Knowledge Management for Electronic-Based Government System Using Semantic Thesaurus

Abstract

Sistem Pemerintahan Berbasis Elektronik (SPBE), the electronic-based government system, is Indonesia’s egovernment policy providing services to citizens through information and communication technology. Knowledge containing SPBE must be managed in various ways, one of which is the creation of an SPBE thesaurus to facilitate access and search for SPBE-related items using words or terms about it. In this study, we provide an overview of the thesaurus development process that has complied with the ISO 25964 standard and uses the Simple Knowledge Organization System (SKOS) as the application of the thesaurus in the web environment. Basic concepts or related terms and relationships between concepts have been linked with similar concepts in other thesauri that have existed before. This research also looks at the process of automating the recognition of related terms in internet articles using Word2Vec and Doc2Vec. In the process of adding terms, we discover challenges in filtering terms, determining relationships between terms, and determining reciprocal relationships between terms.

Full text

Jurnal Ilmu Komputer dan Informasi (Journal of Computer Science and Information)
16/2 (2023), 103-113. DOI: http://dx.doi.org/10.21609/jiki.v16i2.1148
103
Knowledge Management for Electronic-Based Government System Using
Semantic Thesaurus
Nuraisa Novia Hidayati1, Agoeng Srimoeljanto2
1 Research Center for Data and Information Science, National Research and Innovation Agency, Serpong,
Indonesia
2 Research Center for Artificial Intelligence and Cybersecurity, National Research and Innovation
Agency, Serpong, Indonesia
nura017@brin.go.id 1, agoe003@brin.go.id 2
Abstract
Sistem Pemerintahan Berbasis Elektronik (SPBE), the electronic-based government system, is Indonesia’s e-
government policy providing services to citizens through information and communication technology.
Knowledge containing SPBE must be managed in various ways, one of which is the creation of an SPBE
thesaurus to facilitate access and search for SPBE-related items using words or terms about it. In this study,
we provide an overview of the thesaurus development process that has complied with the ISO 25964 standard
and uses the Simple Knowledge Organization System (SKOS) as the application of the thesaurus in the web
environment. Basic concepts or related terms and relationships between concepts have been linked with
similar concepts in other thesauri that have existed before. This research also looks at the process of
automating the recognition of related terms in internet articles using Word2Vec and Doc2Vec. In the process
of adding terms, we discover challenges in filtering terms, determining relationships between terms, and
determining reciprocal relationships between terms.
Keywords: SPBE, thesaurus, ISO 25964, SKOS, semantic, e-government, word2vec, doc2vec
1. Introduction
During the last four years, the Indonesian
government has attempted to re-intensify efforts
to promote the digital transformation to provide
better services to Indonesian citizens through
Presidential Regulation No. 95 of 2018
concerning Electronic-Based Government
Systems (SPBE). SPBE strives for clean,
effective, transparent, accountable governance,
quality, and dependable public services. National
SPBE governance and management are also
required to improve the integration and efficiency
of the electronic-based government system. One
of the major initiatives in SPBE is implementing
knowledge management to improve government
services and decision-making.
The implementation of SPBE has closely
related to business transformation, and the process
is undoubtedly very complex. In the previous
studies, implementing enterprise architecture
based on several frameworks is necessary to
achieve the transformation goal [1][2]. Besides
enterprise architecture, a study to build digital
enterprise architecture compared three models for
managing digital repositories to make SPBE much
more efficient and green-minded [3]. SPBE then
became a fundamental issue regarding digitizing
the government system. We discuss the SPBE
architecture, especially the service and
infrastructure domains.
Digitizing public services is currently a
demand for many governments worldwide [4].
Managing organizational knowledge is one of the
most critical aspects of digital transformation.
Sharing and creating organizational knowledge is
made more effective and efficient thanks to
advancements in information and communication
technologies. It is necessary to conduct
knowledge management related to SPBE to
develop and add value to the government’s
products, services, and administrative processes
[5]. However, there seems to be a lack of
knowledge management awareness in the public

104 Jurnal Ilmu Komputer dan Informasi (Journal of Computer Science and Information), volume 16,
issue 2, June 2023
sector [4].
One of the knowledge management processes
can be through the creation of the SPBE semantic
web. The Semantic Web is increasingly used in e-
government to describe and specify e-government
services to achieve semantic integration and
interoperability. Within the layered structure of
the Semantic Web, thesaurus finds a relevant role
in supporting semantic searches and other value-
added services [6]. Several applications are owned
and operated by central and local government
agencies to support the implementation of SPBE
digital services. Semantic Web-based applications
must be developed to support interoperability
between applications that support SPBE services.
The thesaurus can play a role in managing all that,
like how the previous study assembled it to
manage knowledge related to the smart city [7].
