Challenges of Modeling and Evaluating the Semantics of Technical

Content Deployed in Recommendation Systems for Industry 4.0

Jos Lehmann, Michael Shamiyeh and Sven Ziemer

Bauhaus Luftfahrt e.V., Willy-Messerschmitt-Straße 1, Taufkirchen, 82024, Germany

Keywords:

Modeling, Evaluation, Deployment of Semantics, Web Ontology Language, Industry 4.0, Aviation.

Abstract:

In the context of Industry 4.0 the Smart Factory is enabled by the automation of physical production activities.

The automation of intellectual pre-production activities enables what is here dubbed the “Smart Studio”. A

key-element of the Smart Studio is Semantic Technology. While prototyping an ontology-based recommen-

dation system for technical content about the case-study of the aviation industry, the problem of the readiness

level of Semantic Technology became apparent. This led to the formulation of a Semantic Modeling and Tag-

ging Methodology. The evaluation of both prototype and methodology yielded valuable insight about (i) the

quantity and quality of semantics needed in the Smart Studio, (ii) the different interaction proﬁles identiﬁed

when testing recommendations, (iii) the efﬁciency and effectiveness of the methods required to achieve se-

mantics of right quantity and quality, (iv) the extent to which an ontology-based recommendation system is

feasible and reduces double work for knowledge workers. Based on these results in this paper a position is

formulated about the challenges for the viable application of Semantic Technology to technical content in

Industry 4.0.

1 INTRODUCTION

Research on Industry 4.0 develops and evaluates soft-

ware prototypes to assess how and to what extent dig-

ital technology can make industrial processes more

ﬂexible and efﬁcient (Vogel-Heuser et al., 2017). As

discussed in (Lehmann et al., 2017) and in (Lehmann

et al., 2018), digital technology can potentially ef-

fect two new paradigms: the Smart Factory, an es-

tablished concept in research on Industry 4.0, and

what, by analogy, can be dubbed the “Smart Studio”.

In the Smart Factory Robotics, Cyber-physical Sys-

tems (CPS) and the Internet of Things (IOT) can en-

able the automatic reconﬁguration of Production and

Logistics lines, increasing productivity and reduc-

ing costs. In the Smart Studio Knowledge Manage-

ment (KM) and Artiﬁcial Intelligence (AI) can reshape

pre-production processes, from Conceptual Design to

Prototyping, by supporting or automating the intellec-

tual activities that make enterprise knowledge avail-

able when relevant – a goal comparable to what pur-

sued in Knowledge-based Engineering and Ontology-

based Design (Li and Ramani, 2007).

A key-element of both the Smart Factory and the

Smart Studio is Semantic Technology (ST) (Bifﬂ and

Sabou, 2016). The use of semantics in the Smart Fac-

tory is characterized by a focus on physical activities.

Here observations through sensors can be leveraged in

order to update and correct derivations about a given

state of affairs. The use of semantics in the Smart

Studio, on the other hand, cannot rely on sensing. It

therefore has to rely to a larger extent on logical in-

ference, reasoning or other forms of association, in

order to harmonize and integrate information sources

about multidisciplinary subjects (e.g. textual corpora,

model data, expert knowledge) and to ﬂesh out im-

plicit meanings and consequences.

At present it is unclear how easily semantic tech-

nology, especially its modeling tools, can help estab-

lish the new work ﬂow of the Smart Studio. In (Xu

et al., 2018) this point is made more generally for

many of the technologies that are being applied in

research on Industry 4.0 and that lack the required

readiness level.

While prototyping an ontology-based recommen-

dation system for technical content about the case-

study of the aviation industry, intending to test to

what extent this type of system is feasible and whether

it reduces double work for knowledge workers, the

problem of the readiness level of Semantic Technol-

ogy became apparent. This led to breaking down this

problem into a number of sub-problems (discussed

Lehmann, J., Shamiyeh, M. and Ziemer, S.

Challenges of Modeling and Evaluating the Semantics of Technical Content Deployed in Recommendation Systems for Industry 4.0.

