=Paper=
{{Paper
|id=Vol-3003/paper2
|storemode=property
|title=Structural Alignment Method of Conceptual Categories of Ontology and Formalized Domain
|pdfUrl=https://ceur-ws.org/Vol-3003/paper2.pdf
|volume=Vol-3003
|authors=Eduard Manziuk,Iurii Krak,Olexander Barmak,Olexander Mazurets,Vladyslav Kuznetsov,Olexander Pylypiak
|dblpUrl=https://dblp.org/rec/conf/profitai/ManziukKBMKP21
}}
==Structural Alignment Method of Conceptual Categories of Ontology and Formalized Domain==
https://ceur-ws.org/Vol-3003/paper2.pdf
Structural Alignment Method of Conceptual Categories of
Ontology and Formalized Domain
Eduard Manziuk 1, Iurii Krak 2, Olexander Barmak 1, Olexander Mazurets 1, Vladyslav
Kuznetsov3, Olexander Pylypiak 1
1
Khmelnytskyi National University, 11, Instytuts’ka str., Khmelnytskyi, 29016, Ukraine
2
Taras Shevchenko National University of Kyiv, 60, Volodymyrska str., Kyiv, 01033, Ukraine
3
Glushkov Cybernetics Institute, 40, Glushkov avenue, Kyiv, 03187, Ukraine
Abstract
The problem of the structural method of ontology alignment and the more formally
represented structured domain is considered. The applied area of research belongs to the field
of ethical AI. The ontology developed on the basis of the ISO / IEC TR 24028 standard -
Overview of trustworthiness in Artificial Intelligence, and the formalized research based on
the corpus of gray literature which represents global landscape is investigated. Presented a
structured alignment method used for manual alignment. The method is part of a general
system of alignment and is based on building relationships about the study entity on the
domain of ontology and finding the appropriate structure on the structured domain. The
method is based on the semantic structures of concepts and relationships between them. More
formally, the emphasis is on semantic relationships and the search for appropriate semantic
structures to determine alignment at the level of the structure of relationships.
The aim of the study is to detect the compliance of the trustworthiness ontology with the
current global state of the problem and the existing global trend in the field of AI ethics. The
structural method has shown that semantic relationships with the domain of research are an
important element and stage of alignment. Semantic relationships play an important role
because they can be used to detect the alignments of concepts, despite the fact that the corpus
has been documented in different languages and with a different lexical notation of concepts.
The results of the research showed that the ontology based on the ISO / IEC TR 24028
standard adequately corresponds to the global view on the issue of AI ethics.
Keywords 1
alignment ontology, structural alignment, data integration, trustworthiness, ethic AI.
1. Introduction
Artificial Intelligence (AI) is expanding the scope of its practical application. It allows performing
tasks that are difficult to describe formally. This is one of the most promising computer technologies,
which corresponds to modern trends and the practicality of its application should be highly valued. AI
has been studied for a long time in view of the speed of development of computer technologies, but
recently in the areas of machine learning and deep learning has gained accelerating development.
This is due to the prospects of application in a variety of applications of a wide range of human
activities. Methods are developing visual analytics to use human intellectual capabilities in machine
learning [1]–[3], methods based on ensembles of models are used [4]–[7], methods used to reduce the
dimensionality of features in classification, and clustering [8]–[11] and other areas. The scope of
application is quite wide in information security [12], [13], telecommunication [14], [15], medicine
International Workshop of IT-professionals on Artificial Intelligence (ProfIT AI 2021), September 20–21, 2021, Kharkiv, Ukraine
EMAIL: eduard.em.km@gmail.com (E. Manziuk); krak@univ.kiev.ua (I. Krak); alexander.barmak@gmail.com (O. Barmak);
exechong@gmail.com (O.Mazurets); kuznetsow.wlad@gmail.com (V. Kuznetsov); oleksandrpylypiak7@gmail.com (O.Pylypiak)
ORCID: 0000-0002-7310-2126 (E. Manziuk); 0000-0002-8043-0785 (I. Krak); 0000-0003-0739-9678 (O. Barmak); 0000-0002-8900-0650
(O.Mazurets); 0000-0002-1068-769X (V. Kuznetsov); 0000-0002-3246-3590 (O.Pylypiak)
©️ 2021 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
[16], and so on. Due to the rapid development and intensive spread of AI, the impact on society as a
whole and the individual, in particular, is manifested in a variety of ethical, law, and social
challenges. These challenges can significantly reduce the value and benefits of AI implementation if
these challenges are not properly addressed. In addition, there is the prospect of negative impacts on
human life, which sometimes have critical, threatening, and dangerous consequences. The risks of
widespread AI implementation are already becoming apparent in the early stages.
Risks range from violating an individual's confidentiality, discriminating against him on a set of
features, to the global consequences of the economic crisis from managing e-commerce, managing
mass security systems, and malicious manufacturing, and so on.