This paper introduces the semantic knowledge
thesaurus model using the Simple Knowledge
Organization System (SKOS) to organize
knowledge in SPBE Knowledge Management
System. A semantic thesaurus is a type of
thesaurus that organizes and structures knowledge
by utilizing semantic relationships between terms.
Semantic relationships refer to the meaning and
context of terms and how they relate to one
another. Non-semantic thesauri, on the other hand,
are used in specialized fields where precise
terminology is required, and terms are organized
alphabetically or numerically based on specific
criteria.
We intend to incorporate this SPBE thesaurus
into the knowledge management information
system (SIMPAN) application search engine,
developed and used at the local government level
in version one last year. Implementing a standard
vocabulary for metadata management is a
prospective use of the SKOS thesaurus in
Electronic-Based Government systems, especially
in public service and IT infrastructure. Using the
SKOS thesaurus, government agencies can
establish a unified language for labelling and
organizing digital assets such as documents,
images, and videos. For instance, a government
website that provides information on public
services can use SKOS to establish a standardized
vocabulary for labelling service-related content,
such as forms, guides, and frequently asked
questions. This thesaurus would enable citizens to
quickly locate and access the information they
need while facilitating the government's ability to
manage and maintain its digital assets efficiently.
2. Related Work
Making a thesaurus is collecting information
based on words or terms closely related to
knowledge about the domain. Extraction of the
right words or phrases and the relationship among
them are the core of the thesaurus. Therefore,
sources of information extraction with massive
data, such as Wikipedia, are used [8][9]. Hyperlink
relationships between articles referred to as
structure mining is used to determine the
relationship between words or terms identified as
keywords [8].
The semantic pattern with various Natural
Language Processing (NLP) methods can analyze
words or term relations in a domain. A study
applied sentence simplification, Part Of Speech
tagging (POS), and Finite State Machine (FSM) to
gain information about semantic relation patterns
[9]. The relationships between words or terms first
searched for are hypernyms and synonyms. With
the speed of information related to a predetermined
domain increasing, the thesaurus must
continuously be updated. Searching for words or
other terms in a particular domain in other sources
of information using existing relationships can use
the word embedding model. Previously, the
exploration for synonyms and related words was
expanded using word embedding, specifically
doc2vec, with a determined seed word [10].
Research in Italy for cybersecurity thesaurus uses a
term extraction process with Bidirectional Encoder
Representations from Transformers (BERT) and a
hypernym and synonym relation discovery process
comparing two methods on the word2vec and fast
text models [11]. The study of making a thesaurus
in the cardiology domain identifies hypernym-
hyponym relationships in words and terms and
clusterized them to group concepts with the K-
means method. Also, it involves an expert
intervention to validate the results [12].
3. Methodology
This paper describes the proposed method for
creating an SPBE thesaurus. The proposed
method's stages are depicted in the Fig. 1.
Fig. 1. Research Methodology
A. ISO 25964 vs SKOS
ISO 25964 is a standard for thesaurus
development. Following ISO 25964 ensures that
the thesaurus is designed and developed using
industry-standard practices, making it easier to

Hidayati and Srimoeljanto, Knowledge Management for Electronic-Based Government System 105
share and integrate with other systems. It also
contributes to the accuracy, comprehensiveness,
and ease of use of the thesaurus, which can
improve the effectiveness of information retrieval
and knowledge management.
Meanwhile, the World Wide Web Consortium
(W3C) recommends Simple Knowledge
Organization System (SKOS) as a tool for
creating and managing thesaurus in a web
environment. This chapter explained creating a
thesaurus with SKOS that adhere to the ISO
25964 standard. There are significant differences
between ISO 25964 and SKOS, including the
thesaurus construction and the integrity rules. The
SKOS element category consists of several
categories, each consisting of several
subcategories, as shown in Table 1.
Table 1. SKOS Categories and Subcategories
Categories
Subcategories
Concepts
Concept, Concept Scheme, In
Scheme, Has TopConcept,Top
Concept Of
Labels & Notation
Pref Label, alt Label, hidden Label,
notation
Documentation
Note, change Note, definition,
editorial Note, example, history
Note, scope Note
Semantic Relations
broader, narrower, related, broader
Transitive, narrower Transitive,
semantic relation
Mapping Properties
broad Match, narrow Match, related
Match, close Match, exact Match,
mapping relation
Collections
Collection, ordered Collection,
member, member List
The equivalence between ISO 25964 and
SKOS is discussed in [13]. Some of the mappings
are direct, but some correspondences are more
challenging to establish. There are differences in
thesaurus construction in ISO 25964 and SKOS,
where thesaurus construction based on ISO 25964
does not directly apply to SKOS construction
[6][14]. Following are some of these differences:
Concept Groups: In the construction of
SKOS, concept groups can be applied using a
concept scheme or collection. However, they
cannot represent the whole concept group as in
ISO 25964.
