Keyword Extraction for Search Engine Optimization Using Latent Semantic Analysis

Abstract

It is now difficult to access desired information in the Internet world. Search engines are always trying to overcome this difficulty. However, web pages that cannot reach their target audience in search engines cannot become popular. For this reason, search engine optimization is done to increase the visibility in search engines. In this process, a few keywords are selected from the textual content added to the web page. A responsible person who is knowledgeable about the content and search engine optimization is required to determine these words. Otherwise, an effective optimization study cannot be obtained. In this study, the keyword extraction from textual data with latent semantic analysis technique was performed. The latent semantic analysis technique models the relations between documents/sentences and terms in the text using linear algebra. According to the similarity values of the terms in the resulting vector space, the words that best represent the text are listed. This allows people without knowledge of the SEO process and content to add content that complies with the SEO criteria. Thus, with this method, both financial expenses are reduced and the opportunity to reach the target audience of web pages is provided.

Full text

POLİTEKNİK DERGİSİ
JOURNAL of POLYTECHNIC
ISSN: 1302-0900 (PRINT), ISSN: 2147-9429 (ONLINE)
URL: http://dergipark.org.tr/politeknik
Keyword extraction for search engine
optimization using latent semantic analysis
Gizli anlamsal analiz ile arama motorları için
anahtar kelime çıkarma
Yazar(lar) (Author(s)): Fahrettin HORASAN
ORCID: 0000-0003-4554-9083
Bu makaleye şu şekilde atıfta bulunabilirsiniz(To cite to this article): Horasan F., “Keyword extraction
for search engine optimization using latent semantic analysis”, Politeknik Dergisi, 24(2): 473-479, (2021).
Erişim linki (To link to this article): http://dergipark.org.tr/politeknik/archive
DOI: 10.2339/politeknik.684377

Keyword Extraction for Search Engine Optimization Using Latent
Semantic Analysis
Highlights
❖ SEO processes were performed automatically
❖ The study has been tested with a well-known data set.
❖ It has become easier to create content suitable for SEO
Graphical Abstract
The following figure shows the process of obtaining keywords from textual data used as web content, in stages.
Figure. 1
Aim
In this study, the keywords that best represent the content were obtained using the LSA in order to ensure that the
content on the web pages comply with the SEO.
Design & Methodology
Latent semantic analysis known as vector space based is used in this study.
Originality
Keyword lists used in SEO were examined according to different dimension reduction parameters.
Findings
Better results were obtained with the method based on the proposed term and document similarity.
Conclusion
Keyword lists that best represent textual web content were obtained by considering their similarity with other
terms/sentences.
Declaration of Ethical Standards
The author of this article declare that the materials and methods used in this study do not require ethical committee
permission and/or legal-special permission.

