development of OMNIMOD is informed by insights

from cognitive load theory (CLT).

CLT is a conceptual framework that describes how

the human brain processes and retains new informa-

tion. It offers strategies for organizing and present-

ing information in ways that enhance users’ ability

to assimilate new data effectively while fostering a

deeper understanding of the material. By including

principles of CLT in its design, OMNIMOD lever-

ages achievements in the humanities to enhance its

technical implementation and formulate the theoret-

ical strategy that underpins its approach. The name

”OMNIMOD,” which stands for Ontology Modular-

ization through an Integrated Methodological Design,

reﬂects this comprehensive approach. The efﬁciency

of OMNIMOD has been evaluated based on two crite-

ria: cohesion and coupling. The results are presented

in the paper.

This study is structured as follows: beyond the in-

troduction already presented, section 2 delineates the

state of the art, providing a critical examination of

existing methodologies and identifying the need for

advancements in ontology modularization. Section 3

discusses the methodology employed to develop OM-

NIMOD, which integrates both the theoretical frame-

works that guided its design and the context for its

creation. Finally, section 4 introduces OMNIMOD

itself, detailing both its formal deﬁnition and techni-

cal implementation. The metrics used to evaluate the

method are presented in section 5. The validation pro-

cess of OMNIMOD, tested with CIDOC-CRM as use

case, is outlined in Section 6. Ultimately, section 7

further discusses the results presented in Section 6.

The modularization method presented in this study

gains additional relevance in light of Directive 2013/37/UE.

Enforced in 2021, the directive requires cultural heritage in-

stitutions to make their data machine-understandable, inter-

connected, and openly accessible in digital libraries. Adopt-

ing standardized, comprehensive ontologies like CIDOC-

CRM presents a signiﬁcant challenge for professionals

tasked with navigating and adapting these broad ontolog-

ical frameworks to publish their data on both digital plat-

forms and digital libraries (such as Europeana). There-

fore, the tools and methodologies developed in this study

are not only designed to overcome the current limitations

in ontology modularization but also to support cultural her-

itage professionals in fulﬁlling the mandates of Directive

2013/37/UE, ensuring they are provided with the possibility

to access and understand how to efﬁciently and consistently

employ extensive vocabularies.

2 STATE OF THE ART

Ontology modularization strategies can vary widely

based on several factors. They might use semantic

or syntactic criteria to determine how to divide the

ontology into smaller, more manageable parts. The

approach to modularization can be guided by either

informal or formal speciﬁcations. In terms of im-

plementation, these strategies range from fully auto-

mated processes, where software tools do the work, to

partially automated methods that require some human

intervention, and entirely manual approaches where

the modularization is manually implemented. The

objectives behind modularizing an ontology can also

differ signiﬁcantly, inﬂuencing the chosen strategy.

For instance, the aim might be to enhance the ontol-

ogy’s usability, improve reasoning efﬁciency, or facil-

itate easier maintenance and updates. Another criti-

cal consideration is how the resulting modules inter-

act with each other, whether they should be disjoint,

meaning no overlap in concepts or properties, or if

overlapping modules are permissible.

Methods that employ syntactic strategies are com-

monly referred to as graph-based approaches, focus-

ing on the structural relationships and connections

within the ontology represented as a graph. These ap-

proaches analyze the topology and connectivity of the

ontology graph to identify modular structures. Exam-

ples of the application of these methods are provided

by (Noy and Musen, 2009; Sarkar and Dong, 2011;

Kachroudi et al., 2013). On the other hand, meth-

ods that utilize semantic criteria are known as logi-

cal approaches. These approaches consider the mean-

ing and semantics of the ontology’s concepts and re-

lationships to modularize the ontology based on its

content. Examples of how these methods have been

applied can be drawn from (Vescovo et al., 2011b;

Kontchakov et al., 2010; Grau et al., 2009). Ulti-

mately, a third category, referred to as hybrid meth-

ods, is identiﬁed in (LeClair et al., 2023). These in-

corporate elements of both syntactic and semantic ap-

proaches by leveraging both the structural properties

of the ontology graph and the semantic information

encoded within it. An instance of this approach is ex-

empliﬁed in (Leclair et al., 2019).

According to insights from the ﬁeld, there are

several areas where existing ontology modularization

methods could be signiﬁcantly improved. A primary

suggestion is to extend the focus beyond merely mod-

ularizing the ontology to include the associated data

records (LeClair et al., 2023). This wider scope is

expected to ensure a more exhaustive coverage of the

domain described within the modules. Moreover, the

establishment of clear, unambiguous criteria for eval-

OMNIMOD: Automating Ontology Modularization for Digital Library Data Using CIDOC-CRM as Use Case

103

uating the modularization process is emphasized as a

critical need.