Concept Group Nesting: As of ISO 25964,
the hasSubGroup/hasSuperGroup relationship
cannot be applied to SKOS. The solution is found
in ISO 25964 SKOS extension (ISO-THES),
where a new property is created to add a
subgroup/supergroup.
Top Concept: Different representations of
concept groups affect the top concept. Thesauri
use concept schemes to represent micro thesauri.
However, the relation cannot be used if
collections are used as in ISO-THES to represent
concepts. For this reason, information about top
concepts often cannot be shared explicitly and
depends on the interpretation of individual users
and creators.
Effects of different types of NT/BT
relationships: The representation of the diversity
of Narrower Term/Broader Term (NT/BT)
relationships does not have a definite solution on
SKOS, except for introducing transitivity.
Transitivity can be used if the existing
relationship is generic (kind of), not mixed (part
of).
According to ISO 25964, “When the scope of
one concept completely overlaps the scope of the
other, a hierarchical relationship is established. It
must be based on degrees or levels of
superordination and subordination, with the
superordinate concept representing the class or
whole and the subordinate concept referring to its
members or parts. The following tags should be
used in reverse order: The broader term, BT, is
written as a prefix to the term superordinate; the
narrower term, NT, is written as a prefix to the
term subordinate”. [15]. The reciprocal
relationship required in ISO 25964 does not
automatically happen.
B. Determine SPBE-Related Term
In this study, the information to be recorded is
contained within the scope of the SPBE
infrastructure. The related concepts are
determined using the SPBE regulatory documents
and the implementation of the SPBE evaluation
on the Abacus application, which is a Ministry of
State Apparatus Utilization, and Bureaucratic
Reform used to carry out the SPBE Architecture
preparation. Three experts who assisted in
implementing the SPBE architecture in some
organizations over the last three years carried out
concept extraction and hierarchy as the basis for
making this semantic thesaurus.
The terms that will constitute the top concept
and its derived ideas and hierarchical relationships
must be determined. The core concept grouping is
accomplished by developing a concept Scheme
called “Arsitektur SPBE” that contains all
concepts related to SPBE architecture. The
“Arsitektur SPBE” concept scheme includes six
top concepts: “Proses Bisnis” (Business
Processes), “Layanan” (Services), “Data dan
Informasi” (Data and Information), “Keamanan”
(Security), “Aplikasi” (Applications), and
“Infrastruktur” (Infrastructure). The six top
concepts are “themes”, “domains”, or “subject
categories”, which include the narrower concepts

106 Jurnal Ilmu Komputer dan Informasi (Journal of Computer Science and Information), volume 16,
issue 2, June 2023
of each of these six top concepts. Meanwhile, the
implementation of the collection, which is a
compilation representation in the ISO 25964
thesaurus terminology on “SPBE Thesaurus” by
grouping several concepts under “Layanan
Publik” (Public Service/ Public Administration)”
as a top concept into Collection “G2C
(Government to Citizen)” or “G2B (Government
to Business) as shown in Fig. 2.
Fig. 2. Architecture SPBE Hierarchy Concept
C. Making Thesaurus
The two SPBE architecture domains of
layanan (services) and infrastruktur
(infrastructure) will receive the most attention in
this study. We create thesaurus using VocBench,
an open-source web-based collaborative platform
developed and maintained by the Semantic
Technology Laboratory of the National Research
Council of Italy (CNR). It integrates easily with
other semantic web tools and applications and
supports several standard formats such as SKOS,
RDF, and OWL. VocBench displays the domain’s
hierarchical concepts tree structure and the
resource view of a certain concept. As its derived
notion, each domain is provided with an initiating
term. These ideas will serve as the foundation for
the extensional search for similar terms. The term
initiation as a concept is then applied to SKOS.
D. Connected to Existing Thesauri
There is the possibility of describing a process
for analyzing the features of another current
thesaurus, like previous studies that manage the
knowledge organization systems used in cultural
heritage collections published as Linked Data,
focusing on the quality of SKOS-based models
[16] . The existing thesauri for our research with a
similar scope and related terms are listed below.
1) AGROVOC
FAO publishes AGROVOC, a thesaurus
containing food, nutrition, and agriculture
information. It is organized hierarchically into 25
concepts, such as "activity," "entity," and
"location".
2) UNESCO
The thesaurus of the United Nations
Educational, Scientific, and Cultural Organization
(UNESCO) is available in SKOS Core and
SKOS-XL versions, containing 4,417 concepts
organized into seven domains and 88
microthesauri. It is downloaded as a SKOS
dataset following Linked Open Data (LOD)
principles and complies with ISO 25964.