In order to prevent the creation of negative consequences from the introduction of AI, prevention
of risks and dangers, as well as preserving the benefits and positive effects of AI on human life and
society on a global scale at the present stage of human development are considered AI. As part of this
concept, the EU Commission’s High-Level Expert Group on AI (AI HLEG) European AI Alliance in
the “Ethics Guidelines for Trustworthy AI” [17], [18] proposed the Trustworthy AI concept. Within
the framework of ethical AI, this concept is responsible for defining the boundaries of trust and the
necessary sets of factors for the use of AI in terms of safety, risk minimization, especially in critical,
important for human life areas.
For example, the use of AI when driving a car. The driver must trust the AI decision, only then
he'll agree to implement and use it. In the field of health care, the physician who uses AI in the
diagnosis and as a recommendation system should trust AI, as well as understand why AI gives such a
recommendation. If AI in some areas from the modern point of view cannot fully replace a person, it
can significantly increase the efficiency of his work and bring to a qualitatively new level the
likelihood of correct decisions that people make if they trust AI. There are also industries such as
electronic markets that make extensive use of computer technology. The introduction of AI in this
area can significantly affect these markets. However, this impact can have global negative
consequences, and therefore to prevent it, one must be sure that the introduction of AI will not lead to
this. AI trust in this case also plays a key and decisive role in implementation.
Research on AI trust covers research areas of ethical AI, human-machine interaction, machine
learning and the use of information technology by society, management methods using information
technology, and the role of trust in these areas based on the experience of different levels of AI use.
The level of importance and interest in this direction is indicated by high-level guidelines and
recommendations of gray literature and soft-law documents at the world level. Thus, the issue of AI
trust went beyond study and reached a level of broad discussion in order to develop a general, global
approach to AI trust.
2. Related works
In general, trustworthy AI is an interdisciplinary and sphere of research, and a necessary set of
knowledge on technical and non-technical aspects extends to a wide field of disciplines and
manifested in diverse studies. This circumstance makes it difficult to conduct objective studies that
have fully taken into account all aspects and identified its implementation.
In this study, we aim to determine the structure of trustworthy AI and promote this important
direction. Since the research field is quite wide and known different directions and interpretations of
the semantic structure of trustworthiness. It is also necessary to determine the development of
trustworthy AI at the global level, taking into account the diversity of directions and approaches in the
understanding of this concept. To do this, we use the study of Jobin A. et. al [19], which based on the
created document corps, summarizes world experience for developing ethical AI and ontology of
trustworthy AI [20]. The features of the created corps are determined by the fact that it consists of
soft-law levels and are gray literature from different countries and are presented in a variety of
languages. Accordingly, these documents are trying to highlight public opinion on issues of
importance and significance of ethical AI. They are also used to develop wording, semantic meaning,
informative content, etc., to create a basis for further development of the widespread introduction of
AI in society. In addition, the advantage of using these studies is that the corps documents already
contain a certain level of generalization and represent the opinion, and quite often a wide range of
people, also at the level of state institutions, professional unions, world corporations, and more. The
analysis of corps documents will give an objective knowledge of the trustworthy AI development at
the world level. In order to become a degree of alignment with world trends, use ontology based on
ISO / IEC TR 24028 [21]. In addition, the alignment also pursues the purpose of determining the
importance and significance of structural components of the trustworthy AI concept.
With the development of AI and in accordance with recent trends, numerous new directions and
branches of AI have been formed. For example, research in such areas as beneficial AI [22], [23],
responsible AI [24]–[27], ethical AI [15], [28]–[30], fairness AI [31]–[34] and others. Despite the
diversity of the presented concepts and areas of research in general, the content can be attributed to
the same goals. These goals can be summarized as promoting the development of AI so that the
benefits of its use are maximized, but the risks and dangers have to be eliminated. In order to unify
the wording, recommendations on ethical AI have been developed, and AI standards are being
introduced, for example, ISO / IEC TR 24028: 2020, ISO / IEC DIS 22989, ISO / IEC TR 24030:
2021 and others.
The heterogeneity of semantic structures in the subject information field is largely due to the
modeling of the same conceptual categories of reality by different approaches. This is due to the fact
that the rapid development of AI occurs simultaneously and globally. And to date, no stable
formulations and named designations have been developed. Syntactic constructions are used in
different formats and content. Subject concepts need to be formalized in order to establish
correspondence and harmonize lexical structures. It should be noted that the role of humans in the
processes of analysis is sometimes significant and decisive [35]. One of the most convenient methods
of formalizing knowledge is the development of subject areas ontologies. Ontologies contain a
conceptual description of data as formalized knowledge in the concepts form of their properties and
the relationships between these structures. At the same time, in the context of the study, the ontology
should show world experience and meet current trends. The presentation of concepts should be
objective in semantic terms.
Alignment ontology - a technique that is widely used and based on establishing the
correspondence of individual concepts [36], based on the structure, i.e. taking into account the
structure of the organization of several concepts [37], the use of meta heuristics, data related to
ontological concepts [38].