Therefore, to advance research in current ontol-

ogy modularization methods, OMNIMOD has been

developed to test the efﬁciency of and create a new

modularization approach. This includes assessing

its limitations, exploring potential alternative paths,

and determining applicable evaluation criteria for the

method and its scoped functionalities. The evalua-

tion criteria to assess the modularization process of

OMNIMOD are presented in this paper to further for-

malize and present a possible approach that can be

followed to evaluate methods for ontology modular-

ization. The methodology and theoretical underpin-

nings used to design OMNIMOD are detailed in the

subsequent section.

3 METHODOLOGY

This section presents the methodology behind the de-

velopment of OMNIMOD. Section 3.1 outlines the

contextual factors that inspired the creation of OMN-

IMOD. Section 3.2 delves into the theoretical back-

bone of the methodology, harnessing insights from

CLT to inform the design and structuring principles

followed by OMNIMOD.

3.1 Context

To contribute to the advancement of the ﬁeld, this

study has implemented an automated approach for

modularizing large ontologies into modules. This ap-

proach was developed within the framework of the

research activities conducted under the Norm En-

gneering program, hosted by the netherlands organi-

zation for applied scientiﬁc research (TNO). Within

this program, speciﬁc ontologies have been created

to facilitate the interpretation of norms, such as

FLINT (Breteler et al., 2023), and Source ontology

The Source ontology, designed to manage tex-

tual information found in both digital and physical

libraries regarding norms and regulations, aligns with

key standards in the cultural heritage sector, including

CIDOC-CRM

, and some of its extensions (i.e. the

FLINT and Source ontologies can be accessed at

https://gitlab.com/normativesystems

CIDOC-CRM serves as the standard vocabulary for

structuring data related to the collections of cultural her-

itage institutions. Initially developed in 1994 by an inter-

national committee of experts convened under ICOM, the

model was technically expressed in Telos. Since then, it has

evolved into various formats, including RDF(S) and OWL

standards. Recognized as an ISO standard since 2009, the

Functional Requirements for Bibliographic Records,

FRBR,

, and FRBR’s lightweight variant, LRM)

To align the Source ontology with the standard-

ized conceptual frameworks used within (digital) li-

braries, an in-depth investigation into CIDOC-CRM

was conducted, inter alia. This process required

breaking down the ontology into modules to analyze

each class deﬁnition and its relationships in order to

gain deeper insights into the conceptualizations and

how they complement each other. Given the scarcity

of existing automated methods suitable for modular-

izing ontologies and knowledge graphs, and facilitat-

ing in-depth examination of conceptualizations, OM-

NIMOD was developed. The development of this

method has been steered by foundational principles of

ontological knowledge organization, with a particular

emphasis on class deﬁnitions, hierarchical structures,

and class relationships. Therefore, while the method

considers the semantic scope of each module, its core

approach remains graph-based.

3.2 Theoretical Framework for the

Modularization Strategy

OMNIMOD was developed with the aim to improve

navigability, information retrieval, analysis and ex-

amination of extensive ontologies. On a theoretical

level, the creation of modularization method was in-

formed by ﬁndings from the ﬁeld of CLT, namely, a

theoretical framework that examines how information

is processed within recipients’ working memory and

derives ways to present and organize information to

maximize its uptake. Developed by John Sweller in

the 1980s, CLT suggests that our ability to process

and integrate data is inﬂuenced by the capacity of

our working memory, which can be overloaded if too

much information is presented at once (Sweller et al.,

1998; Sweller et al., 2019). The theory divides cogni-

tive load into three main types: intrinsic, extraneous,

and germane.

Intrinsic load is the inherent difﬁculty associated

with a speciﬁc topic. It is directly related to the com-

model serves as a fundamental framework within digital

European libraries like Europeana, with Europeana EDM

aligning seamlessly with CIDOC-CRM.

FRBR, developed by the International Federation of

Library Associations (IFLA), establishes the Work Expres-

sion Manifestation and Item (WEMI) Structure as its foun-

dation. This framework distinguishes between abstract

Work, its various Expressions, physical Manifestations em-

bodying those Expressions, and individual Items speciﬁc to

Manifestations.

In February 2023, FRBR released the draft model of

its lightweight WEMI conceptualization, LRM, maintain-

ing alignment with CIDOC-CRM.