3) ELSST
The thesaurus is owned and operated by the
Consortium of European Social Science Data
Archives (CESSDA) and is used for data
discovery across Europe. It supports Danish,

Hidayati and Srimoeljanto, Knowledge Management for Electronic-Based Government System 107
Dutch, Czech, English, Finnish, French, German,
Greek, Hungarian, Icelandic, Lithuanian,
Norwegian, Romanian, Slovenian, Spanish, and
Swedish. It covers over 3,300 concepts in politics,
sociology, economics, education, law, crime,
demography, health, employment, information,
communication technology, and environmental
science.
4) LCSH
The Library of Congress Subject Headings
(LCSH) is a monolingual subject title thesaurus
maintained by the US Library of Congress. It
contains 30,000 concepts and is available as data
linked to a database of data. Cross-references are
used with headers for users of new terms, such as
titles for available terms and related topics.
Fig. 3. Concepts Mapping Diagram between Different Concept
Schemes
Fig. 4. Metadata Registry for SPBE-AGROVOC Concepts Mapping
Fig. 5. Namespace and Import for SPBE-ELSST Concepts Mapping
The VocBench tool maps or aligns the SPBE
thesaurus with existing thesauri. This mapping
process is shown in Fig 3, which shows the
mapping of two concepts from two different
concept schemes. The Metadata Registry View
can create a catalogue of known remote datasets
using metadata such as DCAT, VoID/LIME,
Friend of a Friend (FOAF), and Dublin Core
Metadata Initiative (DCMI) Metadata Terms.
AGROVOC is one of the cloud's remote datasets.
Open datasets because AGROVOC is available in
Dataset Catalogs of Linked Open Data (LOD)
1
.
For example, we aimed to search the
AGROVOC thesaurus by utilizing its
Internationalized Resource Identifier (IRI)
2
in the
VocBench tool. A successful search generated one
AGROVOC Catalog Record, confirming the
addition of AGROVOC metadata, as depicted in
Fig. 4. The researchers aligned external resources
1
https://lod-cloud.net
2
http://aims.fao.org/aos/agrovoc

108 Jurnal Ilmu Komputer dan Informasi (Journal of Computer Science and Information), volume 16,
issue 2, June 2023
by mapping concepts from their SPBE thesaurus
to those in AGROVOC. During the mapping
process, they selected one of the five mapping
properties: skos:exactMatch, skos:closeMatch,
skos:broadMatch, skos:narrowMatch, and
skos:relatedMatch. In addition, they used the
Namespaces and Imports feature of VocBench to
import a concept scheme from the European
Language Social Science Thesaurus (ELSST-
CESSDA)
3
in the turtle format and URI. Adding
the namespace as a new reference, a new concept
scheme named "ELSST Thesaurus" appeared on
the Scheme tab in the VocBench tool, as shown in
Fig. 5, enabling the researchers to carry out the
concept mapping process between SPBE
Thesaurus and ELSST Thesaurus.
E. Obtain a new term by using Word2Vec and
Doc2Vec.
Concepts were gathered from various SPBE
infrastructure documents to identify potential new
terms to add to the thesaurus and then compared
to seed words that were defined as similar or
related terms, as shown in Fig. 2. This research
focuses on identifying potential new terms for the
thesaurus by utilizing seed words from the service
and infrastructure domain. The expert then
manually compiles and verifies these seed words.
The SPBE thesaurus, which already contains
seed words, will be used to generate keywords for
articles or data obtained from Wikipedia scrapped
by BeautifulSoup python library. Wikipedia is a
collection of dispersed and constantly updated
articles. We tried several keywords as filters when
retrieving text data on Wikipedia for Indonesian
language articles, such as layanan public (public
services), layanan pemerintahan (government
services), layanan (services), ilmu pengetahuan
dan teknologi (science and technology),
infrastruktur (infrastructure), and pemerintahan
(government), due to the difficulty in filtering
related terms. Meanwhile, articles in English
contain public administration, infrastructure,
government, services, science and technology, and
infrastructure.
The total number of articles gathered is 100.
We prioritize Indonesian articles because this
thesaurus still uses Indonesian terms, with only a
few terms translated into English for easy
mapping into existing thesauruses. Seed words in
the service and infrastructure domain will be
checked for synonyms and related words. The
search process for this SPBE thesaurus concept
can be carried out using a two word-embedding
3
https://elsst.cessda.eu/id/
method as AI helps to automate extraction terms,
Word2vec and Doc2vec. The Word2Vec and
Doc2Vec models will be trained with data from
100 Wikipedia articles and added with several
SPBE documents on the service and infrastructure
domain.