The most well-known method and one that has been continued in many works is OWL (Web
Ontology Language) [39]. The method contains better semantic informativeness than the methods on
the knowledge graphs. It can be used for complex alignment of axioms using logical constructors and
entities from the target to the source ontology. These axioms form an alignment of ontologies.
Matching in OWL is determined by the logic of relationships in the domains of alignment ontologies.
The method POMI (Pattern mining for Ontology Matching-based Instances) [40] allows to obtain
the necessary knowledge based on the correlation of the basic ontology and the resulting using
approaches to clustering and modeling patterns. Clustering is performed using the k-means algorithm.
The peculiarity of the method is that it groups the instances of each ontology, and then alignment the
two ontologies based on a comparison of clusters and the corresponding instances within the clusters.
Another method uses a system of mapping agents to compare ontologies [41]. Agents establish
semantic bridges between concepts in the source and resulting ontologies. Mismatches in alignment
are processed using the confidence values for each semantic bridge and the complexity of processing
the mismatch.
The SFA (Similarity Flooding Algorithm) method [42]-[44] is used to alignment structural
ontologies called models. Models are represented by data schemes that are converted into a directed
graph. The algorithm is based on the position that the concepts are similar, if the adjacent concepts on
the graph are similar. The algorithm begins work on obtaining the initial relations between the
concepts on two graphs using the correspondence function corresponding to the initial state of
comparison [45]-[48]. After obtaining the initial data values [49]-[51], the algorithm continues to
matching based on the fact that the concepts are more likely to be similar the more likely they are to
be related to other concepts. When achieving a stable distribution of probabilities between concepts,
he degree of correspondence between the concepts of ontologies is determined [52].
3. Model of structural alignment ontology and structured domain
To make the structural comparison given by the function structure ( ) , we take as a basis for
research [45]–[52]. The structural relations of entities within domains EntO and EntDs we are
denoted by sets LinkO and Link Ds . Correspondence for structural alignment is defined as follows
A+B
;
2 ent ( Link O )
structure ( ent , ent ) = A = linkDs ent , y :linkO ,Ds , filter x , y linkO ent ,x (xEntO )(yEntDs ) ( ent ( Link Ds ) ) ;
( ) ( ) ( )
B = linkO ,Ds , filter x , y :linkO ent ,x (yEntDs )(zEntO ) ( LinkO , Ds , filter ) , (1)
( ) ( )
s.t. structure ( ent , ent ) structure ( ent , ent ) , structure ( ent , ent ) 0,...,1 ,
linkO , Ds , filter ( ent ,ent ) , ent EntO , ent Ent Ds , !link = max w ( Link ) , w 0,...,1.
Structural alignment is formed using relations. The alignment of the entities ent EntO with
existing relations on the set EntO LinkO( ent , x ) where x x | linkO( ent , x ) . That is, a set of connections
of the comparison entity from the basic set of comparison of the trust ontology with other entities
from the same set is obtained. That is, a set of relations of the alignment entity from the basic set of
alignment of the trust ontology with other entities from the same set is obtained. The entity from
EntO is similar to the entity or subset of entities from the set EntDs . This is determined by the
(
function of selecting relationships between sets EntO and EntDs - ent LinkO , Ds , filter . If the set )
( )
of existing entities ent LinkO , Ds , filter 1 , explores all the similarities in the existing relationships
with the entity ent on the set LinkO , Ds , filter . Relationships are sorted by weights
( )
sortw ent ( LinkO , Ds , filter ) , w 0,...,1 , and then removed from the sorted list. Then the function
of alignment is calculated by (1) and a set of weight values of each of the available similarities is
formed in relation to the essence of the study. This creates a sequence of definitions of relationships
and entities from the base set as related to ent , relationships between entities on the set EntDs , and
relationships between sets EntO and EntDs .Since the relationship is removed, the next relationship
with the maximum weight is selected. This approach is used for all sets of
relationship ( link Link ) : !link = max w ( Link ) . The best case for structural alignment is the
presence for each entity from EntO , which is related to the entity of alignment, a similar entity or a
subset of entities on the set EntDs , x | link (
O ent , x )
, linkO , Ds , filter ( x , y ) , x EntO , y Ent Ds .
From this set, we can select sets of relationships: the set of relationships on EntO with the entity of
the study ent - linkO( ent , x ) | ent , x EntO ; the set of relationships between EntO and EntDs with
entities on EntO that have relationships with the entity of the study -
link O , Ds , filter ( x , y )
| linkO( ent , x ) , x EntO , y Ent Ds , . Since each entity from EntO , at best, that is
related to the entity of the study should have a similar entity from EntDs , and given
( link Link ) : !link = max ( Link ) , i.e. such a relationship at each stage of the study only one,
w
can be written
link ( ) | ent, x Ent =
O ent , x O
(2)
link ( ) | link ( ) , x Ent , y Ent ,
O , Ds , filter x , y O ent , x O Ds
That is, the powers of these sets will be equal. Accordingly, the best-considered case can be
represented as a double value of the projection power entity ent on LinkO
2 ent ( LinkO ) (3)
When studying the relationship structure of the entity of the study, it is necessary to detect the
presence of relationships of compatible entities EntO with similar entities on EntDs . In real tasks,
this may not correspond to the described best case. The existing relationships are determined by
LinkO , Ds , filter , i.e., the set of the entity relations
link O , Ds , filter ( x , y )
| linkO( ent , x ) , x EntO , y Ent Ds , LinkO , Ds , filter . We present this set of
relations in the form of constraints of the selection function on LinkO , Ds , filter .
link (
O , Ds , filter ( x , y ) :linkO( ent , x ) yEnt Ds ) ( zEntO ) ( Link O , Ds , filter ) (4)
On the set EntDs defines all the relationships with relation ent to such a relative entity ent .