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

104

plexity of the material and the recipient’s existing

knowledge (Sweller et al., 1998). As such, intrinsic

load cannot be altered by information organizational

design, but understanding its impact is crucial for pac-

ing and structuring information uptake.

Extraneous load, on the other hand, refers to the

cognitive load imposed by the way information is pre-

sented to recipients (Sweller et al., 1998; Whelan,

2007). It is not essential for information uptake and

can be reduced or managed through effective instruc-

tional design. Extraneous load can arise from disor-

ganized and inadequately presented information that

distracts or confuses recipients, requiring them to ex-

pend effort on processing irrelevant information. Re-

ducing extraneous load is crucial for liberating work-

ing memory capacity to handle intrinsic and germane

loads.

Germane load is the cognitive effort dedicated

to processing, constructing, and automating schemas

(Sweller et al., 1998). It represents the cognitive load

that contributes directly to learning by facilitating the

integration of new information into long-term mem-

ory. Effective organization and presentation of knowl-

edge aim to maximize germane load by using strate-

gies that support schema acquisition

The application of CLT has been signiﬁcant in in-

forming strategies to reduce extraneous load while

maximizing the potential to assimilate information

(Skulmowski and Xu, 2022). Research in multimedia

learning has highlighted several principles aimed at

minimizing extraneous load, thereby optimizing cog-

nitive capacity for processing intrinsic content. No-

table principles include the split-attention effect, ad-

vocating for the placement of related information in

proximity to each other, and guidelines against in-

cluding distracting or irrelevant details in information

presented to end-users to avoid the seductive details

effect. This effect can hinder recipients’ understand-

ing of core data by distracting them with irrelevant

information presented alongside the main material.

When applied to ontology modularization meth-

ods and techniques, these principles may outline the

strategy or approach that the modularization method

should follow. Moreover, they can provide insights

into which criteria should be primarily considered to

evaluate the results of the modularization approach

and, consequently, the approach itself. The focus

points thus become cohesion of the information pre-

sented and grouped in each module, consistency of

the information covered in each module, complete-

ness of each module, domain appropriateness, inte-

gration (i.e. ease of integrating modules back into

the original ontology while ensuring the degree of in-

dependence of each module), and the logical consis-

tency of the modules.

Translating such CLT principles into guidelines,

the following points can be derived for the develop-

ment approach of the modularization method. The

method should create thematically cohesive modules,

aiming to remove all superﬂuous information while

still maintaining the richness of the axiomatization

of the entities in the modules. As a result, cohe-

sion, which refers to the degree to which the ele-

ments within a module are related to each other, and

coupling, which refers to the degree of interdepen-

dence between modules, become primary criteria for

the evaluation of the modularization approach.

In line with these principles, OMNIMOD has

been designed to adhere to the guidelines derived

from CLT, as will be presented in the following sec-

tions. Its efﬁciency is evaluated primarily in terms of

coupling and cohesiveness, providing insights into its

performance and highlighting space for potential im-

provements.

4 OMNIMOD

To effectively apply and implement the principles of

CLT in the creation of ontological modules, it is cru-

cial to ﬁrst comprehend the foundational organiza-

tion of information in OWL (Web Ontology Lan-

guage) and RDF(S) (Resource Description Frame-

work Schema), or more broadly, the underlying phi-

losophy that guides the technical implementation of

these two languages

RDF and OWL semantics are built on the fun-

damental distinction between the concepts of uni-

versality (rdfs:resource and owl:Thing) and nul-

lity (owl:Nothing). The classes rdfs:resource and

owl:Thing serve as universal or ”upper level” classes,

signifying that all instantiable classes fall under them.

Conversely, the concept of the empty set is captured

in OWL with owl:Nothing, and in RDF, a similar con-

cept is implied though not explicitly named, high-

lighting the framework’s capability to represent the

absence of any class instances.

By adhering to this framework, semantic tech-

nologies inherently propose that all entities trace back

to a foundational origin situated in a root class, while

entities that cannot be instantiated are assigned to

an empty set. This dichotomy serves as the ba-

sis for organizing knowledge within these systems.

Subsequently, classes receive further characterization

through the properties (or relationships) they form

For RDFs and OWL documentation cf. respective-

ley: https://www.w3.org/TR/rdf-schema/ and https://www.

w3.org/TR/owl-guide/

OMNIMOD: Automating Ontology Modularization for Digital Library Data Using CIDOC-CRM as Use Case

105

with other groups of individuals. These properties not

only reﬁne but also augment the deﬁnitions of classes.

