Semantic Representation of Physics Research Data

Aysegul Say

1 a

, Said Fathalla

1,2 b

, Sahar Vahdati

1,3 c

Jens Lehmann

1,4 d

and S

oren Auer

5,6 e

Smart Data Analytics (SDA), University of Bonn, Germany

Faculty of Science, University of Alexandria, Egypt

Department of Computer Science, University of Oxford, U.K.

Fraunhofer IAIS, Dresden, Germany

TIB Leibniz Information Center for Science and Technology, Hannover, Germany

L3S Research Center, University of Hannover, Germany

Keywords:

Semantic Web, Domain Ontology, Ontology Engineering, Semantic Publishing, Scholarly Communication,

Physics.

Abstract:

Improvements in web technologies and artiﬁcial intelligence enable novel, more data-driven research practices

for scientists. However, scientiﬁc knowledge generated from data-intensive research practices is disseminated

with unstructured formats, thus hindering the scholarly communication in various respects. The traditional

document-based representation of scholarly information hampers the reusability of research contributions. To

address this concern, we developed the Physics Ontology (PhySci) to represent physics-related scholarly data

in a machine-interpretable format. PhySci facilitates knowledge exploration, comparison, and organization

of such data by representing it as knowledge graphs. It establishes a unique conceptualization to increase

the visibility and accessibility to the digital content of physics publications. We present the iterative design

principles by outlining a methodology for its development and applying three different evaluation approaches:

data-driven and criteria-based evaluation, as well as ontology testing.

1 INTRODUCTION

The advent of the Web has led researchers to a new

era where research paradigms (empirical, theoreti-

cal, and computational) have merged with data-driven

science (Hey et al., 2009). Today most of the sci-

entiﬁc disciplines, especially physics, require data-

driven technologies to integrate large-scale data that

is produced by satellites, telescopes, and sensor net-

works. However, the application of data-intensive

practices has produced a vast amount of unstructured

data on the Web. Even though most of the scholarly

output is meanwhile digitally available, the lack of

coverage of digital content for each scientiﬁc commu-

nity is a pervasive barrier to productive research in the

https://orcid.org/0000-0002-8803-7355

https://orcid.org/0000-0002-2818-5890

https://orcid.org/0000-0002-7171-169X

https://orcid.org/0000-0001-9108-4278

https://orcid.org/0000-0002-0698-2864

physics domain. Since science is multidisciplinary in

nature, researchers need resources that cover various

subjects related to their research interest; however,

with the current search engines, it is difﬁcult to ﬁnd

links with cross-domain publications. As a result, it

is inconvenient to discover, reuse, and process pub-

lished articles with tools or interfaces for researchers.

The traditional system of scholarly communica-

tion is improving with recent developments related

to the use of semantic and AI technologies. Se-

mantic technologies and knowledge graphs offer new

ways for discovering and dissemination of scholarly

content, which leads to better collaboration. In this

context, ontologies support knowledge extraction and

modeling as a speciﬁcation of a conceptualization.

Also, they help to resolve the difﬁculty of handling

the overﬂow of heterogeneous data by organizing and

interlinking the data in a meaningful way.

The objective of this study is to support scholarly

communication and ﬁll the research gap by provid-

ing an ontology for organizing physics-related sci-

Say, A., Fathalla, S., Vahdati, S., Lehmann, J. and Auer, S.

Semantic Representation of Physics Research Data.

DOI: 10.5220/0010111000640075

In Proceedings of the 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2020) - Volume 2: KEOD, pages 64-75

ISBN: 978-989-758-474-9

Figure 1: Knowledge capturing for physics publications. (1) Candidate terms are detected from unstructured text (i.e., physics

publications). (2) Networks of scholarly entities and science discourse elements are created. (3) All concepts are structured

and combined in a knowledge graph. (4) Discover and analyze complex or hidden relations. Enable collaboration between

different scientiﬁc communities.

entiﬁc contributions. In this study, we target the

following research question: How can we facilitate

access to the physics research data in a machine-

understandable way? Therefore, we applied semantic

technologies to represent the outputs of physics re-

search. With the Physics ontology (PhySci), we aim

to support the transformation of scholarly communi-

cation in physics and provide more effective solutions

for scholarly research. To enable a rich representa-

tion of scientiﬁc data, the FAIR principles (Wilkinson

et al., 2016) are applied for rendering data and ser-

vices. The principles emphasize the capacity of com-

putational systems to ﬁnd, access, interoperate, and

reuse data efﬁciently. A knowledge graph scheme is

designed for physics publications to determine how

PhySci can exploit the formal structure of a publica-

tion and the details of research that is saved in the

author’s mind(see Figure 1). This model has three

main contributions. First, the densely interconnected

content of scientiﬁc publications promotes accessing

more convenient scientiﬁc collections and proposals.

Second, improved reusability of materials enables re-

searchers to advance the production of new knowl-

edge. Third, performing semantic queries on orga-

nized knowledge can facilitate the interpretation of

the physics content. To apply ontology-based repre-

sentation, existing ontological resources are aligned

with PhySci. In fact, the (PHYSCI) ontology is one of

the ontologies of the Science Knowledge Graph On-

tologies (SKGO) (Fathalla et al., 2020) suite. Various

RDF serializations of the ontology can be found on

SKGO’s GitHub repository

. Furthermore, human-

readable documentation of PhySci is available via its

Persistent Identiﬁers (https://w3id.org/skgo/physci#).

The preﬁxes are registered in preﬁx.cc

, a name-space

lookup service for RDF developers, under the open

CC-BY 3.0 license.