Politeknik Dergisi, 2021; 24(2) : 473-479 Journal of Polytechnic, 2021; 24 (2): 473-479
473
Keyword Extraction for Search Engine Optimization
Using Latent Semantic Analysis
Araştırma Makalesi / Research Article
Fahrettin HORASAN*
Engineering Faculty, Computer Engineering Department, Kırıkkale University, Kırıkkale, Turkey
(Geliş/Received : 04.02.2020 ; Kabul/Accepted : 11.04.2020)
ABSTRACT
It is now difficult to access desired information in the Internet world. Search engines are always trying to overcome this difficulty.
However, web pages that cannot reach their target audience in search engines cannot become popular. For this reason, search engine
optimization is done to increase the visibility in search engines. In this process, a few keywords are selected from the textual content
added to the web page. A responsible person who is knowledgeable about the content and search engine optimization is required
to determine these words. Otherwise, an effective optimization study cannot be obtained. In this study, the keyword extraction from
textual data with latent semantic analysis technique was performed. The latent semantic analysis technique models the relations
between documents/sentences and terms in the text using linear algebra. According to the similarity values of the terms in the
resulting vector space, the words that best represent the text are listed. This allows people without knowledge of the SEO process
and content to add content that complies with the SEO criteria. Thus, with this method, both financial expenses are reduced and
the opportunity to reach the target audience of web pages is provided.
Keywords: Search engine optimization, keyword extraction, latent semantic analysis, text mining.
Gizli Anlamsal Analiz ile Arama Motorları için
Anahtar Kelime Çıkarma
ÖZ
Devasa bilgi yığınının bulunduğu internet dünyasında artık istenilen bilgiye erişmek zor hale geldi. Arama motorları bu zorluğun
altından kalkmak için çaba sarf etmektedirler. Ancak arama motorlarında hedef kitlesine ulaşamayan bir web sayfası popüler hale
gelememektedir. Arama motorlarındaki görünürlüğün artırılması için arama motoru optimizasyonu yapılır. Bu süreçte web
sayfasına eklenen metinsel içeriklerden anahtar kelimeler seçilir. Bu kelimelerin belirlenmesi için hem içerik hakkında ve hem de
arama motoru optimizasyonu konusunda bilgili bir sorumlu kişi gereklidir. Böyle olmadığı durumlarda etkili bir optimizasyon
çalışması elde edilemez. Bu çalışmada gizli anlamsal analiz tekniği ile metinsel verilerden anahtar kelime çıkarma işlemi
gerçekleştirilmiştir. Gizli anlamsal analiz yöntemi, metin içerisindeki doküman/cümle ve terimler arasındaki ilişkileri lineer cebir
yönüyle modellemektedir. Elde edilen vektör uzayındaki terimlerin benzerlik değerlerine göre metini en iyi temsil edilen kelimeler
listelenir. Bu işlem SEO süreci ve içerik hakkında bilgisi olmayan insanların da SEO kriterlerine uygun içerik eklemelerine imkan
tanıyacaktır. Dolayısıyla, bu yöntemle hem maddi gider azaltılmış hem de web sayfalarının hedef kitlesine ulaşma fırsatı
sağlanmıştır.
Anahtar Kelimeler: Arama motoru optimizasyonu, anahtar kelime çıkarımı, gizli anlamsal analiz, metin madenciliği.
1. INTRODUCTION
The increase in the usage of the internet and the
facilitation of the publishing of web sites lead rapid
increase in the number of web pages published on the
internet [1]. Besides, the most important purpose of
publishing a website is to ensure that the data is delivered
to target users on time. Most of the time ,under standard
circumstances, it is time consuming for users to access
the information they want on the internet since it requires
the users to know all the content on the web and where
the specific content is located. This issue is largely
overcome by web sites called web search engines.
According to World Internet Usage and Population
Statistics data, 58.8 percent of the world's population use
internet technologies in mid-2019 [2]. Figure 1 shows the
increase in the number of hostnames and active websites
in the World. Difficulties are encountered in accessing
correct information in the internet environment where
there are too many internet users and websites [3]. The
search engines are an information-navigation-tool that
enables users to access the information they want on the
internet [4]. On the other hand, in parallel with the
development of web technologies, the importance of the
search engines is increasing day by day [5].
The search engines scan contents of the web pages at
certain time intervals via a software called crawlers or
spiders and add the necessary information about the
contents of the scanned web pages to their databases.
Thus, they make it possible for internet users to access
the desired content or web pages thru themselves [6]. In
order for the crawlers to find and scan the web pages, the
*Sorumlu Yazar (Corresponding Author)
e-posta : fhorasan@kku.edu.tr