Word2Vec is an approach to learning high-
quality word embeddings based on neural
networks [17]. It is trained on a large corpus of
text to learn word distributional patterns
represented by the similarity of two vectors. The
algorithm has two architectures: skip-gram and
continuous bag-of-words (CBOW). The skip-
gram architecture predicts context words when
given a target word, whereas the CBOW
architecture predicts target words when given a
context [18]. To automatically expand thesauri,
the algorithm uses the most similar() method to
find similar words in the corpus and add them as
synonyms to the thesaurus. The model was fine-
tuned for a specific task or domain to improve
performance.
The Doc2Vec algorithm is a Word2Vec
extension that learns vector representations of
documents, also known as document
embeddings[19]. It represents each document in
the corpus as a fixed-length vector that captures
its contextual and semantic relationships with
other documents. The distributed memory (DM)
or distributed bag-of-words (DBOW)
architectures can be used to train the algorithm
[18]. The DM architecture predicts the next word
in a document based on its context, whereas the
DBOW architecture predicts a word's presence in
a document based on its context. To automatically
expand the thesauri using Doc2Vec, the algorithm
searches the corpus for similar documents and
adds them as related topics to the thesaurus. The
algorithm accomplishes this by employing the
most similar() method, which returns a list of
similar documents based on the cosine similarity
of their document vectors.
In this study, we train Word2vec with the
following parameters were used:
 Train epochs: 6
 Vector size: 200
 Minimum count: 3
 Algorithm: Skip-gram
 Negative sampling: 5
 Window size: 5
 Downsampling: 1e-3
The train epochs parameter specifies how
many iterations the model should perform on the
training data. The vector size parameter
determines the size of the vector representations
learned by Word2vec. The minimum count
parameter specifies how many times a word must

Hidayati and Srimoeljanto, Knowledge Management for Electronic-Based Government System 109
appear in the corpus before it is included in the
vocabulary. Infrequent terms are typically
excluded because they do not contribute
significantly to the model's performance.
Negative sampling is a technique for increasing
training efficiency while lowering computational
costs. Instead of updating the weights of all words
in the vocabulary for every training sample,
negative sampling randomly selects a small
number of words as negative examples and only
updates their weights.
The number of negative samples is a
parameter that determines how many negative
examples are used for each training sample. The
number of negative samples was set to 5 in this
case. The downsampling parameter specifies how
frequently words in the corpus are randomly
downsampled during training. This parameter
helps to prevent overfitting and improves model
performance. The downsampling parameter was
set to 1e-3 in this case. Doc2vec uses the same
train epochs, vector size, minimum count,
negative sampling, window size, and
downsampling parameters as Word2vec. The only
difference is that the distributed memory (DM)
algorithm is used. The DM algorithm is a CBOW
variant that can learn vector representations of
documents and words. It predicts the next word
based on the preceding terms and the document
vector. The document vector is updated during
training to capture the contextual and semantic
relationships between documents.
4. Result and Discussion
A. Mapping of Terms Between Thesauri
We could not locate an Indonesian thesaurus
with the same scope, but as stated previously, this
SPBE thesaurus will be linked to several English
thesauri with virtually the same scope. Because we
are constructing our thesaurus exclusively in
Bahasa Indonesia, we translate the terms into
English; for instance, we translate layanan publik
into public administration so that the terms public
and administration can capture similar words or
terms.
A team of experts was involved in this
research, and we performed the translation
procedure manually to ensure that the context was
maintained. The term public administration, which
resembles public service on ELSST and
AGROVOC, is one example of a concept map, as
shown in Fig. 6. As depicted in Fig. 7, the term
ilmu pengetahuan dan teknologi on the ELSST is
comparable to science and technology. The
mapping process is performed on a term-by-term
basis, with each term first being translated into
English for recognition purposes, thereby
rendering the SPBE thesaurus bilingual. This
procedure necessitates vigilance and discernment
of the same scope so that each term has the same
meaning. The collected terms are then arranged
according to the hierarchical structure created in
Fig. 2, resulting in the initial thesaurus structure
depicted in Fig. 8.
Fig. 6. Public Service Concept Relation
Fig. 7. Technology Mapping Term

110 Jurnal Ilmu Komputer dan Informasi (Journal of Computer Science and Information), volume 16,
issue 2, June 2023
Fig. 8 Hierarchical Concepts Tree Structure of the Domains and Resource View of a Certain Concept
B. Word2Vec and Doc2Vec Related Term
Generated
Some examples of words generated
based on the Word2Vec and Doc2Vec models
with several examples of keywords such as
services, public, science and technology,
infrastructure, and government can be seen
in Table 2. We investigate the top 5 similarity
scores of terms generated by Word2Vec and
Doc2Vec algorithms, like the previous study that
used Doc2Vec only [10]. Both algorithms are
commonly used for natural language processing
and are used to analyze text data. Word2Vec is a
shallow neural network-based model that focuses
on word embeddings. At the same time, Doc2Vec
is an extension of Word2Vec that can handle
variable-length texts such as sentences,
paragraphs, or documents by creating document-
level embeddings. Looking at the table, we can
observe that both algorithms generate similar
terms, but the similarity score varies.