The article is organized as follows: In section 2,

we present the fundamental approaches that are ap-

plied in the development of the ontology. Section 3

presents speciﬁc design patterns and the structure of

concepts in PhySci. The evaluation, given in sec-

tion 4, discusses a set of assessment methods for

PhySci. Section 5 introduces state-of-art practices.

Section 6 gives a summary of our approach and an

outlook of future work.

2 METHODOLOGY

In our approach, we applied ontological engineer-

ing practices to systematize the ontology develop-

ment. Moreover, the application of the modeling

methodologies enables to transform the requirements

into a formal language that is designed, evaluated,

and documented by the ontology. We follow the

Ontology development 101 (Noy et al., 2001) and

the Systematic Approach for Building Ontologies

https://github.com/saidfathalla/

Science-knowledge-graph-ontologies

https://preﬁx.cc/

Semantic Representation of Physics Research Data

(SABiO) (de Almeida Falbo, 2014) for the develop-

ment of PhySci Ontology. Our modeling methodol-

ogy is composed of four main phases with support

activities and outlined as follows.

Identiﬁcation of requirements: This phase starts with

the selection of the focus area of the ontology, prepar-

ing and collecting requirements, and determining in-

put sources and raw data. Requirements are ad-

dressed to cover the user issues and to reach a high-

quality model. Those requirements are listed as fol-

lows. Accuracy: All axioms represented in the on-

tology should be aligned with the domain knowledge

of stakeholders. Coherence: The ontology must be

uniﬁed with the terms related to the physics domain.

Consistency: The ontology must be consistent with-

out having any contradiction about input data. Exten-

sibility: The ontology should be extended by merging

new deﬁnitions and information. Reliability: The on-

tology performance should be reliable and be able to

process complex queries. Data Timeliness: The on-

tology should offer reliable and accessible data that

are related to papers published within a speciﬁc pe-

riod. Reusability: The ontology should interoperate

with other ontologies.

Domain Conceptualization and Formalization: Do-

main concepts that will be integrated into the ontol-

ogy are determined. Conceptual modeling activity

starts with selecting ontological and non-ontological

knowledge resources.

Resource: Scientiﬁc publications are used as the pri-

mary non-ontological data source for capturing the

domain concepts while creating the ontology. Those

publications are selected from IOP

and APS

science

journals with their whole context (e.g., abstract and

introduction).

Topical Coverage: Physics, as a scientiﬁc disci-

pline, is divided into different sub-disciplines (Feyn-

man et al., 1965). We deﬁned our corpus with mostly

particle physics, and high energy physics since these

topics involve popular investigations and have many

relations to other sub-disciplines of physics. This ac-

tivity produces a complete dictionary that includes

classes, instances, and properties with dictionary ta-

bles. Using tables of classes and properties helps to

gather all the useful and potentially usable domain

concepts, their meanings, relations, labels, and URI’s.

Informal axioms extracted from resources are trans-

formed into formal axioms. The formalization phase

proposes to have clarity and correctness within the on-

tology.

Design and Development: Ontology is deﬁned in a

https://iopscience.iop.org/

https://journals.aps.org/

formal language. Web Ontology Language (OWL 2)

is chosen to formalize the PhySci ontology. OWL

provides different elements (e.g., classes, annotations,

properties, and instances) that can be used for formal-

ization and development tasks (McGuinness et al.,

2004). We used Graffoo Editor

to design the formal

ontology and to visualize the structure of the ontol-

ogy. Prot

e ontology editor (Musen et al., 2015) of-

fers strong functionality such as modiﬁcation, query-

ing, and reasoning. Thus, Prot

e is used for the de-

velopment of the PhySci ontology.

Ontology Testing: This phase executes all the require-

ments that are designed for the behavioral character-

istics of the ontology. A set of test cases are formed

from competency questions to ensure that the on-

tology satisﬁes the expected behavior regarding the

competency questions. In addition to the develop-

ment process listed above, SABiO (de Almeida Falbo,

2014) considers some supporting processes: docu-

mentation, reuse (section 3.1), knowledge acquisition,

and evaluation.

Knowledge Acquisition: The process of knowledge

acquisition starts by extracting terms and their syn-

onyms from the underlying text. For this process,

scientiﬁc publications about physics are set as a

non-ontological resource to form a corpus

. This

corpus comprises 125.592 words and 5.083 sen-

tences in total. Each section of the papers, rhetor-

ical terms (e.g., conclusion, abstract), and scien-

tiﬁc discourse elements (e.g., equations, theories)

are identiﬁed in this task. We applied statistical

techniques; TF-IDF (Term Frequency/Inverse Doc-

ument Frequency) (Ramos et al., 2003) weighting

scheme in combination with Latent Semantic Anal-

ysis (LSA) (Landauer et al., 1998) techniques for the

knowledge acquisition process to capture latent con-

cepts and discover a coherent knowledge base that

devises an effective knowledge representation. Key

terms that have the highest scores based on the TF-

IDF scores are utilized to create a semantic space

where terms are associated with one another. There-

fore, most relevant terms are investigated using the

LSA method.

3 PhySci ONTOLOGY

The purpose of developing the PhySci Ontology is

to increase the value of the physics research data

https://www.w3.org/TR/owl2-overview/

http://www.essepuntato.it/graffoo

https://github.com/aysegulsay/PhySci/blob/master/

Datasets

KEOD 2020 - 12th International Conference on Knowledge Engineering and Ontology Development

Figure 2: Linked Scholarly Entities. The interpretation of main classes, object properties and data properties in PhySci

ontology.

by creating an ontology in regards to FAIR princi-

ples (Wilkinson et al., 2016). The PhySci ontology is

further reﬁned by reusing other vocabularies and inte-

grating extracted concepts that are determined from

knowledge acquisition. Moreover, we demonstrate

that PhySci ontology maintains the structure that is

represented in Figure 1 by providing classes and re-

lations to link those concepts. Figure 2 bestows the

central part of the ontology; it describes the related

entities with their object properties.