The nature of the relationships between individuals

dictates their classiﬁcation, determining whether they

belong to one class or another.

OWL class restrictions, along with RDFS domain

and range deﬁnitions, signiﬁcantly augment the ax-

iomatic description of classes. More speciﬁcally,

OWL Class Restrictions provide a powerful means to

deﬁne conditions that individuals must satisfy to be-

long to a particular class. These restrictions can spec-

ify necessary and sufﬁcient conditions for class mem-

bership based on the presence or absence of certain

properties and their values. On the other hand, RDFS

domain and RDFS range deﬁnitions indirectly inﬂu-

ence class membership by asserting that individuals

with certain properties belong to particular classes

or that the values of speciﬁc entities characterized

through the use of properties must ﬁt within deﬁned

categories.

Summarily, the above described organizational

logic follows a class-centric approach where sets and

groups of individuals are categorized into speciﬁc

classes based on the attributes and relationships they

exhibit. Subclass relationships, in particular, estab-

lish a hierarchical structure among classes, implying

that properties and attributes are inherited and shared

down the hierarchy from upper classes to their sub-

classes.

The method proposed in this study respect the

aforementioned class-centric approach to modularize

knowledge representation. This means that it breaks

down complex hierarchical structures into distinct,

manageable modules based on the organization and

deﬁnition of the classes. Each module not only in-

cludes its own set of properties (where the classes act

as either the domain or the range) but also encom-

passes the individuals (data records) and the relation-

ships these individuals instantiate. The formal deﬁ-

nition of the modularization method used to develop

OMNIMOD can be described as presented below, be-

ginning with the notation utilized and a description of

this notation, as detailed in Table 1.

The modularization method followed by OMNI-

MOD aims to break the ontology into distinct mod-

ules based on the direct subclasses of the root class R.

Each module is structured as follows:

• The root class R and its deﬁnitions.

• Each direct subclass C within Sub(R), including

their complete subclass hierarchy (Sub

∗

(C)).

• Deﬁnitions, restrictions, and properties associated

with R, C, and subclasses within Sub

∗

(C).

• All individuals that instantiate R, C, or any sub-

Table 1: Notation Overview.

Notation Description

O Ontology.

C The set of all classes within O.

P All properties in O.

I All individuals in O.

Sub(C) The set of direct subclasses of

any class C in O, used speciﬁ-

cally for the root class R.

Sub

∗

rect) under each direct subclass

C within Sub(R).

Def(C) The deﬁnition of class C, in-

cluding OWL restrictions and

class-speciﬁc properties.

Dom(P) The domain of property P.

Range(P) The range of property P.

Triples(I) RDF triples associated with in-

dividual I.

R The root class for modulariza-

tion purposes.

class within Sub

∗

(C), along with the RDF triples

involving these individuals.

• Properties where R, C, or any subclasses within

Sub

∗

• Properties where the domain or range is not spec-

iﬁed, and therefore potentially applicable to all

classes.

The modularization approach implemnted in OMNI-

MOD can be formally described as presented in Table

Table 2: Overview of the OMNIMOD Method’s Modular

Structure.

Category Modularization Method

Classes: C

= {R} ∪ {C} ∪Sub

∗

(C)

Properties: P

= {P ∈ P : ((Dom(P) ∈ C

∨

Range(P) ∈ C

) ∨ (Dom(P) =

0 ∨

Range(P) =

0))}

Instances: I

= {I ∈ I : InstanceOf(I, X), X ∈ C

}

Triples

= {triples involving I : I ∈ I

}

The modularization method delineated above is

technically implemented as a Python function, mak-

ing it accessible for anyone to try, reuse, and adapt as

needed. This function, which is fully detailed below,

requires two inputs: the ontology in question, and the

root class from which the user wants to initiate the

modularization process. The design of this function

provides a degree of adaptability for subsequent in-

depth analysis. Moreover, it outputs the modules in

a Turtle (.ttl) ﬁle format, thus ensuring broad com-

patibility and facilitating further utilization within the

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

106

semantic web framework.

import rdflib

from rdflib import Graph, URIRef, BNode

from rdflib.namespace import RDF, RDFS, OWL

def get_local_name(uri):

return uri.split(’/’)[-1].split(’#’)[-1]

def OMNIMOD(ontology_file, root_class_uri):

g = rdflib.Graph()

g.parse(ontology_file, format=rdflib.util.

guess_format(ontology_file))

# Ensure root_class_uri is a URIRef

if isinstance(root_class_uri, str):

root_class_uri = URIRef(root_class_uri)

# Find immediate subclasses of the root

class

immediate_subclasses = set(g.subjects(RDFS

.subClassOf, root_class_uri))

all_properties = set(g.subjects(RDF.type,

RDF.Property)) | set(g.subjects(RDF.

type, OWL.ObjectProperty)) | set(g.

subjects(RDF.type, OWL.