DOI: 10.5220/0008348503590366

In Proceedings of the 11th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2019), pages 359-366

ISBN: 978-989-758-382-7

359

in Section 2 of this paper) about the modeling and

evaluation of semantics when deployed in the Smart

Studio. This in turn led to the formulation of a Se-

mantic Modeling and Tagging Methodology (Section

3). The evaluation of both prototype and methodol-

ogy (Section 4) yielded valuable insight about (i) the

quantity and quality of semantics needed in the Smart

Studio, (ii) the different interaction proﬁles that may

be identiﬁed when testing recommendations, (iii) the

efﬁciency and effectiveness of the methods required

to achieve semantics of right quantity and quality,

(iv) the extent to which an ontology-based recommen-

dation system is feasible and reduces double work for

knowledge workers. Based on these results in Section

5 a position is formulated about the challenges for the

viable application of Semantic Technology to techni-

cal content in Industry 4.0.

2 HYPOTHESES, PROTOTYPE,

PROBLEMS

Research on the Smart Studio usually assumes that

the automated fostering of enterprise knowledge is

both possible and essential to increasing the produc-

tivity of knowledge workers. In order to pin down and

partly test this assumption research hypotheses RH1

and RH2 were formulated.

RH1. It is technically feasible to implement an

ontology-based recommendation system (OBRS) that

provides in real time references to legacy data to

knowledge workers as they compile new technical re-

ports.

RH2. An OBRS increases the knowledge workers’

productivity by supporting the reuse of existing

knowledge and avoid double work.

Despite existing successful examples of ontology-

based recommendation systems in many domains, the

emphasis of RH1 is on the ontological integration

of technical content. This type of content is highly

structured and detailed, making it more challenging

to strike the right balance between the abstraction re-

quirements of ontological integration and the level of

precision required by the users of technical content.

In order to test RH1 and RH2 the design, imple-

mentation and evaluation of the prototype PR1 was

undertaken:

PR1. A prototype OBRS that provides in real-time ref-

erences to legacy technical data to knowledge work-

ers of an aviation ﬁrm as they compile in a text editor

new technical reports about aircraft components. The

prototype is developed around an existing linguistics-

and statistics-based recommendation engine, origi-

nally designed for recommendations of short non tech-

nical text items. The transition to a OBRS for technical

content is accomplished by semantic-linguistic mod-

eling and tagging, not by re-engineering the recom-

mendation engine.

PR1’s development was based on the generic ar-

chitecture for multilingual semantic applications sup-

porting enterprise knowledge reuse shown in Figure

1. As reported in (Lehmann et al., 2018) with the

same formulation but more detail, the architecture

comprises four main phases (the alternating gray ver-

tical bars, read left to right).

Raw Data Acquisition. Textual data, e.g. PDF ﬁles

of technical reports and related non-textual data, e.g

STEP ﬁles of component models, are selected for in-

tegration.

Semantic Modeling. Raw data are sampled in order

to extract domain knowledge and compile ontological

structures that enable semantic tagging.

Linguistic-Semantic Integration. Textual data are

tagged based on term frequency of terms described

in the ontology. The ontology is also used to semi-

automatically tag non-textual data. All tagged data

are then stored in a database.

User Interaction. The database is dynamically ac-

cessed by a text editor to provide end-users with rec-

ommendations of existing reports or models that may

contain information that is relevant to the contents be-

ing typed in.

(Chen and Wu, 2008) discusses an architecture for

a comparable class of document recommendation sys-

tems, though based to a larger extent on user prefer-

ences rather than semantic modeling.

Working on PR1 raised two groups of problems

about the scaling-up of established approaches to data

integration. On the one hand, data-related problems

(DP1 through DP6 below) had to do with the optimiza-

tion of software components that, as shown in Fig-

ure 1, are located in the parts of the architecture, that

support the preparation and enrichment of the tens of

thousands of documents available to the OBRS plug-

in. Although not further discussed in this paper the list

of these problems provides some context on issues of

data preparation required by an OBRS.

DP1. Large Quantity of Documents.

DP2. Layout of Documents.

DP3. Large Size of Documents.

DP4. Integration of additional and relevant Metadata

avalaible to Data Owner.

DP5. Suitable Client-Server Architecture.