The development of an ontology requires an anal-

ysis of the types of concepts and relations. We applied

statistical analysis techniques to the textual data to

achieve an effective knowledge representation. Statis-

tical techniques, which are deﬁned in section 2, have

been used to extract candidate terms and identify spe-

ciﬁc sentence patterns from the corpus to conform the

triples in the ontology.

3.1 Reuse of Existing Resources

The PhySci ontology imports seven ontologies and

provides 110 OWL classes and 77 object properties.

We aligned similar and new concepts for the physics

domain from existing ontologies to achieve interoper-

ability with other systems and ontologies. This tech-

nique is mostly used for constructing domain-speciﬁc

ontologies in ontology engineering practices (Corcho

et al., 2007). Indeed, it helps to provide better cover-

age of the domain. While adapting new ontologies to

PhySci, we have followed the ontological levels (up-

per, middle, lower) to determine the semantic inter-

operability of PhySci. We begin deﬁning rhetorical

terms that establish the general skeleton of the publi-

cation. npg:Thing, npg:Publication, npg:Issue,

npg:Person, npg:Journal, and npg:Agent are

selected from Nature Publishing Group ontology

(NPG) (Hammond and Pasin, 2015). From Dublin

Core (Weibel et al., 1998), we selected entities to de-

scribe annotations of classes and relations between in-

stances such as terms:creator, terms:publisher

and data properties such as terms:Abstract and

terms:Date Modified. Semantic Web for Earth

and Environmental Terminology (SWEET) (Raskin

and Pan, 2003) is developed for the Earth sys-

tem science domain. Terms relevant to physical

models and components of evidence are aligned

from SWEET ontology, such as sol:Simulation,

mod:ScientificModel, and phen:Phenomena.

Extensible Observation Ontology(OBOE) (Madin

et al., 2007) arranges semantic subtleties of com-

plex ecological data. To deﬁne physical quantities

that are used in equations, we reused entities such

as oboe-core:Unit, oboe-core:Measurement, and

oboe-core:Force from OBOE. Semantic Sensor

Network (SSN) (Compton et al., 2012) ontology

represents sensors and their observations with re-

lated procedures, samples, and actuators. SOSA

(Sensor, Observation, Sample, and Actuator) is

another module of SSN. We mostly reused con-

cepts such as sosa:Observation and about ob-

servations from the SOSA module to deﬁne sci-

entiﬁc instruments in research. The Ontology

of Astronomical Object Types (IVOA) (Cambr

esy

et al., 2010) introduces astronomy and forma-

tion terms. To deﬁne characteristics of a for-

mation, ivoao:horizon, physics:collision, and

physics:spectrum classes are aligned from this on-

tology.

3.2 Representing Metadata in PhySci

In this section, we describe classes, instances, and

properties that are implicitly developed in PhySci

Semantic Representation of Physics Research Data

Figure 3: Scientiﬁc Model. Scientists try to draw their scientiﬁc knowledge by using scientiﬁc models to understand and

deﬁne features of speciﬁc patterns that occur in the universe. Thus, related classes and object properties to the scientiﬁc

model have been deﬁned in PhySci ontology with its instances for the publication (Medeiros et al., 2018).

Figure 4: Triple Patterns of instances and relations deﬁned in PhySci for scientiﬁc paper (Medeiros et al., 2018).

ontology. Each concept has the following features: a

URI, a preferred synonymous label, and a deﬁnition.

Figure 3 depicts the Scientific Model class and

its related classes, instances, and relations described

in PhySci. All instances have been directly extracted

from research papers. Figure 4 shows relations be-

tween instances for a scientiﬁc publication (Medeiros

et al., 2018). Triple patterns are created by con-

structing domain and range constraints for each

object property such as physci:ResearchWork

physci:considerCase physci:Case. To spec-

ify different use cases, relations are asserted for

each characteristic of properties (e.g., transitive,

asymmetric). For example, physci:isformedfrom

and physci:yieldEquation are set as reﬂexive

relations. Other properties are deﬁned as func-

tional relations and inverse functional such as

physci:relyOnMeasurement is functional while

physci:hasSource, and physci:hasObservation

are deﬁned as inverse functional.

Object properties are created concerning clusters of

similar instances to other instances. For example,

terms:Publication class is deﬁned for the physics

articles and it is set as a domain of the object property

physci:addressesResearchWork. The range of

this property is Research Work class. Another

example is the instance physci:CombinedField

Electricity of physci:Solution class connected

to instance physci:GeneralTheoryofrelativity

of class ivaoa:theory via the object property

physci:usesTheory. Instances of observation

class can be related to the observatory, obser-

vational data, observer, duration, and measure-

ment classes. For example, the observation class

can be related to observatory (Observation v

∃hasObservatory.Observatory) and forma-

tion (Observer u ∃detectFormation.Formation).

Observation class has relations with other classes

(Observation v ∃hasObservationalData.Observatio-

nalData). The instance physci:EHTVLBICampaign

Observation belonging to the sosa:Observation

class is connected to the instance physci:EHTData

KEOD 2020 - 12th International Conference on Knowledge Engineering and Ontology Development

Figure 5: Terms that are captured from the publication (Abbott et al., 2016) and their matched classes in PhySci.

of the class physci:ObservationalData by the ob-

ject property physci:hasObservationalData. We

speciﬁed data properties for the Publication class

to deﬁne the publication’s title and abstract. Fur-

thermore, the data property npg:publicationYear

can be used to associate a published year with a

publication. Physical quantities of the equations and

scientiﬁc properties can be deﬁned by using the data

properties such as physci:has Mass, physci:has

Condition, physci:hasParameter, physci:has

Velocity, physci:hasState, physci:hasEnergy,

and physci:has Temperature in PhySci to distin-

guish the equations.