DatatypeProperty))

def recursively_add_triples(subject,

module_graph):

"""

Recursively add triples; specially

handle blank nodes to capture all

related triples.

"""

for s, p, o in g.triples((subject, None

, None)):

module_graph.add((s, p, o))

if isinstance(o, BNode):

recursively_add_triples(o,

module_graph)

def add_class_hierarchy_and_properties(

class_uri, module_graph):

# Handle class hierarchy

for s, p, o in g.triples((None, RDFS.

subClassOf, class_uri)):

module_graph.add((s, p, o))

add_class_hierarchy_and_properties(

s, module_graph)

# Include all direct triples and

recursively handled blank nodes

linked via complex properties

recursively_add_triples(class_uri,

module_graph)

# Handle properties where the class is

the domain or range, or no domain/

range specified

handled_properties = set()

for prop in set(g.subjects(RDFS.domain,

class_uri)) | set(g.subjects(RDFS.

range, class_uri)):

handled_properties.add(prop)

recursively_add_triples(prop,

module_graph)

# Include properties with no domain or

range in each module

for prop in all_properties -

handled_properties:

if not list(g.objects(prop, RDFS.

domain)) and not list(g.objects

(prop, RDFS.range)):

recursively_add_triples(prop,

module_graph)

# Include individuals that are

instances of the class

for individual in g.subjects(RDF.type,

class_uri):

for s, p, o in g.triples((

individual, None, None)):

module_graph.add((s, p, o))

for subclass in immediate_subclasses:

module_graph = Graph()

add_class_hierarchy_and_properties(

subclass, module_graph)

filename = f"{get_local_name(subclass)}

_module.ttl"

module_graph.serialize(destination=

filename, format=’turtle’)

print(f"Module created: {filename}")

The code displayed above consists of several key

parts. First, the necessary RDFLib modules are im-

ported to handle RDF data. Then, the utility function

get local name is provided to extract the local name

from a URI. The main function, OMNIMOD, parses

the ontology ﬁle and identiﬁes immediate subclasses

of the speciﬁed root class. It deﬁnes helper functions

to recursively add triples and handle class hierarchies

and properties, and creates separate modules for each

subclass by capturing all relevant triples and proper-

ties.

OMNIMOD can be used by providing the path

to the ontology that needs to be modularized and the

root class from which the modularization process may

start, as in the following example of usage

please note that OMNIMOD segments the ontology

OMNIMOD: Automating Ontology Modularization for Digital Library Data Using CIDOC-CRM as Use Case

107

# Example usage

OMNIMOD(’CIDOC_CRM_v7.1.3.rdf’, ’http://www.

cidoc-crm.org/cidoc-crm/E1_CRM_Entity’)

5 EVALUATION METRICS

OMNIMOD has undergone testing and validation by

applying it to CIDOC-CRM ontology

with and with-

out instances (i.e., data records). To evaluate OMNI-

MOD’s efﬁcacy, and therefore the effectiveness of the

modularization method, measures for cohesion and

coupling have been derived.

5.1 Cohesion and Coupling

Cohesion and coupling metrics evaluate the strength

of relationships within and between modules(Khan,

2016). Cohesion refers to the relatedness of ele-

ments within a single module(Garc

ıa et al., 2010)(Oh

et al., 2011). It is inferred from the intra-module re-

lationship density. Higher density indicates higher

cohesion, suggesting well-connected classes within

the module. Coupling measures the interdependen-

cies between different modules(Oh and Ahn, 2009).

Lower coupling values indicate fewer dependencies,

which is desirable for maintainability and scalability.

High cohesion and low coupling, as measured by

average intra-module and inter-module relationship

densities, align with CLT by reducing extraneous cog-

nitive load and minimizing the split-attention effect.

5.1.1 Cohesion

Cohesion in the context of ontology modularization

measures how well-connected the classes within a

module are. The density formula (1) provides a way

to quantify this by considering the number of logical

axioms and the number of instantiated classes within

the module. Logical axioms represent the relation-

ships between classes, so a higher number of axioms

generally indicates better connectivity.