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

360

Figure 1: Architecture of multilingual semantic application, OBRS PR1, supporting enterprise knowledge reuse.

DP6. Connection to Data Sources.

On the other hand, semantics-related problems

(SP1 through SP7) surfaced in various parts of the ar-

chitecture shown in Figure 1. They in large part had

to do with providing the OBRS plug-in with semantics

of the right quality in the right quantity.

SP1. Coverage of the Domain Knowledge required

by the Use Cases.

SP2. Representation of the High Level of Detail of

the Vocabulary of the Model Data.

SP3. Representation of the Lexical Variations and

Multilingualism of the Documents.

SP4. Extraction of Large Quantities of Knowledge.

SP5. Ontology Reusability.

SP6. Semantic Tagging of Textual Data.

SP7. Semantic Tagging of Model Data.

3 SEMANTIC MODELING AND

TAGGING

As reported in (Lehmann et al., 2018) the engineer-

ing of appropriate ontologies requires to accurately

model the information ecosystem in which the OBRS

is meant to operate. On the one hand, a normative

modeling approach makes sure that documents are

correctly classiﬁed with respect to the domain of ref-

erence. On the other hand, when deploying an ontol-

ogy onto the corpus, issues speciﬁc to the corpus in-

terfere with correct interpretation: lexical variations,

multilingualism, abbreviations, technical data.

Figure 2 presents an updated version of the

methodology introduced in (Lehmann et al., 2018),

which supports the transition of information from

sample data to an ontology. After sampling a de-

scriptive document, e.g. a nomenclature for a landing

gear extension/retraction system, a conversion into

a spreadsheet takes place in step SM2a. Then, the

contents (found in headers, lists, indexes, tables etc.)

are classiﬁed as individuals of a minimal number of

classes (e.g., system, component, part) and ordered

by an hyponym property (e.g., narrower-than). For

instance, the individual of class component retrac-

tion actuator is narrower-than, i.e. an aspect of or

a functional part of, an individual landing gear exten-

sion/retraction system of class system. In turn such

system is narrower-than, i.e. an aspect of or a func-

tional part of, individuals nose landing gear resp.

main landing gear of class component. Finally, these

components are aspects of an individual of class mas-

tersystem, which classiﬁes the most generic systems

for inferential convenience. Similarly, classes mas-

tercomponent and masterpart group the most generic

components resp. parts.

The resulting hierarchy mixes-up class/subclass,

class/instance, whole/part hierarchies, which are dis-

entangled in step SM2(b)iD. After ontological checks

and name assignments (SM2(b)i to SM2(b)iii), the

OWL ontology (DOCO-R) is set-up via a mapper.

Challenges of Modeling and Evaluating the Semantics of Technical Content Deployed in Recommendation Systems for Industry 4.0

361

SM1. Sample Data

sample reports and design models that are relevant to use-

cases

SM2. Master-Spreadsheet

(a) Prepare Sources

export content of existing multilingual documentation (such

as pdf ﬁles of reports as textual data, spreadsheets of com-

ponent hierarchies generated from design models as non-

textual data) into a spreadsheet and delete irrelevant parts

(b) Create Ontology’s Basic Version

i. Ontological Modeling

ﬁrst version of ontology is created in reiﬁed form

A. Create Classes and Hyponym Property

from sources’ section names, table headers

B. Create Individuals

from sources’ section content, table entries

C. Assert Hyponym Property between Individuals

from sources’ section content, table entries

D. Qualify Hyponym Property between Individuals

disentanglement of relationships class/subclass,

class/instance, whole/part

ii. Ontological Checks

A. Translate Individuals IRI’s

into main language of ontology

B. Find Duplicate Individuals

as exact match, partial match, no match

iii. Assign names to Individuals as Annotations

including synonyms, abbreviations, their grammatical

variations (plural, cases)

SM3. OWL Ontology

(a) Export ontology

from Master Spreadsheet to DocO-R

(b) Integrate ontology

i. Abstract Individuals to Classes

from DocO-R to DocO-A

ii. Match Classes

between DoC-A and DesO-A resulting in

AviO-A

iii. Reify Classes to Individuals

from AviO-A to AviO-R

(d) Merge Ontology

(e) Export Ontology

SM4. Semantic Tagging

(a) Textual Data Enrichment

(b) Data Model Integration

Figure 2: Semantic Modeling and Tagging Methodology.