Example candidate terms captured from scientiﬁc

publication (Abbott et al., 2016) and their deﬁned

classes and instances in PhySci can be seen in Fig-

ure 5. Additionally, extracted instances with related

class names are listed in the tables Table 1 and Ta-

ble 2.

3.3 Reasoning

The Semantic Web Rule Language (SWRL) (Hor-

rocks et al., 2004) establishes expressive representa-

tion formalisms and helps to reveal new inference in

an ontology. Therefore, SWRL rules are deﬁned for

PhySci by executing Drools reasoner (Proctor, 2011)

to infer alternative linking triples, to discover incon-

sistencies, and to improve the expressivity. The rule-

set of PhySci comprises the following SWRL rules.

From Equation 1, we can infer that if two publica-

tions relate to each other, then one of them might cite

the other. In Equation 2, the research work demon-

strates a scientiﬁc model to explain a solution by ap-

plying speciﬁc methods, and thus this solution might

include the scientiﬁc method. Equation 3 suggests

that research work reveals a formation, and it explains

a case related to the formation. For instance, research

work that investigates black holes might explain some

cases, such as the collision of neutrinos and massive

star collapsing to describe the occurrence of forma-

tions.

Publication(?x) ∧ Publication(?y)∧

relation(?x, ?y) ∧ addressesResearchWork(?y, ?z)

→ addressesResearchWork(?x, ?z) (1)

Scienti ficModel(?x) ∧ provideSolution(?x, ?y)∧

hasScienti ficMethod(?y, ?z) →

useScienti ficMethod(?x, ?z) (2)

ResearchWork(?x) ∧ Formation(?y)∧

argueFormation(?x, ?y) ∧ f ormedBy(?y, ?z)

→ ConsiderCase(?x, ?z) (3)

Semantic Representation of Physics Research Data

Table 1: Classes and related data properties in PhySci with captured terms from the research paper (Abbott et al., 2016).

Class Data Property Literal Value Data Type

Publication title GW150914:Implications for the

Stochastic Gravitational-Wave

Background from Binary Black

Holes

rdfs:Literal

Publication publicationDate 2016-03-31 xsd:date

Publication doi DOI:10.1103 rdfs:Literal

Publication creator B.P. Abbott et. al rdfs:Literal

Journal issue PRL 116, 131102(2016) rdfs:Literal

Observation ObservationTime 2015-10-14 xsd:dateTime

Formation hasScenario unresovable events combine to cre-

ate stochastic background

rdfs:Literal

Table 2: Classes and related instances in PhySci for the research paper (Abbott et al., 2016).

Class Instance

Publisher American Physical Society

Creator B.P. Abbott et al.

ResearchWork Implications for Gravitational Wave

Phenomena Gravitational waves

Observation LIGO detection of gravitational waves

Observatory The Laser Interferometer Gravitational Wave Observatory (LIGO)

Observer Hanford and Livingston detector

Formation Binary Black Holes

Wave GW150914

Spectrum Energy density spectrum

4 EVALUATION

The evaluation is a fundamental task for ontol-

ogy engineers to verify and validate the quality

of the ontology. According to SABIO methodol-

ogy (de Almeida Falbo, 2014), the evaluation pro-

cess has two main activities; (1) to check the ontol-

ogy requirements are being met with speciﬁc crite-

ria and (2) to ensure that the veriﬁed requirements

are compatible with the intended use of the ontology.

Therefore, we assessed the content by applying data-

driven (Rospocher et al., 2012) and criteria-based ap-

proaches (Gangemi et al., 2005). The ontology test-

ing technique is practiced by executing a test case for

each competency question to verify the requirements.

These approaches have been applied to assess in what

dimensions PhySci can bring value to the scholarly

community and how high can the impact be on the se-

mantic publishing.

(1) Ontology Content Evaluation: Quality crite-

ria are deﬁned to assess if the content of ontol-

ogy contains any anomalies or redundant informa-

tion (Gangemi et al., 2005) (Lovrencic and Cubrilo,

2008). The main goal of this step is to resolve if the

ontology deﬁnes concepts accurately, does not deﬁne,

or even deﬁnes inaccurately (G

omez-P

erez, 2001). A

set of criteria (i.e., consistency, completeness, con-

ciseness, expandability, and sensitiveness) are applied

through expert evaluation to assess the quality of the

ontology, the rate of its performance, and the deﬁni-

tion of the concepts. Each quality metric should con-

form to associated questions to check if the ontology

satisﬁes conditions or not. Table 3 shows the corre-

spondence between the metric, questions, and results.

(2) Ontology Testing: Competency questions are

prepared to determine the behavioral characteristics

of the ontology that relate to the knowledge repre-

sented in PhySci ontology. They are used to ensure

that the ontology implementation compatible with the

scope of PhySci. We created these questions from the

content of the text corpus. To evaluate the complete-

ness, instantiation queries that represent the compe-

tency questions are prepared. Therefore, CQs are

transformed into SPARQL queries that can be exe-

cutable within the framework. Table 4 presents a sam-

ple of 10 competency questions out of 20. CQ1 is ex-

pressed with SPARQL query in Listing 1 to ﬁnd pub-

lications that contain theories for a speciﬁc problem

KEOD 2020 - 12th International Conference on Knowledge Engineering and Ontology Development

Table 3: Quality Criteria for ontology evaluation.