Cohesion was calculated using the following met-

ric for density in modules that had more than two in-

stantiated classes:

into modules starting from a speciﬁed root class down

through its branches. If users wish to start the modular-

ization from all upper root classes, they should begin by in-

dicating OWL:Thing or rdfs:Resource as the starting point

for modularization. If the root classes are not explicitly in-

stantiated as subclasses of these, users need to add such def-

initions to modularize from the upper part of the ontology.

The version used is: v.7.1.3. It can be

downloaded here: https://www.cidoc-crm.org/

versions-of-the-cidoc-crm

Density =

Number of Logical Axioms

Number of Classes × (Number of Classes − 1)

(1)

In formula (1), the terms are deﬁned as follows:

• Number of Logical Axioms: This is the to-

tal count of logical relationships between classes

(such as subclass, equivalent class, property re-

strictions, object properties) deﬁned within the

module.

• Number of Classes: This is the total count of in-

stantiated distinct classes present within the mod-

ule. Note that this does not include distinct classes

that are not instantiated and are derived from the

domain and range external to the module.

• (Number of Classes - 1): This accounts for the

potential relationships between the instantiated

classes. Subtracting 1 ensures the density mea-

sure considers the proportion of actual relation-

ships relative to the maximum possible relation-

ships between different classes, normalizing the

value based on the size of the module

For modules including fewer than 2 instantiated

classes, the strength of relation for each entity is cal-

culated based on the farness centrality measure from

graph theory, as proposed by(Freeman, 1978) and

summarized in (Khan, 2016)

The results have been evaluated against the bench-

mark displayed in table 3:

Table 3: Benchmark to assess cohesion levels.

Cohesion Level Cohesion Range

High Cohesion ≥ 0.5

Moderate Cohesion 0.2 ≤

Normalized Cohesion <

0.5

Low Cohesion < 0.2

5.1.2 Coupling

Coupling was derived by calculating the ratio between

A) the number of external classes with which the in-

stantiated classes within the module established one

or more relationships, and B) the number of instanti-

ated classes within the module. The results were then

normalized to provide an indication measure.

The following benchmark was used to assess cou-

pling:

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

108

Table 4: Benchmarks for Coupling Levels.

Coupling Level Coupling Range

High Coupling ≥ 0.5

Moderate Coupling 0.2 ≤

Average coupling value <

0.5

Low Coupling < 0.2

High coupling indicates strong dependencies be-

tween modules, which can be problematic as changes

in one module might impact others. Conversely, low

coupling suggests that the modules are relatively in-

dependent, which is desirable for maintainability and

scalability.

6 RESULTS

OMNIMOD and its effectiveness have been tested

on CIDOC-CRM, starting the modularization process

from the ”Entity” class. The resulting modules were

named after the respective upper subclass under the

CIDOC-CRM Entity from which they were derived.

The modules created using OMNIMOD are as fol-

lows, each stored in a separate TTL ﬁle:

• E53 Place Module: Contains geographical data

related to various locations.

• E54 Dimension Module: Includes measurements

data.

• E92 Spacetime Volume Module: Integrates spatial

and temporal data.

• E77 Persistent Item Module: Assists in managing

artifact catalogues.

• E52 Time-Span Module: Provides detailed

chronological information.

• E2 Temporal Entity Module: Focuses on the tim-

ing aspects of events.

The modules were evaluated based on the metrics for

computing cohesion and coupling presented in the

previous section. The average measure for cohesion

across modules was 0.6, while that for coupling was

0.5. When tests were conducted with instance data,

the data were correctly mapped within the module of

the classes in which they were instantiated. Based

on a comparative analysis across the modules and the

core module, no information regarding properties and

classes under the branches of the provided root class

was lost. Logical consistency was maintained, as veri-

ﬁed by running the Hermit 1.4.3.456 reasoner on each

generated module, which revealed no contradictions.

The inference was limited to the instantiated classes

and their relative relations, without extending to the

classes that had no instantiation but were still present

in the module due to being part of object properties’

axiomatic deﬁnitions derived from the use of domain

and range. Their inclusion allowed for integration of

the module back into the core ontology, thereby meet-

ing the requirement for evaluating possible reintegra-

tion into the main ontology as well.