Fahrettin HORASAN / POLİTEKNİK DERGİSİ, Politeknik Dergisi,2021;24(2): 473-479
474
URL details of the web sites need to be recorded in the
database of the search engines [6]. For this reason, it is
important for web developers to implement this addition
process and regularly update the site contents in order for
the websites to get more visits [6,7]. The whole set of
operations performed to ensure that websites are indexed
in the search engines in the best way is called Search
engine optimization (SEO)[8,9].
Figure 1. The increase in the number of hostnames and active websites in the World [2]
During the SEO process, some additional edits are made
in the required places of the web sites. These edits may
look like simple operations, but they have a significant
impact on the site's search results. They may improve
performance, but it should also be taken into account that
they may reduce performance so, the edits need to be
done carefully [6,7]. The web developers need to do a
separate SEO work for each site since content and
content presentation are different on each site. Moreover,
the person performing the SEO process must be
knowledgeable about the content posted on the web page,
otherwise, the SEO process may cause the website to
become a site that does not appeal to the target audience
[7].
The SEO operations can be facilitated by designing the
necessary processes automatically. There are many
studies in this subject. First of all, the SEO operations
should be carried out for the visitors, not the website
publisher. The SEO directing users to the correct web
pages are called White hat SEO [10]. Therefore, the main
principle in SEO studies is to be user oriented. There are
many studies in the literature that automatically perform
the SEO process as user-oriented or facilitate this
process. For example, a content management system that
performs white hat SEO has been developed for an active
web page called Fragfornet [11]. This system, in which
content is added and managed on the web page, realizes
SEO automatically. One of the platforms that attract
attention in the internet world is electronic market sites.
A product-content optimization was developed in a study
for the electronic market site. The basis of this study is
multi-criteria optimization model such as discount time,
visual presentation and product relations [12]. In another
study, researchers have done a new heuristic scanning
process with additional learning techniques that learn by
looking at the data that affects their web site's ranking on
the search engine. They have proved that a system that
automatically combines intuitive scans based on the data
coming from users makes a system that gets better rank
in search engines [13].
The search engines make recommendations according to
the query sentence consisting of keywords or keyphrases
[14]. Studies under the name of keyword extraction
[15,16,17], query suggestion [18] and query
classification [19] are based on this.
Latent semantic analysis (LSA) can be useful in finding
keywords or keyphrases that best represent the semantic
structure in the text [20,21]. In this study, the keywords
used in the seo process of the contents added to the web
pages were determined with the LSA model. The LSA,
which is used in many fields such as text mining, image
processing, data mining, signal processing, voice
analysis, is a dimension reduction based approach
[21,22,23]. The tests performed according to the
parameters in the technique called rank-k in the
dimension reduction stage were examined. Thus, a more
efficient and faster recommendation process was
obtained. The contributions of the developed model can
be listed as follows.
• SEO processes are performed automatically for each
content added to the web pages.
• Someone who has nothing to do with the content or
seo stages can also add content that complies with
the SEO criteria.
• The study has been tried with a well-known data set.
It is comparable to future studies

KEYWORD EXTRACTION FOR SEARCH ENGINE OPTIMIZATION USING LATENT SEMAN… Politeknik Dergisi, 2021; 24 (2) : 473-479
475
• It was examined according to different dimension
reduction parameters.
• The study can shed light on studies such as topic
extraction, query classification.
In the next part of the paper, the realization of the seo
process with the latent semantic analysis is explained. In
Section 3, experiments and test parameters are explained.
The last part is the discussion and conclusion section.
2. SEO VIA LATENT SEMANTIC ANALYSIS
The Latent semantic analysis is a statistical /
mathematical technique that reveals latent relations
between term-term, term-document and document-
document [22,23]. The LSA, which is a dimension
reduction based approach, aims to consider only the
important data groups in the dataset. Data included in the
data stack that does not contribute to the meaning or
negatively affect the meaning are not included in this
process. For this, the low-rank approach of the term-
document matrix is used to find the latent semantic
structure between the term and the documents [23].
Terms and documents are represented by elements of the
row and column of the matrix, respectively. This matrix
is called as term-document matrix. i-th row and j-th
column of the term-document matrix contains the
mathematical value of i-th term in j-th document. This
mathematical value is known as the weight. The
calculation of this mathematical value is known as
weighting. The weight of the document for each term is
calculated according to some methods [21,23].
Usually, the textual dataset is processed in the LSA. By
passing through this dataset through the pre-process, stop
words and punctuation marks are removed. Stemming is
applied for each term. Then the weights in the term and
document matrix are obtained. In this study, the most
used TF * IDF method was chosen among the weighting
methods [21,22,24]. Thus, the term-document matrix ( )
is obtained by using the weight of the term in each
document. The SVD is applied to the obtained matrix.
The rank k approach is applied to the matrices obtained
as a result of the matrix decomposition in order to reduce
the dimension. After applying the Rank k approach, the
term matrix and the document matrix are obtained by
multiplying the left and right orthogonal matrices by the
singular value matrix, respectively [25,26]. Each row in
the term matrix represents the vector of the same indexed
term in the term-document matrix. Each column in the
document matrix represents the vector of the same
indexed document in the term-document matrix. Thus,
the term and document vectors are represented in the
same vector space. After obtaining the vector space,
documents/terms are listed according to the query from
most similar to least similar. Ultimately, the documents /
terms associated with the query (This can be a document,
term or sentence) are discovered [23,27,28].
In this study, a single text is considered as a data stack
and each sentence in this text is used as a document. The
term document (sentence) matrix is obtained for terms in
each sentence. As mentioned in the previous paragraph,
the SVD is applied to this obtained matrix. The vectors
of terms and documents are determined by the value of k
in the rank k approach. The terms are listed according to
their proximity to each other, taking into account their
similarities to all terms and phrases mentioned in the text.
Thus, word lists that have a very similar resemblance to
terms and documents can be accessed in this way. These
are the words that have the most discrimination in the text
and can represent the text. Figure 2 shows the flow chart
of the study.
Figure 2. Flow chart of keyword extraction technique with LSA
2.1. Singular Value Decomposition
The SVD of the
mn
A