Criteria Questions Results

Consistency Does the documentation of ontology

meet the speciﬁcation? Is there any en-

coding bias related to the transforma-

tion from the knowledge level to the en-

coding? Is the representation be made

genuinely for the beneﬁt of implemen-

tation?

Yes, ontology is consistent since it

does not include any contradictory

conclusions. Reasoner shows no er-

ror.

Completeness Is the domain of interest properly cov-

ered? Can the ontology answer all the

competency questions? Does the on-

tology include all related concepts to

the domain and their lexical represen-

tations?

Yes, the ontology is complete re-

garding the requirements speciﬁ-

cations that are designed in the

identiﬁcation of requirements, and

all competency questions are an-

swered.

Conciseness Are there any irrelevant axioms con-

cerning the domain to be covered?

Does it support a minimal ontological

commitment? Are there any weak as-

sumptions regarding the ontology’s un-

derlying domain?

Yes, the ontology is concise since

it does not store any unnecessary or

useless deﬁnitions.

Expandability Does the ontology ﬂexible enough to

support new deﬁnitions and axioms? Is

the ontology be expanded by adding

new knowledge to classes without al-

tering the already deﬁned concepts?

Yes, the ontology is expandable

since adding or modifying the con-

cepts does not inﬂuence other ax-

ioms and classes.

Sensitiveness How is the ontology affected regarding

altering the semantics of the ontology?

The ontology is not sensitive since

it is expandable, meaning that

changes in deﬁnitions of different

concepts did not affect the other de-

ﬁned concepts.

from PhySci ontology. CQ1.Which publications use

relativity theory to solve the particle problem?

The output of query CQ1 gives the publication ti-

tle that is discussed for the solution of the particle

problem. The results are listed as follows; publica-

tion: “The Particle Problem in the General Theory

of Relativity”, problem:“Particle Problem”, and solu-

tion: “A Special Kind of Singularity and Removal”.

(3) Data-driven Approach: The corpus-based ter-

minological ontology approach assesses the coverage

and the capacity of the ontology (Rospocher et al.,

2012). This approach has many advantages to deter-

mine the uncertainty of domain-speciﬁc terminology;

therefore, it can provide a precise output to rank the

relevancy of a knowledge domain.

Listing 1: SPARQL example for query CQ1 (in Table 4).

SELECT DISTINCT ? pbl ? p r blm ? sol

{

? pbl phy s ci : a dd r es ses Re s ea r ch Wo r k ? wrk .

? wrk phy s ci : h a sP r obl e m ? p r blm .

? sol phy s ci : so l v e ? pr b l m .

? sol phy s ci : u s es T heo r y ? the o r y .

? th e ory rd f s : la bel ? label .

? pr b l m rdfs : l a b el ? plabe l .

FILTER (regex( ? l a bel , " r e lat i vit y ")

&&regex(? p labe l , " Par t icle " ))

}

It starts with the extraction of concepts from the

deﬁned corpus by applying TF-IDF, then each ex-

tracted concept is compared with the ontology to ﬁnd

similar class terms. Next, the number of overlapped

concepts between the ontology and corpora are listed.

After that, metrics(precision, recall, F1) can be ap-

plied by using the number of classes in the ontology

and the number of matched concepts. Two different

corpora are generated to examine how far PhySci cov-

ers different topics of physics. The main objective of

this approach is to select corpora that contain different

topics than the corpus that is used in the development

of the PhySci. Additionally, we compared different

types of ontologies against the text corpus to see how

well PhySci is suitable for the domain to be repre-

sented with respect to other ontologies. All ontolo-

gies are assessed to check that they adequately deﬁne

Semantic Representation of Physics Research Data

Table 4: Competency questions for PhySci ontology.

Query Competency Question

CQ1 Which publications use theory X to solve problem Y?

CQ2 Which concepts is used in publications for the theory Y?

CQ3 What are the results of Research Work X?

CQ4 List all the equations that are used in publication X?

CQ5 Which publications use scientiﬁc model X?

CQ6 Who are the authors of publications that use research work X for problem Y?

CQ7 Which equations, theories are used in Scientiﬁc Method X?

CQ8 Which observational data is used in observation X with observatory Y?

CQ9 Which phenomena are observed in Observation X with scientiﬁc method Y?

CQ10 Give me the velocity, energy of the signal that is found in the horizon X of the

Formation Y formed from Formation Z?

the terminology and represent the most relevant con-

cepts appropriately. We chose ontologies that are the

most current ontologies in their ﬁeld and closest to the

physics domain.

OM Ontology (Ontology of Units of Measure and

Related Concepts) (Rijgersberg et al., 2013) is an

ontology about the science domain and developed to

improve the alignment and representation of quantita-

tive research data.

OPB Ontology (Ontology of Physics for Biol-

ogy) (Cook et al., 2008) is a reference ontology of

physical principles(classical physics and thermody-

namics) that can be applied to the bioinformatics

modeling. It is developed to bridge the gap between

the biosimulation, biological processes, and physical

domains (e.g., ﬂuid dynamics and particle diffusion)

to annotate biosimulation models.

ENVO Ontology (Environment Ontology)

(Buttigieg et al., 2013) is an ontology for deﬁning a

broad range of environments related to ecosystems,

environmental processes, and habitats.

We established a corpus-based evaluation based

on these ontologies and Corpus1

and Corpus2

evaluate the coverage of each ontology against each

other. Corpus1 has produced from the scientiﬁc pub-

lications that involve the topics of atomic, molec-

ular, and optical physics (Physical review A) pub-

lished in APS. A total of 25 articles are added to the

dataset. The search results of Google Scholar gener-

ated Corpus2 by performing the keyword “black holes

in string theory”. The top 50 keywords are captured

by applying TF-IDF to the datasets.