For example, Word2Vec generates the
term pelayan (waiter) with a similarity score of
0.80, while Doc2Vec generates pelayanan
(service) with a similarity score of 0.77. Similarly,
Word2Vec generates the term komputer
(computer) with a similarity score of 0.80, while
Doc2Vec generates inovasi with a similarity score
of 0.81. One key difference between the two
algorithms is the level of analysis. Word2Vec
focuses on word-level embeddings, while
Doc2Vec considers document-level embeddings.
As a result, Doc2Vec can better capture the
context of the entire document, while Word2Vec
is better suited for analyzing individual words in
isolation. Another difference is the nature of the
similarity scores.
Word2Vec tends to generate higher similarity
scores than Doc2Vec, as it focuses more on the
relatedness of words, while Doc2Vec considers
both the relatedness and similarity of the
documents. In terms of the generated terms, both
algorithms generate terms that are commonly used
in academic and technical writing. For example,
pemerintah (government), teknologi informasi
(information technology), riset (research), and
kebijakan (policy) are all commonly used in
academic writing. However, there are some
differences in the terms generated, with Word2Vec
generating terms like komputer (computer) and
data mining , which are more technical. In
contrast, Doc2Vec generates terms like publikasi
(publication) and jasa (services), which are more
general and encompass a more comprehensive
range of topics. In summary, the table shows that
both Word2Vec and Doc2Vec algorithms generate
similar terms, but the similarity scores and the
nature of the generated terms differ between the
two algorithms. Word2Vec is better suited for
analyzing individual words in isolation, while
Doc2Vec is better for analyzing documents.
The process of filtering words and terms that
appear has its challenges. Because the scope set is
SPBE, while the articles used as training data are
articles with top concepts as keyword terms, these
terms may have a general meaning and may not be
related to SPBE. Words and terms found from
Word2Vec and Doc2Vec modeling process also
need to identify the relationships among them. The
relationship between generated related terms and
the top concept terms needs to be determined,
whether it is NT, BT, or RT. In the end, the expert
must manually determine the relationship.
It should be noted that ISO 25964 requires a
reciprocal relationship between NT and BT, which
SKOS does not automatically support. When the

Hidayati and Srimoeljanto, Knowledge Management for Electronic-Based Government System 111
top concept has the concept below it, there will
only be a BT relationship between the concepts.
Meanwhile, the NT process from the previous
concept must be manually entered. We must keep
this in mind when adding terms that every time
terms are added, the reciprocity of each related
term must be ensured. Despite their similarities,
words not within the scope of SPBE appear during
the training process for searching for synonyms
and related terms in Indonesian. The term
“layanan”(service) brings up words like
“pelayan”(waiter) and “pelanggan”(customer),
which are not in the SPBE scope. The expert has
eliminated the words like that.
Table 2. Example of Words generated by Word2Vec and Doc2Vec
Layanan (service)
Word2Vec
Doc2Vec
Word
Similarity
Score
Word
Similarity
Score
Pelayan (waiter)
0.80
Pelayanan (service)
0.77
Pelanggan (customer)
0.78
Pelanggan
(customer)
0.75
Fasilitas (facility)
0.76
Konsumen
(consumer)
0.75
Inovasi (innovation)
0.75
Teknis (technical)
0.70
Penyedia (provider)
0.74
Jasa (service)
0.70
Publik (public)
Masyarakat (people)
0.73
Publikasi
(publication)
0.78
Pemerintah
(government)
0.64
Masyarakat (people)
0.77
Layanan (service)
0.63
Umum (public)
0.74
Administrasi
(administration)
0.63
Pengunjung (visitor)
0.74
Birokrasi
(bureaucracy)
0.62
Pemerintah
(government)
0.72
Ilmu Pengetahuan
dan Teknologi
(science and
information
technology)
Komputer (computer)
0.80
Inovasi (Inovation)
0.81
teknologi informasi
(information
technology)
0.74
teknologi informasi
(Information
Technology)
0.72
Sains (science)
0.74
Riset (research)
0.65
Teknologi
(technology)
0.74
Pengembangan
(development)
0.65
data mining
0.72
cybersecurity
0.64
Infrastruktur
(Infrastructure)
Pembangunan
(development)
0.84
Jaringan (network)
0.78
Prasarana
(infrastructure)
0.76
Transportasi
(transportation)
0.67
Proyek (project)
0.75
Komunikasi
(communication)
0.65
Modernisasi
(modernization)
0.74
Teknologi
(technology)
0.65
Penyediaan
(provision)
0.70
Fasilitas (facility)
0.63
Pemerintah
(government)
Birokrasi
(bureaucracy)
0.76
Lembaga
(organization)
0.78
Kebijakan (policy)
0.69
Negara (country)
0.72
Daerah (regional)
0.69
Kebijakan (policy)
0.69
Pusat (central)
0.68
Politik (political)
0.65
Peraturan (regulation)
0.66
organisasi
0.64

112 Jurnal Ilmu Komputer dan Informasi (Journal of Computer Science and Information), volume 16,
issue 2, June 2023
C. New Terms Obtained as an Expansion of Seed
Words
Further investigation revealed that the
algorithms generated 210 unique words with
multiple keyword associations in the services
domain. However, only 75 of these words were
approved after expert review, indicating that the
remaining 135 words were either irrelevant to the
services domain or needed to be clearer to be
considered valid.