is found by the formula
.
0T
A U V


=

(1)
Here m>n,
TT
m
U U UU I==
and
TT
n
V V VV I==
. In addition,

matrix of diagonal
and containing singular values of A is in the format
1 2 1 0
k k n
    
+
     =  = =
. (2)
2.2. Rank-k Approach
The rank k approach is applied to reduce the cost of
calculation and increase efficiency in Formula 2. When
the rank of A matrix is k
1 2 1k k n
     
+
        =
. (3)

Fahrettin HORASAN / POLİTEKNİK DERGİSİ, Politeknik Dergisi,2021;24(2): 473-479
476
Here,

represents the threshold value. In order for

to be the most optimal, the difference between
k

and
1k

+
should be significantly higher.
If the rank approach of the matrix A(
k
A
) is used instead
of the matrix A in LSA, Ak is represented by the equation
T
k k k k
A U V=
. (4)
k
U
and
T
k
V
in the formula represent the first k columns
of
U
and
T
V
matrices, respectively.
k

is the diagonal
matrix
1 2 3
( , , , , )
kk
diag
   
 = 
.
2.3. Obtaining Vector Space
Representatives of terms and documents in vector space
are obtained as
k
T
ve
k
D
, respectively, with equations
k k k
TU=
(5)
T
k k k
DV=
(6)
Herein, the i'th row of the
k
T
matrix is the vector
symbolizing the term i'th, and the j'th column of the
k
D
matrix is the vector symbolizing the j'th document.
2.4. Listing Words
At this stage, words are listed in three ways. According
to the first, only other terms are taken into account when
calculating the similarity of the terms. According to the
similarity value, the terms are listed from the least value
to the most value. This listing method is Term Similarity
Based listing (TSBL). The second is that the words are
listed according to the similarity of the documents. The
name of this method is Document Similarity Based
Listing (DSBL). Another method of listing is the method
in which the similarity of terms and documents are
calculated together. The name of this method is Term and
Document Similarity Based Listing (TDSBL).
In calculating each of these methods, the cosine
similarity technique was used. The cosine similarity
technique was preferred because it makes an angular
similarity measurement. The cosine similarity technique
takes into account the cosine value of the angles at which
two vectors intersect with each other in the same vector
space [21,24]. The similarity between the two vectors is
calculated by the formula
1
22
11
Cos_ ( , )
n
ii
i
nn
ii
ii
AB
XY
Similarity X Y XY AB
=
==