7 DISCUSSION

OMNIMOD, and consequently its theoretical modu-

larization approach, has scored highly in both cohe-

sion and coupling. While the high cohesion score in-

dicates that the richness of the axiomatization within

each module is well-maintained, the high coupling

value reveals potential scalability and maintainability

issues. Speciﬁcally, if modiﬁcations are made within

a module rather than in the core ontology, the high

coupling may lead to difﬁculties in maintaining the

integrity and consistency of the entire ontology.

These insights also prompt a point of reﬂection

and help in better scoping the modularization ap-

proach. In fact, OMNIMOD can be a valuable support

tool for users to understand, inspect, and analyze large

ontological frameworks, as indicated by the cohesion

score. Each generated module is self-contained and

focuses on a speciﬁc thematic area, functioning as

a standalone database. This possibly makes it eas-

ier to conduct targeted academic and professional ex-

ploration. Each module can show the main hierarchy

and relationships of the branches of the root class pro-

vided as input to start the modularization, and also al-

low for the inclusion of instance data and their triples

in the core module where their class is instantiated.

Moreover, by instantiating and providing axiom-

atization only for the classes that are the focal point

of the module, while presenting the classes at the

level of range and domain simply as such without

further instantiation, the CLT-described split-attention

effect and seductive details effect are potentially min-

imized. This approach aims to enable users to fo-

cus directly on data pertinent to their ﬁeld without

distractions. The results can span from improved

user engagement through organized and easily navi-

gable modules, to enhancing efﬁciency and accuracy

in data retrieval and analysis. Ultimately, OMNI-

MOD may represent a valuable support for profes-

sionals seeking familiarity with ontologies to struc-

ture their data more effectively. It can certainly assist

professionals in evaluating an ontology’s expressiv-

ity coverage, enabling them to align their conceptu-

alizations with established standards and potentially

OMNIMOD: Automating Ontology Modularization for Digital Library Data Using CIDOC-CRM as Use Case

109

extend these knowledge representations.

However, as previously addressed, from a techni-

cal perspective, OMNIMOD segments ontologies and

instance data by creating duplicates. An instance in a

triple within one module might also appear in a triple

in another module. This principle applies to relation-

ships and classes as well. Although OMNIMOD’s du-

plication strategy may streamline exploration, it may

also create issues if changes are made in one module

without updating another module. Therefore, when

changes are made, they should take place at the level

of the integrated ontology rather than at the module

level to avoid intensifying further work and creating

possible problems.

The limitation described above also highlights key

areas for further development. If readers of this paper

wish to use and expand the core function provided in

section 4, they should consider enabling OMNIMOD

for users who want to directly work on the modules

and have automated updates applied to all other mod-

ules. Speciﬁcally, a strategy for the automated up-

dating of all modules when a change is made in one

of them needs to be implemented. This enhancement

would ensure consistency and integrity across the en-

tire ontology, facilitating easier maintenance.

ACKNOWLEDGEMENTS

This research activity was conducted as part of the

Norm Engineering Program at TNO, funded by the

Dutch Ministry of the Interior and Kingdom Rela-

tions.

REFERENCES

Breteler, J., van Gessel, T., Biagioni, G., and van Doesburg,

R. (2023). The ﬂint ontology: An actor-based model

of legal relations. In SEmantics2023:Proceedings.

Commission, E. (2022). Moving towards a euro-

pean data space: New eu law for data-sharing.

https://data.europa.eu/en/news-events/news/moving-

towards-european-data-space-new-eu-law-data-

sharing, last accessed 2024/04/18.

Corcho, O. and Simper, E. (2022). The role of

data.europa.eu in the context of european com-

mon data spaces. https://data.europa.eu/sites/default/

ﬁles/report/data.europa.eu theRoleofdata.europa.

euinthecontextofEuropeancommondataspaces.pdf,

last accessed 2024/04/18.

d’Aquin, M., Schlicht, A., Stuckenschmidt, H., and Sabou,

M. (2009). Criteria and evaluation for ontology mod-

ularization techniques. In Modular Ontologies: Con-

cepts, Theories and Techniques for Knowledge Mod-

ularization, pages 67–89. Springer Berlin Heidelberg,

Berlin, Heidelberg.

Freeman, L. C. (1978). Centrality in social networks con-

ceptual clariﬁcation. Social Networks, 1(3):215–239.

Garc

ıa, J., Pe

nalvo, F. J. G., and Ther

on, R. (2010). A sur-

vey on ontology metrics. In Lytras, M. D., de Pablos,

P. O., Ziderman, A., Roulstone, A., Maurer, H. A.,

and Imber, J. B., editors, Third World Summit on the

Knowledge Society, WSKS’10, volume 111 of Com-

munications in Computer and Information Science,

pages 22–27, Corfu, Greece. Springer. September 22-

24.