Associated entities with ent on EntDs must also have similar entities, i.e. relationships with entities
on EntO . This determines the set of relationships Link Ds relative to ent
link ( Ds ent , y )
| linkO , Ds , filter ( x , y ) linkO( ent , x ) , x EntO , y Ent Ds , Link Ds . We present this
set of relations in the form of constraints of the selection function on the projection on the entity ent
on Link Ds
link ( (
Ds ent , y ) :linkO , Ds , filter ( x , y ) linkO( ent , x ) xEntO ) ( yEntDs ) ( ( Link ) )
ent Ds (5)
On a set of entities EntDs in a structured domain Ds , the number of relationships ent relative to
other entities in the same set may exceed the number of relationships relative to a similar study entity
ent on EntO with entities on the same set EntO . This is due to the broader representation of
conceptual structures in the structured domain, which is presented as an area of comparison, ie
Dom O Dom Ds . Structural alignment is based on a direct alignment of entities, as well as on the
alignment of related entities within the relative sets.
The relationship between the entities ent and ent not taking into account the direct relationship
of similarity entRent , R = linkO , Ds , filter ( ent ,ent ) i.e. entlinkO , Ds , filter ( ent ,ent ) ent can be written as a non-
transitive relationship
( ent , ent ) , ent EntO , ent Ent Ds : ( entR1 x ) ( xR2 y ) ( yR3ent ) ( entRent ) ,
(6)
R1 LinkO , R2 LinkO , Ds , filter , R1 Link Ds , x EntO , y Ent Ds
The best case of the degree of entities relationship ent and ent is to consist of the ratio of
transitivity of entities EntO realted to ent and related to ent
( x, ent ) , x EntO , ent Ent Ds : ( entR1 x ) ( xR2 y ) ( yR3ent ) ( entRent ) ,
(7)
R1 LinkO , R2 LinkO , Ds , filter , R1 LinkDs , ent EntO
To detect the degree of the ratio between entities ent EntO and ent EntDs present an
example in the form of a graphical representation of the scheme of entities, their sets, and the related
relationships within the sets of relationships LinkO , Link Ds and LinkO , Ds , filter .Entity relationships
are denoted as graph edges, with the corresponding relation vertices as entities. Relationships that are
represented by edges denote within the affiliation of the corresponding sets of entities.
Figure 1: Scheme of relationships of structural alignment of entities ent4 EntO and ent3 EntDs
for detection of the alignment degree
Thus, according to the figure, we define three types of relationships that combine entities in
alignment system:
1. Relationships within the trust ontology AI domain dom O , and connect entities that have a
certain relationship to the study entity, i.e. the entity relatively which the structure of relations with
other entities of this domain.
2. Relationships within the structured domain of the corpus dom Ds , and connect entities that
have a certain relationship to the entity that in the current step is maximum similar as possible to the
study entity. The structure of relations to other entities of the domain Ds is built concerning this
entity. The structure is built using informativeness, which is limited exclusively to this domain
without taking into account similarity to the ontology domain O . This is due to the fact that the
informativeness present in the field of ontology did not affect the structure of relations in the field of
alignment.
3. Relationships between entities of domains O and Ds . In the study of detecting the degree of
alignment with the structural approach, we build links between the entities of domains that have the
appropriate links within the respective domains with the entities of the baseline comparison, i.e. the
study entity and the closest relative entity at the current stage of alignment.
To detect the degree of alignment concerning the entity of trust ontology ent4 EntO and a
similar entity from a structured domain ent3 EntDs . The possibility of comparison is due to existing
the similarity linkO , Ds , filter ( ent 4,ent 3) LinkO , Ds , filter . It should be noted that studies are conducted based
on the value of the weighting relationships. That is, at the current stage, the only similarities between
the entities are established to the one that corresponds to the only one connection represented by the
edges between these entities. That is, at the current stage, we study one similarity that exists between
the entities from EntO and EntDs that correspond to the only one connection represented by the
edges between these entities. If the entity has several similar elements, an alignment scheme is studied
for each case. In the final case, the choice is made by the numerical value of the degree of the greatest
compliance.