==


. (7)
In the formula,
X
and
Y
represent
1m
dimensional
vectors.
2.5. Choosing Words and Evaluating
It is the selection of the best
N
of the words listed in the
previous stage.
N
keywords are determined by selecting
the best of the terms listed in order from least similar
to most similar. The keyword lists obtained were
examined according to the TSBL, DSBL and TDSBL
techniques. In addition, according to rank k approach,
their performances according to different
k
values were
examined. As an example, Table 1 shows the 20 most
similar word lists according to the different rank-
values of a text according to the TBDL technique.
3. EXPERIMENTAL ANALYSIS
In this study, BBC news collection was used as a data set.
In this collection, there are 1313 documents and 15393
words under 5 classes in total.
The performances of the keyword lists obtained
according to the TSBL, DSBL and TDSBL techniques
were examined according to the rank k approach. In the
word groups listed according to two different
k
values,
the number of similar words (
n
) and the similarity ratio
(
sr
) were examined. sr is calculated according to the
equation
The number of similar terms
N
sr =
. (8)
Here, N is the number of the best N terms.
The algorithm complexity of the keyword extraction
technique using the
mn
dimensional term document
matrix is
2
O( ).mn
In this study, where the Rank
k
approach is used, the algorithm complexity is
O( ).mnk
It is also considered to be
( )
k<min m,n
. Thus, a less
costly system was developed.
During the analysis process, keyword extraction was
performed for all documents in the dataset. Firstly,
figures 3.a., figures 4.a. and figures 5.a., which show
N
k
Table 1. Words listed according to different k values
k
Ordered Word List
20
boost, profit, timewarn, year, earlier, high, speed, internet, aol, revenu, catwoman, contrast, third,
final, lord, ring, trilog, full, chief, execut
15
boost, profit, timewarn, year, earlier, internet, aol, revenu, help, box, offic, alexand, catwoman,
sharp, contrast, final, ring, trilog, full, post
Similar Words
boost, profit, timewarn, year, earlier, internet, aol, revenu, catwoman, contrast, final, ring, trilogy,
full

KEYWORD EXTRACTION FOR SEARCH ENGINE OPTIMIZATION USING LATENT SEMAN… Politeknik Dergisi, 2021; 24 (2) : 473-479
477
similarity changes according to the TSBL, DSBL and
TDSBL techniques mentioned in Section 2.4, should be
examined. In these figures, the terms are sorted according
to their similarity values to all documents. As can be
seen, there is an increasing similarity change. In figure
3b, figures 4b, and 5b, which show the performances of
the 20 words that resemble the best, the similarity change
of the term groups that can represent the document better
is seen.
a. All of the listed terms b. Top 20 of the listed terms
Figure 3. Similarity changes according to the TSBL technique
a. All of the listed terms b. Top 20 of the listed terms
Figure 4. Similarity changes according to the DSBL technique
a. All of the listed terms b. Top 20 of the listed terms
Figure 5. Similarity changes according to the TDSBL technique

Fahrettin HORASAN / POLİTEKNİK DERGİSİ, Politeknik Dergisi,2021;24(2): 473-479
478
In Table 2, Table 3, and Table 4, the similarities of the
keywords according to the TSBL, DSBL and TDSBL
techniques were examined according to different rank k
values. In the TSBL technique, the best result was
observed when k 15 and 20. According to the DSBL
technique, good results are observed when k is in the
range of 20-25. In the TDSBL technique, which uses both
techniques, good results were obtained when k value is
between 15 to 20.
4. CONCLUSION
Firms/ people have to work with Search Engine
Optimization consultants or companies for their web
pages. They need this to reach their target audience or
increase their audience in the e-commerce environment.
As a result, SEO process of websites causes both labor
and financial expenses. With this study, in order to
eliminate/ reduce these losses, keyword determination
processes in the seo transaction were performed
automatically. The Latent semantic analysis and keyword
extraction method used in the study will shed light on
future studies, especially in areas such as question
answering, topic detection, and text classifying.
DECLARATION OF ETHICAL STANDARDS
The author(s) of this article declare that the materials and
methods used in this study do not require ethical
committee permission and/or legal-special permission.
AUTHORS’ CONTRIBUTIONS
Fahrettin HORASAN: performed the design and
implementation of the research, analysis of the results
and writing the article.
CONFLICT OF INTEREST
There is no conflict of interest in this study.
Table 2. Performances of the TSBL technique
rank k
rank k
The number of similar terms
sr
rank 2
rank 10
1,3
6,5
rank 2
rank 15
0,5
2,5
rank 2
rank 20
0,3
1,5
rank 2
rank 25
0,1
0,5
rank 10
rank 15
8,1
40,5
rank 10
rank 20
7
35
rank 10
rank 25
8,6
43
rank 15
rank 20
14,2
71
rank 15
rank 25
11
55
rank 20
rank 25
13,1
65,5
Table 3. Performances of the DSBL technique
rank k
rank k
The number of similar terms
sr
rank 2
rank 10
6,1
30,5
rank 2
rank 15
5,4
27
rank 2
rank 20
3,7
18,5
rank 2
rank 25
3,3
16,5
rank 10
rank 15
12,1
60,5
rank 10
rank 20
12,1
60,5
rank 10
rank 25
11
55
rank 15
rank 20
12,2
61
rank 15
rank 25
10,5
52,5
rank 20
rank 25
16
80
Table 4. Performances of the TDSBL technique
rank k
rank k
The number of similar terms
sr
rank 2
rank 10
2
10
rank 2
rank 15
1,1
5,5
rank 2
rank 20
0
0
rank 2
rank 25
1,1
5,5
rank 10
rank 15
10,2
51
rank 10
rank 20
7,3
36,5
rank 10
rank 25
8,4
42
rank 15
rank 20
15,2
76
rank 15
rank 25
14,1
70,5
rank 20
rank 25
16,4
82