(a) AviO-A (b) AviO-R

Figure 3: Retraction Actuator in AviO.

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

362

DOCO-R is then abstracted as DOCO-A via a con-

version function, which outputs a representation of

the domain knowledge, which exploits the full po-

tential of DESCRIPTION LOGIC (DL) and does away

with classes such as mastersystem, mastercomponent,

masterpart, introduced in the reiﬁed version of the

ontology for inferential convenience. In DOCO-A

the appropriate parts of the contents of DOCO-R are

converted into a DL class hierarchy, while the rest is

represented by appropriate object or data properties.

DOCO-A can be matched with relevant ontologies,

e.g. a Design Ontology DESO-A, to get to a version

that is called Aviation Ontology (AVIO-A in Figure

3a). In this example the class aircraft of DOCO-A is

matched with the class Aircraft of DESO-A thereby

acquiring a part-of relation to the class Fuselage of

DESO-A, not found in the original sample data, given

their focus on the extension/retraction system under

consideration. AVIO-A is then reiﬁed as AVIO-R,

shown in 3b. AVIO-R reduces the number of proper-

ties and classes, reducing the expressivity of the on-

tology, while retaining all annotations.

Both conversions (Abstract, Reify) are akin to

so-called meta-modeling techniques presented for in-

stance in (Welty, 2006), (Glimm et al., 2010), (Jek-

jantuk et al., 2011).

Steps SM3c through SM3e involve checks of

AVIO’s coverage with respect to external knowledge

sources, as well as the transformation of its import

structure and of its serialization, to enable its use by

the tagging modules. These steps are not further dis-

cussed in this paper.

Finally, AVIO-R is used by the Semantic Tex-

tual Data Enrichment module to integrate the term-

frequency-based tagging of the documents with extra

tags derived from the hyponym hierarchy. AVIO-A is

used by the Semantic Data Model Integration module

to provide an expert user of the Data Model Extrac-

tion & Management module (not shown in Figure 1)

with semantic tags for the non-textual data.

4 EVALUATION

The evaluation of PR1 was based on (i) user feed-

back given through questionnaires, (ii) screen shot se-

quences of user interactions that included the assess-

ment by users of the relevance of recommendations,

(iii) interviews, (iv) developer feedback given through

reporting. All this provided evidence on the following

issues:

1. How do PR1’s quantity of semantics (in terms

of domain coverage) or quality of semantics (in

terms of level of detail or of lexical variations)

score on the following scale?

[too poor, poor, ok, rich, too rich]

Note that values poor and rich can either have a

positive or a negative connotation, therefore too

poor resp. too rich are used to indicate a value

beyond what is considered practical by the evalu-

ator.

2. Which types of interaction can be observed be-

tween PR1 and its users and how productive are

such interactions?

3. How do the methods described in Figure 2 score

on the following scales?

[partially efficient, efficient]

[partially effective, effective]

These scales are intended for the assessment of

the amount of effort needed (efﬁciency) when ap-

plying a given method for the solution of a seman-

tic problem i.e. for modeling the minimal amount

of knowledge required by the use cases of a OBRS

(effectiveness).

4. Does the evaluation of points 1 through 3 above

conﬁrm PR1’s feasibility (RH1) and its role in

reducing double-work for knowledge workers

RH2)?

4.1 Evaluation of Semantics in PR1

As shown in Table 1 the quality (in terms of domain

coverage) and the quantity (in terms of level of detail

and lexical variations) of the semantics of PR1 was

evaluated from the perspectives of three stakehold-

ers: the ontology developer, the plug-in developer, the

plug-in user.

The coverage of the domain knowledge required

by the use cases was evaluated as ok by the ontol-

ogy developer and the end user but poor by the plug-

in developer. The ontology developer was satisﬁed

by the amount of domain knowledge provided to PR1

by sampling representative documents. The end user

did not notice obviously lacking semantic tags among

those the OBRS plug-in provided with any given rec-

ommendation – although the end user did not always

agree with some tags that were used for a recommen-

dation. The plug-in developer who, as opposed to

the plug-in user, had direct access to the ontology,

expected more domain knowledge to drive the rec-

ommendation engine, beyond what was harvested by

sampling the data pool.