Similarly, the algorithms generated 150 unique
words with several keyword associations in the
infrastructure domain, but experts approved only
56. A precision score is a metric to assess a natural
language processing system's ability to identify
relevant information correctly [11]. In this case,
the precision score is the ratio of approved words
to total words generated by the algorithms. As a
result, the precision score in the services domain
would be 75/210 = 0.36, and in the infrastructure
domain, it would be 56/150 = 0.37. The precision
score provides valuable information about how
well the Word2Vec and Doc2Vec algorithms
generate relevant words for a given domain. A low
precision score indicates that the algorithms
generated many irrelevant or ambiguous words. In
contrast, a high precision score suggests that the
algorithms could capture the domain's relevant
language. As a result, the precision score is an
essential factor to consider when evaluating the
performance of natural language processing
systems in domain-specific applications.
This thesaurus will store all keywords related
to the SPBE architecture to find related
information quickly. However, expanding the
SPBE thesaurus was difficult because the
sentences caught were outside the scope of SPBE
despite using keywords related to SPBE. The
process of filtering and relations is still very
manual and requires human power, in this case,
experts and us as researchers, so creating a
thesaurus takes a long time and is prone to errors.
Another external issue is that the keywords in
SPBE sometimes need to be more specific and are
used for purposes other than explaining the SPBE
architecture.
D. SPBE Thesaurus Maintenance Strategy
According to our findings, we propose a
maintenance strategy for an SPBE SKOS
thesaurus. The strategy entails establishing a
system to automatically extract new terms from
online articles using word2vec and doc2vec
models, validating the context of the newly
extracted terms, and integrating the thesaurus with
the website's search engine. The system can be
configured to run periodically to add new terms to
the thesaurus.
Keeping track of changes in the digital
government domain, maintaining consistency and
organization in the thesaurus, regularly reviewing
and updating scope notes and definitions, and
providing clear documentation on how to use the
thesaurus are additional considerations for
ensuring the accuracy and utility of the thesaurus.
This strategy ensures that the thesaurus remains
useful and pertinent for users navigating the
complex terminology of digital government.
5. Conclusion and Future Work
Thesaurus development in the web
environment must adhere to the SKOS structure
and the ISO 25964 standard. In this study, the
SPBE thesaurus’s SKOS construction applied
concept schemes, top concepts, concepts, and the
relationship between NT and BT. In this study, we
look for related terms scattered throughout articles
on the internet. The initial terms in the main
hierarchies are in the SPBE scope and domain and
have been identified as keywords in advance. The
term search process employs Word2Vec and
Doc2Vec, with data training articles containing
Wikipedia keywords. Using this method to add
terms to the thesaurus presents three challenges,
one of which is that the related terms produced by
the Word2vec and Doc2Vec models are outside
the scope of SPBE.
The precision value for each generated word
that can be entered into the thesaurus still needs to
be higher in the focused domains, namely 0.36 for
the service domain and 0.37 for the infrastructure
domain. The relationship between terms, namely
NT, BT, and RT, is then difficult to ascertain.
Finally, due to SKOS limitations, the reciprocal
relationship between NT and BT to follow the
ISO 25964 standard must always be ensured.
In the future, we will concentrate on e-
government, online services, and IT
infrastructure-related content in Wikipedia and
other online articles. Avoid extracting irrelevant
keywords from irrelevant content to maintain
precision. Identify frequently used phrases and
terms in digital government, such as "open data,"
"digital citizenship," and "smart cities," and
employ domain-specific knowledge to improve
keyword extraction. Participation by experts may
necessitate knowledge of digital government
procedures, policies, and technologies, as well as
assuring proper NT and BT relationship patterns
among the terms as concepts.