The entity ent4 EntO has a relationship to the entities from EntO . Relationships are defined on
LinkO .
ent4 EntO
→ linkO( ent 4,ent1) → ent1
→ linkO( ent 4,ent 2) → ent2
→ linkO( ent 4,ent 3) → ent3
→ linkO( ent 4,ent 5) → ent5
We get a set of entities ent1 , ent4 , ent6 Ent Ds and a set of relationships
link (
Ds ent 3,ent 1)
, linkDs ( ent 3,ent 4) , linkDs (ent 3,ent 6) Link Ds . The entity ent3 EntDs has relationships to
the entities from EntDs . Relationships are defined on Link Ds .
ent3 EntDs
→ link Ds( ent 3,ent1) → ent1
→ link Ds( ent 3,ent 4) → ent4
→ link Ds( ent 3,ent 6) → ent6
We get a set of entities ent1 , ent4 , ent6 Ent Ds and a set of relationships
link (
Ds ent 3,ent 1)
, linkDs ( ent 3,ent 4) , linkDs (ent 3,ent 6) Link Ds .
ent4 EntO → linkO , Ds , filter ( ent 4,ent 3) → ent3 EntDs
ent1 → linkO , Ds , filter ( ent1,ent 6) → ent6
ent2 → linkO , Ds , filter ( ent 2,ent 4) → ent4
ent3 → linkO , Ds , filter ( ent 3,ent1) → ent1
ent5 → linkO , Ds , filter ( ent 5,ent 8) → ent8
Next we note that
ent3 EntDs
→ link Ds( ent 3,ent 8) → ent8
That is the corresponding relation between the entities ent3 EntDs and ent8 EntDs is not
detected.
There may also be cases of lack of relationships, i.e. a similar element from EntDs for the entities
that belong to EntO and are related to the study entity ent EntO .
In the general case, this can be noted as follows
ent EntO → linkO , Ds , filter ( ent ,ent ) → ent EntDs
entx → linkO , Ds , filter ( entx ,enty ) → ent y
However, it should be noted the following: to detect the degree of alignment of the study entity
ent EntO , there must always be a similar entity ent EntDs and at the current stage only one
relationship is explored !linkO , Ds , filter ( ent ,ent ) . This link is selected from a list of links sorted by the
maximum previous similarity level. Subsequently, the processed relationship is removed from this list
and with the corresponding numerical indicator, defined as the result of the structural method of
alignment and relocated to the list of studied relationships for further analysis.
Also, entities x EntO that have nothing to do with the study entity ent EntO ,
linkO( ent , x ) cannot be considered, although they may have similar entities y EntDs and links to
it linkO , Ds , filter ( x , y ) .
4. Experimental studies
The analysis using a structural alignment of the ontology and the structured domain allowed us to
determine the importance of the concepts of AI trustworthiness.
Figure 2: The relative importance of the concept based on the data of Jobin A. (2019) [19]
distribution of ethical principles of AI using structural alignment
The change of the named designations of concepts was also taken into account, i.e. concepts that
had a similar structure could have different lexical names. At the level of categories of concepts
{Transparency} and {Privacy}, in accordance, the importance is determined by the share of 20% and
13%. The structure of the {Explainability} attachment has a complete match with {Transparency}.
Next, alignments were made using lexical variability using codes for named designations {strategies
for reducing Bias} is the code of the concept of {Justice & fairness} and {functional Safety} is the
code of the concept {Non-maleficence} and are relevant to 18% and 16% respectively. For smaller
shares of similarity, i.e. not complete structural equivalents are {Controllability} has a lexical
equivalent {Freedom & autonomy} and has an importance of 9%. Further, with less overlap of
structural relationships, the concept of {Reliability, Resilience, Robustness} partially corresponds to
{Trust} has a relative importance of 8%, the concept of {Testing, Evaluation} partially corresponds to
{Justice & fairness} has a relative importance of 18% with almost similar structural relationships. The
categories {mitigating system hardware Faults}, {Use}, {Applicability} are presented at the level of
group proximity by structural links. The alignment was made in order to maximize the inclusion of
concepts and generalizations. The alignment showed the effectiveness of the method of structural
alignment of ontology and structured domain.
5. Discussion and conclusions
The developed method of structural alignment of ontologies showed the effectiveness of practical
application on the example of alignment of ontology and structured domain. It should be noted that
the structured domain is presented as a result of content analysis of the corps of gray literature
provides ample opportunities for practical alignment. A structured domain is less formalized and more
flexible in finding matches to ontology concepts. The presence of semantic correspondence codes and
descriptions allows for structural alignment. Given that the method was developed for manual
alignment of formalized structures, it makes it possible to obtain a convenient tool for alimenting
ontologies.
Among other methods, it allows a comprehensive and objective approach to alimenting concepts
that have structural relationships. The proposed method represents only one approach based on
structural relationships. It can be effectively applied to structural ontologies as areas with common
principles. It is also an effective method that, along with methods based on other principles of
alignment, provides objectivity and comprehensiveness. One of the advantages is the formalization of
practical application.