The representation of the high level of detail of

the vocabulary of the model data was rich for the

ontology developer, because such data provides a lot

of information about class and part-of hierarchies for

Challenges of Modeling and Evaluating the Semantics of Technical Content Deployed in Recommendation Systems for Industry 4.0

363

Table 1: Evaluation of semantics in PR1.

Semantics View of View of View of

Problem Ontology Plug-in Plug-in

Developer Developer User

Coverage of

ok poor ok

Domain

Knowledge

SP1

Representation

rich too rich too rich

of Detail Level

of Model Data

SP2

Representation

ok poor n/a

of Lexical

Variations

SP3

components, down to their smallest and most com-

mon parts (e.g. O-Ring). From the perspectives of

both the plug-in developer and user this level of detail

was too rich, i.e. problematic, because the list of

recommendations could at times be clogged with rec-

ommendations that were irrelevant despite containing

many occurrences of a term, e.g. O-Ring, which ap-

peared in the input text and was very frequent in the

data pool.

Finally, the representation of the lexical varia-

tions and multilingualism of the documents varied:

for some entities in AVIO lexical variations of their

names was evaluated as rich, having been modeled

extensively, while for other entities the lexical vari-

ation was poor. This led to some dissatisfaction on

the plug-in developer’s part again based on the expec-

tation of as much structure as possible to drive the

recommendation engine.

4.2 Interaction Proﬁling

The relevance assessment of recommendations by the

users allowed to identify the interaction proﬁles de-

scribed below.

Unaimed search-Experimental tester (UE): In this

type of interaction the OBRS users saw the sys-

tem as a means to understanding more about a

theme and to ﬁnding documents beyond what

they already knew or expected as recommenda-

tions given the text they input. Users might not

always be satisﬁed with the recommendations –

especially regarding the ontological or linguistic

categories used by the system to tag a given doc-

ument. They appreciated though the interaction

with a system that helps to explore the data pool.

Frequency: 30%

Aimed search-Prudent tester (AP): In this type of in-

teraction users – similarly to UE interactions – saw

the system as a means to exploring the data pool,

aiming though at getting a speciﬁc document high

up in the recommendation list.

Frequency: 10%

Very aimed search-Conservative tester (VC): In this

type of interaction users saw the OBRS from the

standpoint of their usual way of working. They

sought recommendations through a controlled in-

teraction with the system, i.e., they often typed

in just the name of a speciﬁc report, a part num-

ber, a single concept or even only an acronym in

order to test whether the system was able to de-

liver a very speciﬁc document they expected. In

these interactions users were interested in testing

whether the system could perform as well as a

user who, based on expertise and knowledge of

the data pool, knows exactly what to look for and

how to ﬁnd it.

Frequency: 60%

In general, users who preferred VC interactions eval-

uated PR1 less favorably than users who preferred UE

and AP interactions. In this respect, the increase in

productivity yielded by avoiding double work seems

to be more readily available to users who are ready to

invest time in exploring the data pool through the rec-

ommendation list – either because they are interested

in exploring new aspects of the data pool to avoid du-

plication of effort or because they lack expertise and

are uncertain which existing documents they need for

a given task.

4.3 Evaluation of Methods

As shown in Table 2 the modeling and deployment

methods employed for the semantics of PR1 was

mainly evaluated from the perspective of the ontol-

ogy developer.

Sampling textual and non-textual data has proven to

be an efficient way of achieving the minimal cov-

erage and level of detail required by the use cases.

The ﬁrst two rows of Table 1 suggest though that sam-

pling can only bring so far in terms of a satisfactory

behavior of an OBRS. As the number of use cases han-

dled by the system grows, more ontological modeling

is needed to semantically consolidate the system.

Assigning names to individuals as annotations for lex-

ical variations is effective, as it increases the sys-

tem’s reach on the data pool with limited ontological

modeling. As opposed to ontological modeling, an-

notating the ontology is still time-consuming though

and more likely to create redundancies.