Precision =

hits

class

(4)

http://www.environmentontology.org

https://github.com/aysegulsay/PhySci/blob/master/

Datasets/DatasetCorpus1.tsv

https://github.com/aysegulsay/PhySci/blob/master/

Datasets/DatasetCorpus2.tsv

Recall =

hits

|List|

(5)

F1 =

2 × Precision ×Recall

Precision + Recall

(6)

Then, precision, recall, and f1 values are calcu-

lated, which are depicted in Equation 4, Equation 5,

and Equation 6, respectively. We perform analysis us-

ing the Corpus1, Corpus2, OM, OPB, and ENVO on-

tology. Table 5 presents the output of assessments. It

includes the results of precision, recall, F1, the total

number of classes, and hits as the number of matched

concepts among the ontology and the corpus. Many

classes in PhySci have matched with top-ranked con-

cepts extracted from each corpus (17 hits and 15 hits

among the top 50 key-concepts) while the compared

ontologies have a lower number of hits among the

top 50 key-concepts even though they contain more

classes than PhySci. Although the precision and re-

call values of PhySci are still relatively low, it scores

signiﬁcantly higher than the benchmark ontologies.

However, to increase the precision and recall values,

PhySci would be extended by aligning with different

ontologies or adding new concepts to capture more

terms from the scientiﬁc literature. Also, the results

show that the F1 value of PhySci is greater than 0.15,

which means that the PhySci covers more knowledge

than other ontologies. This method helps to conﬁrm

that PhySci is sufﬁciently aligned with the deﬁned do-

main of interest.

5 RELATED WORK

The availability of encyclopedic and factual knowl-

edge representation in machine-actionable form is in-

creased and resulted in different knowledge graphs,

such as DBpedia (Lehmann et al., 2015). However,

there is a scarcity of developing science-based ontolo-

gies, especially for the physics domain. Thus, most

KEOD 2020 - 12th International Conference on Knowledge Engineering and Ontology Development

Table 5: Data-driven approach, corpus-based evaluation results.

Corpus1 Corpus2

PhySci OM OPB ENVO PhySci OM OPB ENVO

Classes 110 808 846 6,240 110 808 846 6,240

Precision 0.15 0.01 0.01 0.002 0.13 0.01 0.01 0.002

Recall 0.34 0.26 0.30 0.30 0.30 0.28 0.32 0.28

F1 0.20 0.03 0.03 0.004 0.18 0.03 0.03 0.004

approaches do not cover scholarly data with physics

knowledge; but only interpret articles or scholarly

outputs in the light of more general rhetorical ele-

ments. For the scholarly domain, semantic publishing

is applied as an approach to undertake the challenges

of scholarly communication by utilizing the meta-

data concepts such as Semantic Publishing Referenc-

ing Ontologies (SPAR) (Peroni and Shotton, 2018).

SPAR covers different ontology modules (e.g., DoCo,

FaBiO, and DEO) to support distinctive features of

the scholarly publishing domain together with seman-

tic technologies, for example, document description,

bibliometric data, and workﬂow processes. Springer

Nature Publishing’s SN SciGraph

focuses primarily

on bibliographic data in the scholarly domain. It pro-

vides a rich semantic fabric of bibliographic metadata

for the visualization of the scholarly domain. Many

attempts have been made (Fathalla et al., 2017; Ja-

radeh et al., 2019a; Vogt et al., 2020; Say et al., 2020)

with the purpose of representing research contribu-

tions as knowledge graphs aiming at improving sci-

entiﬁc data management and retrieval. The Seman-

tic Survey Ontology (Semsur) (Fathalla et al., 2017;

Fathalla et al., 2018) is one of the preliminary at-

tempts to design an ontology for systematizing and

linking research ﬁndings presented in surveys in com-

puter science. The Open Research Knowledge Graph

(ORKG) (Jaradeh et al., 2019b; Jaradeh et al., 2019c)

is a semantic publishing platform presenting a knowl-

edge graph to retrieve and explore scientiﬁc knowl-

edge that is described in scholarly literature. It aims

to represent research in a structured manner for easier

access by changing the document-oriented workﬂows

in scholarly communication.

For the science domain, PhySH (Physics Subject

Headings)

(Smith, 2019) is a physics taxonomy

that is presented by the American Physical Society to

manage subject indexes in physics. The ultimate ob-

jective of PhySH is to provide a fully open and high-

quality classiﬁcation scheme in the ﬁeld of physics.

This data model is developed to connect subjects with

papers submitted and published in Physical Review

https://www.springernature.com/gp/researchers/

scigraph

https://physh.aps.org

journals. There are numerous ontologies presented in

the life science domain, such as MeSH (Medical Sub-

ject Headings) (Lipscomb, 2000) is a thesaurus devel-

oped for indexing articles from the Medline database.

OntoBio (Albuquerque et al., 2016) is a biodiver-

sity domain ontology, and it is designed by adopt-

ing SABIO methodology (de Almeida Falbo, 2014).

It is a formal ontology for biological collection and

ﬁeld data collection of biotic entities. Eventually,

concerning its coverage, PhySci incorporates features

related to all physics and scholarly domains of re-

search works that are published and found as hetero-

geneous data. Thus, PhySci, in comparison with all

those works, fulﬁlls the uncovered requirements of a

physics and scholarly communication domain by rep-

resenting document-based information in the form of

metadata.