The practical implementation of the structural method of alignment made it possible to determine
the level of importance of the structural components of the concept of AI trustworthiness. Also,
generalize the structural relationship of the concept with the definition of directions for further
practical implementation in specific embodiment of AI. That is, related concepts are practically
provided by one tool within the implementation of AI. This has a significant advantage in terms of a
systems approach with the allocation and integration of areas of responsibility in areas without losing
aspects of implementation.
The accordance of the ontology of trustworthy AI with the global landscape of ethical AI has
shown the correctness of the need to further formalize AI methods at a practical level while ensuring
compliance with ethics AI.
6. References
[1] J. Yuan, C. Chen, W. Yang, M. Liu, J. Xia, and S. Liu, “A survey of visual analytics techniques
for machine learning,” Comp. Visual Media, vol. 7, no. 1, pp. 3–36, Mar. 2021, doi:
10.1007/s41095-020-0191-7.
[2] A.V. Barmak, Y. V. Krak, E. A. Manziuk, and V.S. Kasianiuk, “Information technology of
separating hyperplanes synthesis for linear classifiers”, J. Autom. Inf. Sci. 51(5), 54–64 (2019).
doi:10.1615/JAutomatInfScien.v51.i5.50
[3] I.G. Kryvonos, I. V. Krak, O.V. Barmak, and A.I. Kulias, “Methods to create systems for the
analysis and synthesis of communicative information”, Cybern. Syst. Anal. 53(6), 847–856
(2017). doi:10.1007/s10559-017-9986-7
[4] S. Fort, H. Hu, and B. Lakshminarayanan, “Deep Ensembles: A Loss Landscape Perspective,”
pp. 1–15, Dec. 2019.
[5] O. V. Barmak, Y. V. Krak, and E. A. Manziuk, “Characteristics for choice of models in the
ansables classification,” in CEUR Workshop Proceedings, 2018, vol. 2139, pp. 171–179,
doi:10.15407/pp2018.02.171
[6] O. Barmak, I. Krak, and E. Manziuk, “Diversity as The Basis for Effective Clustering-Based
Classification,” in CEUR Workshop Proceedings, Odessa, Ukraine, Sep. 2020, vol. 2711, pp.
53–67.
[7] E. A. Manziuk et al., “Approach to creating an ensemble on a hierarchy of clusters using model
decisions correlation,” Przegląd Elektrotechniczny, vol. 96, no. 9, pp. 108–113, 2020, doi:
10.15199/48.2020.09.23
[8] A. Qayyum, A. Malik, N. M Saad, and M. Mazher, “Designing deep CNN models based on
sparse coding for aerial imagery: a deep-features reduction approach,” European Journal of
Remote Sensing, vol. 52, no. 1, pp. 221–239, Jan. 2019, doi: 10.1080/22797254.2019.1581582.
[9] I.G. Kryvonos, I. V. Krak, O.V. Barmak, and R.O. Bagriy, “New tools of alternative
communication for persons with verbal communication disorders”, Cybern. Syst. Anal. vol. 52,
no. 5, pp. 665–673, Sep. 2016, doi:10.1007/s10559-016-9869-3
[10] I. Krak, O. Barmak, E. Manziuk, and A. Kulias, “Data Classification Based on the Features
Reduction and Piecewise Linear Separation,” in Advances in Intelligent Systems and
Computing, 2020, vol. 1072, pp. 282–289, doi: 10.1007/978-3-030-33585-4_28.
[11] I. V. Krak, G.I. Kudin, and A.I. Kulias, “Multidimensional scaling by means of pseudoinverse
operations”, Cybern. Syst. Anal. vol. 55, no. 1, pp. 22–29, Jan. 2019, doi:10.1007/s10559-019-
00108-9
[12] G. Markowsky, O. Savenko, S. Lysenko, and A. Nicheporuk, “The Technique for Metamorphic
Viruses’ Detection Based on its Obfuscation Features Analysis,” vol. 2014, pp. 680–687.
[13] A. Nicheporuk, O. Savenko, A. Nicheporuk, and Y. Nicheporuk, “An Android Malware
Detection Method Based on CNN Mixed-data Model,” vol. 2732, pp. 198–213.
[14] I.R. Parhomey, J.M. Boiko and O.I. Eromenko, "Features of digital signal processing in the
information control systems of multipositional radar", Journal of Achievements in Materials and
Manufacturing Engineering, vol. 77, no. 2, pp. 75-84, 2016.
https://doi.org/10.5604/17348412.1230101.
[15] J. Boiko, I. Pyatin, O. Eromenko and M. Stepanov, “Method of the adaptive decoding of self-
orthogonal codes in telecommunication,” Indonesian Journal of Electrical Engineering and
Computer Science, vol. 19, no. 3, pp. 1287-1296, Sept. 2020,
http://doi.org/10.11591/ijeecs.v19.i3.pp1287-1296.
[16] J. H. Kamdar, J. Jeba Praba, and J. J. Georrge, “Artificial Intelligence in Medical Diagnosis:
Methods, Algorithms and Applications,” in Machine Learning with Health Care Perspective:
Machine Learning and Healthcare, V. Jain and J. M. Chatterjee, Eds. Cham: Springer
International Publishing, 2020, pp. 27–37. doi: 10.1007/978-3-030-40850-3_2.