Tabulating and mapping for knowledge extraction is

very efficient, as spreadsheets provide a ﬁrst cut

of the ontology in a format in which it is easier than

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

364

Table 2: Evaluation of methods for handling semantics.

Method View of View of

for Problem Ontology Plugin

Developer Developer

Sampling for efficient,

n/aCoverage and Detail partially

SM1 for SP1, SP2 effective

Annotating partially

n/afor Lexical Variations efficient,

SM2(b)iii for SP3 effective

Tabulating and Mapping efficient,

n/afor Knowledge Extraction partially,

SM2 for SP4 effective

Abstraction and Reiﬁcation partially

n/afor Reusability efficient,

SM3(b)i SM3(b)iii for SP5 effective

Textual Data Enrichment efficient, efficient,

for Semantic Tagging partially effective

SM4a for SP6 effective

Data Model Integration partially

n/afor Semantic Tagging efficient,

SM4b for SP7 effective

in ontology editors to sort and to identify explicit or

semi-implicit duplicates in the sample data. Tabula-

tion is not effective for more advanced modeling

and does not support the logical checks provided by

ontology editors.

Abstraction and reiﬁcation are effective ways of

achieving reusabilty of semantic modeling results,

when starting from sample data and a hyponym-type

of relation. On the other hand, implementing and

maintaining the abstraction and reiﬁcation conversion

procedures is time-consuming and requires non trivial

ontological choices.

The two semantic tagging methods employed on tex-

tual resp. non-textual data have complementary ef-

ﬁciency and effectiveness. SM4a can rely on a high

degree of automation because it is based on Natural

Language Processing and on term frequency, yielding

results though that do not always effectively mirror

the core semantics of a given document. Conversely,

SM4b is driven by user interaction thereby tagging

component models more effectively, at the cost of

time consuming tagging sessions.

4.4 Evaluation of RH1 and RH2

The overall evaluation of problems DP1 through DP6

(not discussed in this paper) and SP1 though SP7

allowed to establish that PR1 is technically feasible

(RH1) and that the PR1 improves the productivity of

knowledge workers (RH2). PR1 is most effective in

helping knowledge workers to learn about the seman-

tic structure of the data pool as well as about the po-

sition of single documents within that structure.

5 CONCLUSION

Work on the development and the evaluation of PR1

has helped identify a number of challenges for the vi-

able application of semantic technology in Industry

4.0.

On the one hand, many of the steps that allow for

the preparation and consolidation of semantic struc-

tures for a given organization (e.g. the steps described

in Figure 2) should be integrated in normal, day-to-

day Knowledge Management activities. Organiza-

tions that undertake the transition to a paradigm like

the Smart Factory or the Smart Studio need to develop

the necessary expertise to be able to assess the opti-

mal ratio between effort and semantics quantity and

quality when deployed in recommendation or other

computing systems. These Knowledge Management

activities should be supported by the automation of

the following functionalities:

1. Knowledge extraction in a tabular format (e.g. a

spreadsheet), rather than in an ontology editor, to

make it easier to bootstrap ontologies from sam-

ple data by operating (sorting, ﬁltering, match-

ing, editing) on a large amount of raw input in-

formation. This is needed in order to make a se-

mantic structure speciﬁc to its context of use and

enable its deployment in a speciﬁc information

eco-system. An OWL editor becomes essential at

the later stages of ontological analysis, reﬁnement

and general management of a consolidated ontol-

ogy.

2. Meta-modeling, i.e. (class) abstraction and reiﬁ-

cation to learn and reuse ontologies.

3. Automated matching and automated coverage as-

sessment, to enrich ontologies.

4. Gold standard preparation of a representative sub-

set of the data pool through systematic semantic

tagging, to enable the user-independent evaluation

of ontology-based systems.

5. Automated monitoring of user-interactions, to

achieve a behavioral deﬁnition of the values in the

scales used in Section 4 (e.g. too poor, poor,

etc.) and of the interaction proﬁles.

On the other hand, the end-user needs to be sup-

ported in the interaction with the recommendations by

means of the following functionalities:

6. Explanations of the recommendations with re-

spect to the input text.