References
[1] R. Rigin and I. Dua Reja, “Survey Paper tentang
Enterprise Architecture di Sektor Publik,”
KONSTELASI Konvergensi Teknol. dan Sist. Inf.,
vol. 2, no. 1, pp. 56–70, 2022, doi:

Hidayati and Srimoeljanto, Knowledge Management for Electronic-Based Government System 113
10.24002/konstelasi.v2i1.5365.
[2] S. Supriyanto, A. Ridwan, R. Tamam, M. I. Santoso,
D. Satria, and A. I. S. Mutaqin, “Perancangan sistem
pemerintahan berbasis elektronik (SPBE) yang
berkelanjutan di Provinsi Banten,” J. Ind. Serv., vol.
7, no. 1, p. 171, 2021, doi: 10.36055/12952.
[3] I. S. Rozas, K. Khalid, N. Yalina, N. Wahyudi, and
D. Rolliawati, “Digital Enterprise Architecture for
Green SPBE in Indonesia,” CCIT J., vol. 15, no. 1,
pp. 26–42, 2022, doi: 10.33050/ccit.v15i1.1366.
[4] A. Alvarenga, F. Matos, R. Godina, and J. C. O.
Matias, “Digital transformation and knowledge
management in the public sector,” Sustain., vol. 12,
no. 14, 2020, doi: 10.3390/su12145824.
[5] H. Khanchel, “The Impact of Digital Transformation
on Banking,” J. Bus. Adm. Res., vol. 8, no. 2, p. 20,
2019, doi: 10.5430/jbar.v8n2p20.
[6] M. M. Martínez-González and M. L. Alvite-Díez,
“Thesauri and Semantic Web: Discussion of the
Evolution of Thesauri Toward Their Integration with
the Semantic Web,” IEEE Access, vol. 7, no.
October, pp. 153151–153170, 2019, doi:
10.1109/ACCESS.2019.2948028.
[7] K. Nataliia, “Application of the ‘ Smart City ’ data
domain thesaurus as the tool for representing
knowledge while improving the problem-oriented
Web search effectiveness,” pp. 17–20, 2019.
[8] K. Nakayama, T. Hara, and S. Nishio, “A Thesaurus
Construction Method from Large Scale Web
Dictionaries,” no. Aina, 2007.
[9] P. Arnold, P. Arnold, and E. Rahm, “Extracting
Semantic Concept Relations from Wikipedia
Extracting Semantic Concept Relations from
Wikipedia,” no. January 2016, 2014, doi:
10.1145/2611040.2611079.
[10] D. A. Koutsomitropoulos, “Subject Classification of
Learning Resources Using Word Embeddings and
Semantic Thesauri,” pp. 3–8, 2019.
[11] C. Lanza and A. Hazem, “Towards Automatic
Thesaurus Construction and Enrichment,” no. May,
pp. 62–71, 2020.
[12] N. Lagutina, I. Paramonov, I. Vorontsova, and N.
Kasatkina, “An Approach to Automated Thesaurus
Construction Using Clusterization-Based Dictionary
Analysis.”
[13] I. S. O. Tc, T. A. Isaac, and S. Recommendations,
“Correspondence between ISO 25964 and SKOS /
SKOS ‐ XL Models Contributors,” pp. 1–13, 2012.
[14] M. M. Martínez-gonzález, M. Alvite-díez, T. I. T.
Edificio, and C. M. Delibes, “Computer Standards &
Interfaces The support of constructs in thesaurus
tools from a Semantic Web perspective : Framework
to assess standard conformance,” Comput. Stand.
Interfaces, vol. 65, no. July 2018, pp. 79–91, 2019,
doi: 10.1016/j.csi.2019.02.003.
[15] ISO 25964-1:2011(E), information and
documentation — Thesauri and interoperability with
other vocabularies — Part 1:Thesauri for
information retrieval, First edition, 2011-08-15
[16] D. Díaz-corona, J. Lacasta, M. Á. Latre, F. J.
Zarazaga-soria, and J. Nogueras-iso, “Profiling of
knowledge organization systems for the annotation
of Linked Data cultural resources,” vol. 84, pp. 17–
28, 2019, doi: 10.1016/j.is.2019.04.008.
[17] T. Mikolov, K. Chen, G. Corrado, and J. Dean,
“Efficient estimation of word representations in
vector space,” 1st Int. Conf. Learn. Represent. ICLR
2013 - Work. Track Proc., pp. 1–12, 2013.
[18] S. Martinčić-Ipšić, T. Miličić, and L. Todorovski,
“The influence of feature representation of text on
the performance of document classification,” Appl.
Sci., vol. 9, no. 4, 2019, doi: 10.3390/app9040743.
[19] Q. Le and T. Mikolov, “Distributed representations
of sentences and documents,” 31st Int. Conf. Mach.
Learn. ICML 2014, vol. 4, pp. 2931–2939, 2014.