[17] “Ethics guidelines for trustworthy AI | Shaping Europe’s digital future.” URL: https://digital-
strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai.
[18] “Assessment List for Trustworthy Artificial Intelligence (ALTAI) for self-assessment | Shaping
Europe’s digital future.” URL: https://digital-strategy.ec.europa.eu/en/library/assessment-list-
trustworthy-artificial-intelligence-altai-self-assessment.
[19] A. Jobin, M. Ienca, and E. Vayena, “The global landscape of AI ethics guidelines,” Nat Mach
Intell, vol. 1, no. 9, pp. 389–399, Sep. 2019, doi: 10.1038/s42256-019-0088-2.
[20] E. Manziuk, O. Barmak, I. Krak, O. Mazurets, and T. Skrypnyk, “Formal Model of Trustworthy
Artificial Intelligence Based on Standardization.,” in CEUR Workshop Proceedings,
Khmelnytskyi, Ukraine, Mar. 2021, vol. 2853, pp. 190–197.
[21] “ISO/IEC TR 24028:2020,” ISO. URL:
https://www.iso.org/cms/render/live/en/sites/isoorg/contents/data/standard/07/76/77608.html.
[22] P. M. Fernandes, F. C. Santos, and M. Lopes, “Norms for beneficial A.I.: A computational
analysis of the societal value alignment problem,” AI Communications, vol. 33, no. 3–6, pp.
155–171, Jan. 2020, doi: 10.3233/AIC-201502.
[23] S. D. Baum, “On the promotion of safe and socially beneficial artificial intelligence,” AI & Soc,
vol. 32, no. 4, pp. 543–551, Nov. 2017, doi: 10.1007/s00146-016-0677-0.
[24] A. Barredo Arrieta et al., “Explainable Artificial Intelligence (XAI): Concepts, taxonomies,
opportunities and challenges toward responsible AI,” Information Fusion, vol. 58, pp. 82–115,
Jun. 2020, doi: 10.1016/j.inffus.2019.12.012.
[25] R. Clarke, “Principles and business processes for responsible AI,” Computer Law & Security
Review, vol. 35, no. 4, pp. 410–422, Aug. 2019, doi: 10.1016/j.clsr.2019.04.007.
[26] A. Askell, M. Brundage, and G. Hadfield, “The Role of Cooperation in Responsible AI
Development,” arXiv:1907.04534 [cs], pp. 1–23, Jul. 2019.
[27] B. Rakova, J. Yang, H. Cramer, and R. Chowdhury, “Where Responsible AI meets Reality:
Practitioner Perspectives on Enablers for Shifting Organizational Practices,” Proc. ACM Hum.-
Comput. Interact., vol. 5, no. CSCW1, pp. 1–23, Apr. 2021, doi: 10.1145/3449081.
[28] E. Gibney, “The battle for ethical AI at the world’s biggest machine-learning conference.,”
Nature, vol. 577, no. 7791, p. 609, Jan. 2020.
[29] J. T. Hancock, M. Naaman, and K. Levy, “AI-Mediated Communication: Definition, Research
Agenda, and Ethical Considerations,” Journal of Computer-Mediated Communication, vol. 25,
no. 1, pp. 89–100, Mar. 2020, doi: 10.1093/jcmc/zmz022.
[30] H. Muller, M. T. Mayrhofer, E.-B. Van Veen, and A. Holzinger, “The Ten Commandments of
Ethical Medical AI,” Computer, vol. 54, no. 7, pp. 119–123, doi: 10.1109/MC.2021.3074263.
[31] S. Caton and C. Haas, “Fairness in Machine Learning: A Survey,” arXiv:2010.04053 [cs, stat],
pp. 1–33, Oct. 2020.
[32] A. Chouldechova and A. Roth, “A snapshot of the frontiers of fairness in machine learning,”
Commun. ACM, vol. 63, no. 5, pp. 82–89, Apr. 2020, doi: 10.1145/3376898.
[33] S. Kafle, A. Glasser, S. Al-khazraji, L. Berke, M. Seita, and M. Huenerfauth, “Artificial
intelligence fairness in the context of accessibility research on intelligent systems for people
who are deaf or hard of hearing,” SIGACCESS Access. Comput., no. 125, p. 4:1, Mar. 2020,
doi: 10.1145/3386296.3386300.
[34] N. Mehrabi, F. Morstatter, N. Saxena, K. Lerman, and A. Galstyan, “A Survey on Bias and
Fairness in Machine Learning,” ACM Comput. Surv., vol. 54, no. 6, p. 115:1-115:35, Jul. 2021,
doi: 10.1145/3457607.
[35] I. Krak, O. Barmak, and E. Manziuk, “Using visual analytics to develop human and machine-
centric models: A review of approaches and proposed information technology,” Computational
Intelligence, pp. 1–26, 2020, doi: 10.1111/coin.12289.