7. Migration of OBRS to applications other than text-

editors.

8. Conﬁguration of queries relative to user proﬁle or

function in the organization.

Challenges of Modeling and Evaluating the Semantics of Technical Content Deployed in Recommendation Systems for Industry 4.0

365

ACKNOWLEDGMENTS

The project underlying this

research (EFFPRO 4.0 – In-

tegration and Analysis of De-

sign and Production Data for

a more efficient Development

Process Chain) has received

funding from the German

Federal Ministry for Economic Affairs and Energy

under grant agreement no. 20Y1509E.

Ontology editing and presentation in Prot

http://protege.stanford.edu, US NIGMS/NIH grant

no. GM10331601.

REFERENCES

Bifﬂ, S. and Sabou, M., editors (2016). Semantic Web Tech-

nologies for Intelligent Engineering Applications.

Springer.

Chen, D. and Wu, C. (2008). An ontology-based docu-

ment recommendation system: Design, implementa-

tion, and evaluation. In Paciﬁc Asia Conference on In-

formation Systems, PACIS 2008, Suzhou, China, July

4-7, 2008, page 206. AISeL.

Glimm, B., Rudolph, S., and V

olker, J. (2010). Integrated

metamodeling and diagnosis in OWL 2. In Patel-

Schneider, P. F., Pan, Y., Hitzler, P., Mika, P., Zhang,

L., Pan, J. Z., Horrocks, I., and Glimm, B., editors,

International Semantic Web Conference, Revised Se-

lected Papers, Part I, volume 6496 of Lecture Notes

in Computer Science, pages 257–272. Springer.

Jekjantuk, N., Pan, J. Z., and Qu, Y. (2011). Diagnosis

of software models with multiple levels of abstrac-

tion using ontological metamodeling. In Proceedings

of the International Computer Software and Appli-

cations Conference, pages 239–244. IEEE Computer

Society.

Lehmann, J., Heußner, A., Shamiyeh, M., and Ziemer, S.

(2018). Extracting and modeling knowledge about

aviation for multilingual semantic applications in in-

dustry 4.0. In Sixth International Conference on En-

terprise Systems, ES 2018, Limassol, Cyprus, October

1-2, 2018, pages 56–60. IEEE.

Lehmann, J., Shamiyeh, M., and Ziemer, S. (2017). To-

wards integration and coverage assessment of on-

tologies for knowledge reuse in the aviation sector.

In Fensel, A., Daniele, L., Aroyo, L., de Boer, V.,

Dar

anyi, S., Elloumi, O., Garc

ıa-Castro, R., Hollink,

L., Inel, O., Kuys, G., Maleshkova, M., Merdan, M.,

Mero

no-Pe

nuela, A., Moser, T., Keppmann, F. L.,

Kontopoulos, E., Petram, L., Scarrone, E., and Ver-

borgh, R., editors, Joint Proceedings of SEMANTiCS

2017 Workshops co-located with the 13th Interna-

tional Conference on Semantic Systems (SEMANTiCS

2017), Amsterdam, Netherlands, September 11 and

14, 2017., volume 2063 of CEUR Workshop Proceed-

ings. CEUR-WS.org.

Li, Z. and Ramani, K. (2007). Ontology-based design infor-

mation extraction and retrieval. AI EDAM, 21(2):137–

154.

Vogel-Heuser, B., Bauernhansl, T., and ten Hompel, M., ed-

itors (2017). Handbuch Industrie 4.0 Bd.1-4. Springer

Reference Technik. Springer.

Welty, C. A. (2006). Ontowlclean: Cleaning OWL on-

tologies with OWL. In Bennett, B. and Fellbaum,

C., editors, Formal Ontology in Information Systems,

Proceedings of the Fourth International Conference,

FOIS 2006, Baltimore, Maryland, USA, November 9-

11, 2006, volume 150 of Frontiers in Artiﬁcial Intelli-

gence and Applications, pages 347–359. IOS Press.

Xu, L. D., Xu, E. L., and Li, L. (2018). Industry 4.0: state

of the art and future trends. International Journal of

Production Research, 56(8):2941–2962.

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

366