6 CONCLUSION

In this study, we examined the possibility of apply-

ing linked data principles to physics research data by

developing PhySci ontology. We provide an ontology

that allows researchers to reuse, access, and ﬁnd sci-

entiﬁc knowledge that will assist their research. The

dynamic content of PhySci enables the exploration of

non-obvious information found in publications. In

this work, we leveraged semantic technologies and

knowledge graphs to transform single query patterns

of physics research data into a sophisticated ongo-

ing conversation between computers and researchers.

Thus, PhySci can support the organization of the con-

tent of Physics publications by describing scientiﬁc

information semantically. PhySci helps to deal with

the information overload and facilitates the transfor-

mation of the document-oriented workﬂows in schol-

arly communication by enabling extensibility, ﬂexi-

bility, and interoperability of scientiﬁc data. It also

allows the indexing of articles that are used for search,

curation, and augmentation. The PhySci ontology sat-

isﬁes the quality requirements according to the results

of the ontology testing, data-driven, and content eval-

uation.

In future work, we will target the adaptation of

Semantic Representation of Physics Research Data

PhySci within the Open Research Knowledge Graph

(ORKG)

to facilitate the discoverability of physics-

related publications. Besides, the precision and recall

of the PhySci ontology will be improved by cover-

ing more topics and sub-topics related to physics re-

search such as electricity and magnetism, or mechan-

ics in the future. Furthermore, we will extend this

work for other scientiﬁc disciplines and envision a

science knowledge graph covering various scientiﬁc

ﬁelds (e.g., life science, earth science) for scholarly

publishing.

ACKNOWLEDGEMENTS

This work has been supported by ERC project Sci-

enceGRAPH (grant no. 819536).

REFERENCES

Abbott, B. P., Abbott, R., Abbott, T., Abernathy, M., Ac-

ernese, F., Ackley, K., Adams, C., Adams, T., Ad-

desso, P., Adhikari, R., et al. (2016). Gw150914: Im-

plications for the stochastic gravitational-wave back-

ground from binary black holes. Physical review let-

ters, 116(13):131102.

Albuquerque, A. C. F., Dos Santos, J. L. C., and de Cas-

tro J

unior, A. N. (2016). Ontobio: Designing new

features to improve modeling and implementation. In

ONTOBRAS, pages 169–174.

Buttigieg, P. L., Morrison, N., Smith, B., Mungall, C. J.,

Lewis, S. E., Consortium, E., et al. (2013). The en-

vironment ontology: contextualising biological and

biomedical entities. Journal of biomedical semantics,

4(1):43.

Cambr

esy, L., Derriere, S., Padovani, P., Preite Mar-

tinez, A., and Richard, A. (2010). Ontology of

astronomical object types. International Virtual

Observatory Alliance. Available at: http://www. ivoa.

net/Documents/Notes/AstrObjectOntology/20100117/

NOTE-AstrObjectOntology-1.3-20100117. html.

Accessed on, 11(06):2011.

Compton, M., Barnaghi, P., Bermudez, L., Garc

ıA-Castro,

R., Corcho, O., Cox, S., Graybeal, J., Hauswirth, M.,

Henson, C., Herzog, A., et al. (2012). The ssn on-

tology of the w3c semantic sensor network incubator

group. Web semantics: science, services and agents

on the World Wide Web, 17:25–32.

Cook, D. L., Mejino Jr, J. L., Neal, M. L., and Gennari, J. H.

(2008). Bridging biological ontologies and biosimu-

lation: the ontology of physics for biology. In AMIA

Annual Symposium Proceedings, volume 2008, page

136. American Medical Informatics Association.

Corcho, O., Fernandez-Lopez, M., and Gomez-Perez, A.

(2007). Ontological engineering: what are ontologies

https://projects.tib.eu/orkg/

and how can we build them? In Semantic web ser-

vices: Theory, tools and applications, pages 44–70.

IGI Global.

de Almeida Falbo, R. (2014). Sabio: Systematic approach

for building ontologies. In ONTO. COM/ODISE@

FOIS.

Fathalla, S., Auer, S., and Lange, C. (2020). Towards the se-

mantic formalization of science. In Proceedings of the

35th Annual ACM Symposium on Applied Computing,

pages 2057–2059.

Fathalla, S., Vahdati, S., Auer, S., and Lange, C. (2017). To-

wards a knowledge graph representing research ﬁnd-

ings by semantifying survey articles. In International

Conference on Theory and Practice of Digital Li-

braries, pages 315–327. Springer.

Fathalla, S., Vahdati, S., Auer, S., and Lange, C. (2018).

Semsur: a core ontology for the semantic representa-

tion of research ﬁndings. Procedia Computer Science,

137:151–162.

Feynman, R. P., Leighton, R. B., and Sands, M. (1965). The

feynman lectures on physics; vol. i. American Journal

of Physics, 33(9):750–752.

Gangemi, A., Catenacci, C., Ciaramita, M., and Lehmann,

J. (2005). Ontology evaluation and validation: an

integrated formal model for the quality diagnostic

task. On-line: http://www. loa-cnr. it/Files/OntoE-

val4OntoDev Final. pdf.

omez-P

erez, A. (2001). Evaluation of ontologies. Interna-

tional Journal of intelligent systems, 16(3):391–409.

Hammond, T. and Pasin, M. (2015). The nature. com on-

tologies portal. In LISC@ ISWC, pages 2–14.

Hey, A. J., Tansley, S., Tolle, K. M., et al. (2009). The

fourth paradigm: data-intensive scientiﬁc discovery,

volume 1. Microsoft research Redmond, WA.