[36] I. Leshcheva and A. Begler, “A method of semi-automated ontology population from multiple
semi-structured data sources,” Journal of Information Science, pp. 1–17, Aug. 2020, doi:
10.1177/0165551520950243.
[37] H. Belhadi, K. Akli-Astouati, Y. Djenouri, and J. C.-W. Lin, “Data mining-based approach for
ontology matching problem,” Appl Intell, vol. 50, no. 4, pp. 1204–1221, Apr. 2020, doi:
10.1007/s10489-019-01593-3.
[38] Q. Lv, C. Jiang, and H. Li, “Solving Ontology Meta-Matching Problem Through an
Evolutionary Algorithm With Approximate Evaluation Indicators and Adaptive Selection
Pressure,” IEEE Access, vol. 9, pp. 3046–3064, 2021, doi: 10.1109/ACCESS.2020.3047875.
[39] L. Zhou et al., “The Enslaved Dataset: A Real-world Complex Ontology Alignment Benchmark
using Wikibase,” in Proceedings of the 29th ACM International Conference on Information &
Knowledge Management, New York, NY, USA, Oct. 2020, pp. 3197–3204. doi:
10.1145/3340531.3412768.
[40] Y. Djenouri, H. Belhadi, K. Akli-Astouati, A. Cano, and J. C.-W. Lin, “An ontology matching
approach for semantic modeling: A case study in smart cities,” Computational Intelligence, pp.
1–27, doi: 10.1111/coin.12474.
[41] V. Nascimento, M. J. Viamonte, A. Canito, and N. Silva, “An agent-based electronic market
simulator enhanced with ontology matching services and emergent social networks,”
International Journal of Simulation and Process Modelling, vol. 10, no. 3, pp. 265–278, Jan.
2015, doi: 10.1504/IJSPM.2015.071376.
[42] A. M. Shahhosseini, H. Ye, G. Maughan, and T. Foster, “Implementation of Similarity Flooding
Algorithm to Solve Engineering Problems Using Diagnostic Skills Training Technique,”
presented at the ASME 2014 International Mechanical Engineering Congress and Exposition,
Mar. 2015. doi: 10.1115/IMECE2014-39698.
[43] L. Pan, Z. Wang, J. Li, and J. Tang, “Domain Specific Cross-Lingual Knowledge Linking Based
on Similarity Flooding,” in Knowledge Science, Engineering and Management, Cham, 2016,
pp. 426–438. doi: 10.1007/978-3-319-47650-6_34.
[44] J. Liu, Y. Wang, P. Xie, Y. Wang, and Z. Huang, “Malware Similarity Analysis Based on Graph
Similarity Flooding Algorithm,” in Advances in Computer Science and Ubiquitous Computing,
Singapore, 2015, pp. 31–37. doi: 10.1007/978-981-10-0281-6_5.
[45] M. N. Arifin and D. Siahaan, “Structural and Semantic Similarity Measurement of UML Use
Case Diagram,” LKJITI, vol. 11, no. 2, pp. 88–100, Jul. 2020, doi:
10.24843/LKJITI.2020.v11.i02.p03.
[46] T. T. A. Nguyen and S. Conrad, “Ontology Matching using multiple similarity measures,” in
2015 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and
Knowledge Management (IC3K), Nov. 2015, vol. 01, pp. 603–611.
[47] M. Chmielewski, M. Paciorkowska, and M. Kiedrowicz, “Ontology similarity assessment based
on lexical and structural model features extraction.,” Wseas, vol. 14, pp. 134–144, 2017.
[48] O. Udrea, L. Getoor, and R. J. Miller, “Leveraging data and structure in ontology integration,”
in Proceedings of the 2007 ACM SIGMOD international conference on Management of data,
New York, NY, USA, Jun. 2007, pp. 449–460. doi: 10.1145/1247480.1247531.
[49] C. Zhou et al., “Geoscience knowledge graph in the big data era,” Sci. China Earth Sci., vol. 64,
no. 7, pp. 1105–1114, Jul. 2021, doi: 10.1007/s11430-020-9750-4.
[50] Yu.G. Krivonos, Yu. V. Krak, Yu. V. Barchukova, and B.A. Trocenko, “Human hand motion
parametrization for dactilemes modeling”, J. Autom. Inf. Sci. vol. 43, no 12, pp. 1–11, Dec.
2011, doi:10.1615/JAutomatInfScien.v43.i12.10
[51] Iu.V. Krak, O.V. Barmak, and S.O. Romanyshyn, “The method of generalized grammar
structures for text to gesture computer-aided translation”, Cybern. Syst. Anal. vol. 50, no. 1, pp.
116–123, Jan. 2014, doi:10.1007/s10559-014-9598-4
[52] A. Essayeh, and M. Abed, “Towards Ontology Matching Based System Through
Terminological, Structural and Semantic Level,” Procedia Computer Science, vol. 60, pp. 403–
412, Jan. 2015, doi: 10.1016/j.procs.2015.08.154.