Gatens, W., Konev, B., and Wolter, F. (2014). Lower and up-

per approximations for depleting modules of descrip-

tion logic ontologies. Frontiers in Artiﬁcial Intelli-

gence and Applications, 263:345–350.

Ghafourian, S., Rezaeian, A., and Naghibzadeh, M. (2013).

Graph-based partitioning of ontology with semantic

similarity. In Proceedings of the 3rd International

Conference on Computer and Knowledge Engineer-

ing, ICCKE 2013.

Grau, B. C., Horrocks, I., Kazakov, Y., and Sattler, U.

(2009). Extracting modules from ontologies: A logic-

based approach. In Modular Ontologies: Concepts,

Theories and Techniques for Knowledge Modulariza-

tion, pages 159–186. Springer, Berlin, Heidelberg.

Kachroudi, M., Zghal, S., and Yahia, S. B. (2013). On-

topart: At the cross-roads of ontology partitioning

and scalable ontology alignment systems. Interna-

tional Journal of Metadata, Semantics and Ontolo-

gies, 8:215–225.

Khan, Z. C. (2016). Evaluation metrics in ontology mod-

ules. Description Logics.

Konev, B., Walther, D., and Wolter, F. (2009). Forgetting

and uniform interpolation in extensions of the descrip-

tion logic el. In CEUR Workshop Proceedings, vol-

ume 477.

Kontchakov, R., Wolter, F., and Zakharyaschev, M. (2010).

Logic-based ontology comparison and module extrac-

tion, with an application to dl-lite. Artif. Intell.,

174(15):1093–1141.

Leclair, A., Kh

edri, R., and Marinache, A. (2019). Toward

measuring knowledge loss due to ontology modular-

ization. In Scitepress.

LeClair, A., Marinache, A., Ghalayini, H. E., MacCaull, W.,

and Khedri, R. (2023). A review on ontology modular-

ization techniques - a multi-dimensional perspective.

IEEE Transactions on Knowledge and Data Engineer-

ing, 35(5):4376–4394.

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

to extract views. In Modular Ontologies: Concepts,

Theories and Techniques for Knowledge Modulariza-

tion, pages 245–260. Springer, Berlin, Heidelberg.

Oh, S. and Ahn, J. (2009). Ontology module metrics. In

International Conference on e-Business Engineering

(ICEBE’09), pages 11–18, Macau, China. IEEE Com-

puter Society.

Oh, S., Yeom, H. Y., and Ahn, J. (2011). Cohesion and

coupling metrics for ontology modules. Information

Technology and Management, 12(2):81–96.

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

110

Parent, C. and Spaccapietra, S. (2009). An overview of

modularity. In Modular Ontologies: Concepts, The-

ories and Techniques for Knowledge Modularization,

pages 5–23. Springer Berlin Heidelberg, Berlin, Hei-

delberg.

Sarkar, S. and Dong, A. (2011). Characterizing modularity,

hierarchy and module interfacing in complex design

systems. In Proceedings of the ASME Design Engi-

neering Technical Conference, volume 9.

Skulmowski, A. and Xu, K. M. (2022). Understanding cog-

nitive load in digital and online learning: A new per-

spective on extraneous cognitive load. Educational

Psychology Review, 34:171–196.

Sweller, J., van Merrienboer, J. J. G., and Paas, F. (2019).

Cognitive architecture and instructional design: 20

years later. Educational Psychology Review, 31:261–

292.

Sweller, J., van Merrienboer, J. J. G., and Paas, F. G. W. C.

(1998). Cognitive architecture and instructional de-

sign. Educational Psychology Review, 10:251–296.

Vescovo, C. D., Gessler, D. D., Klinov, P., Parsia, B., Sat-

tler, U., Schneider, T., and Winget, A. (2011a). De-

composition and modular structure of bioportal on-

tologies. In The Semantic Web-ISWC 2011, pages

130–145, United States. Springer Nature.

Vescovo, C. D., Parsia, B., Sattler, U., and Schneider,

T. (2011b). The modular structure of an ontology:

Atomic decomposition and module count. Frontiers

in Artiﬁcial Intelligence and Applications, 230.

Whelan, R. R. (2007). Neuroimaging of cognitive load in

instructional multimedia. Educational Research Re-

view, 2(1):1–12.

OMNIMOD: Automating Ontology Modularization for Digital Library Data Using CIDOC-CRM as Use Case

111