Horrocks, I., Patel-Schneider, P. F., Boley, H., Tabet, S.,

Grosof, B., Dean, M., et al. (2004). Swrl: A semantic

web rule language combining owl and ruleml. W3C

Member submission, 21(79):1–31.

Jaradeh, M. Y., Auer, S., Prinz, M., Kovtun, V.,

Kismih

ok, G., and Stocker, M. (2019a). Open re-

search knowledge graph: Towards machine action-

ability in scholarly communication. arXiv preprint

arXiv:1901.10816.

Jaradeh, M. Y., Oelen, A., Farfar, K. E., Prinz, M., D’Souza,

J., Kismih

ok, G., Stocker, M., and Auer, S. (2019b).

Open research knowledge graph: Next generation in-

frastructure for semantic scholarly knowledge. In Ke-

jriwal, M., Szekely, P. A., and Troncy, R., editors,

Proceedings of the 10th International Conference on

Knowledge Capture, K-CAP 2019, Marina Del Rey,

CA, USA, November 19-21, 2019, pages 243–246.

ACM.

Jaradeh, M. Y., Oelen, A., Prinz, M., Stocker, M., and Auer,

S. (2019c). Open research knowledge graph: A sys-

tem walkthrough. In Doucet, A., Isaac, A., Golub, K.,

Aalberg, T., and Jatowt, A., editors, Digital Libraries

for Open Knowledge - 23rd International Conference

on Theory and Practice of Digital Libraries, TPDL

2019, Oslo, Norway, September 9-12, 2019, Proceed-

KEOD 2020 - 12th International Conference on Knowledge Engineering and Ontology Development

ings, volume 11799 of Lecture Notes in Computer Sci-

ence, pages 348–351. Springer.

Landauer, T. K., Foltz, P. W., and Laham, D. (1998). An in-

troduction to latent semantic analysis. Discourse pro-

cesses, 25(2-3):259–284.

Lehmann, J., Isele, R., Jakob, M., Jentzsch, A., Kon-

tokostas, D., Mendes, P. N., Hellmann, S., Morsey,

M., Van Kleef, P., et al. (2015). Dbpedia–a large-

scale, multilingual knowledge base extracted from

wikipedia. Semantic Web, 6(2):167–195.

Lipscomb, C. E. (2000). Medical subject headings

(mesh). Bulletin of the Medical Library Association,

88(3):265.

Lovrencic, S. and Cubrilo, M. (2008). Ontology evaluation-

comprising veriﬁcation and validation. In Central

European Conference on Information and Intelligent

Systems, page 1. Faculty of Organization and Infor-

matics Varazdin.

Madin, J., Bowers, S., Schildhauer, M., Krivov, S., Pen-

nington, D., and Villa, F. (2007). An ontology for de-

scribing and synthesizing ecological observation data.

Ecological informatics, 2(3):279–296.

McGuinness, D. L., Van Harmelen, F., et al. (2004). Owl

web ontology language overview. W3C recommenda-

tion, 10(10):2004.

Medeiros, L., Lauer, T. R., Psaltis, D., and

Ozel, F. (2018).

Principal component analysis as a tool for characteriz-

ing black hole images and variability. The Astrophys-

ical Journal, 864(1):7.

Musen, M. A. et al. (2015). The prot

e project: a look

back and a look forward. AI matters, 1(4):4.

Noy, N. F., McGuinness, D. L., et al. (2001). Ontology

development 101: A guide to creating your ﬁrst ontol-

ogy.

Peroni, S. and Shotton, D. (2018). The spar ontologies. In

International Semantic Web Conference, pages 119–

136. Springer.

Proctor, M. (2011). Drools: a rule engine for complex event

processing. In Proceedings of the 4th international

conference on Applications of Graph Transforma-

tions with Industrial Relevance, pages 2–2. Springer-

Verlag.

Ramos, J. et al. (2003). Using tf-idf to determine word rele-

vance in document queries. In Proceedings of the ﬁrst

instructional conference on machine learning, volume

242, pages 133–142. Piscataway, NJ.

Raskin, R. and Pan, M. (2003). Semantic web for earth and

environmental terminology (sweet). In Proc. of the

Workshop on Semantic Web Technologies for Search-

ing and Retrieving Scientiﬁc Data, volume 25.

Rijgersberg, H., Van Assem, M., and Top, J. (2013). Ontol-

ogy of units of measure and related concepts. Seman-

tic Web, 4(1):3–13.

Rospocher, M., Tonelli, S., Seraﬁni, L., and Pianta, E.

(2012). Corpus-based terminological evaluation of

ontologies. Applied Ontology, 7(4):429–448.

Say, Z., Fathalla, S., Vahdati, S., Lehmann, J., and Auer,

S. (2020). Ontology design for pharmaceutical re-

search outcomes. In International Conference on The-

ory and Practice of Digital Libraries, pages 119–132.

Springer.

Smith, A. (2019). Physics subject headings (physh). ISKO

Encyclopedia of Knowledge Organization.

Vogt, L., D’Souza, J., Stocker, M., and Auer, S. (2020).

Toward representing research contributions in schol-

arly knowledge graphs using knowledge graph cells.

In Proceedings of the ACM/IEEE Joint Conference on

Digital Libraries in 2020, pages 107–116.

Weibel, S., Kunze, J., Lagoze, C., and Wolf, M. (1998).

Dublin core metadata for resource discovery. Internet

Engineering Task Force RFC, 2413(222):132.

Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., Apple-

ton, G., Axton, M., Baak, A., Blomberg, N., Boiten,

J.-W., da Silva Santos, L. B., Bourne, P. E., et al.

(2016). The fair guiding principles for scientiﬁc data

management and stewardship. Scientiﬁc data, 3.

Semantic Representation of Physics Research Data