KEYWORD EXTRACTION FOR SEARCH ENGINE OPTIMIZATION USING LATENT SEMAN… Politeknik Dergisi, 2021; 24 (2) : 473-479
479
REFERENCES
[1] Leavitt, Neal ,“Network-usage changes push internet
traffic to the edge.”, Computer, 43.10: 13-15 (2010).
[2] Internet World Stats Internet users of the world: World
Internet Usage And Populatıon Statıstıcs, “2019 Mid-
Year Estimates”, www.internetworldstats.com ( 2019).
[3] Wood, Steve, “Web of Deception: Misinformation on the
Internet”. New Library World, (2003).
[4] Yan, L., Gui, Z., Du, W., & Guo, Q., “An improved
PageRank method based on genetic algorithm for web
search.”, Procedia Engineering, 15: 2983-2987, (2011).
[5] Cui, M., & Hu, S., “Search engine optimization research
for website promotion”. In 2011 International Conference
of Information Technology, Computer Engineering and
Management Sciences, 4: 100-103, (2011).
[6] Killoran, J. B., “How to use search engine optimization
techniques to increase website visibility.”, IEEE
Transactions on professional communication, 56(1):
50-66, (2013).
[7] Yalçın, N., & Köse, U., “What is search engine
optimization: SEO?.”, Procedia-Social and Behavioral
Sciences, 9: 487-493, (2010).
[8] Malaga, R. A. , “Worst practices in search engine
optimization. Communications of the ACM, 51(12):
147-150, (2008).
[9] “Google's Search Engine Optimization Starter Guide”,
(2013).
[10] Mittal, M. K., Kirar, N., & Meena, J. “Implementation of
Search Engine Optimization: Through White Hat
Techniques.”, In 2018 International Conference on
Advances in Computing, Communication Control and
Networking (ICACCCN), 674-678), (2018).
[11] Gandour, A., & Regolini, A., “Web site search engine
optimization: a case study of Fragfornet.”, Library Hi
Tech News, (2011).
[12] Asllani, A., & Lari, A., “Using genetic algorithm for
dynamic and multiple criteria web-site optimizations.”,
European journal of operational research, 176(3):
1767-1777, (2007).
[13] Boyan, J., Freitag, D., & Joachims, T., “A machine
learning architecture for optimizing web search
engines.”, In AAAI Workshop on Internet Based
Information Systems, (1996).
[14] Kiritchenko, S., & Jiline, M., Keyword optimization in
sponsored search via feature selection. In New
Challenges for Feature Selection in Data Mining and
Knowledge Discovery, 122-134, (2008).
[15] Zimniewicz, M., Kurowski, K., & Węglarz, J.,
"Scheduling aspects in keyword extraction problem.”,
International Transactions in Operational Research,
25(2): 507-522, (2018).
[16] Joshi, A., & Motwani, R., “Keyword generation for
search engine advertising.”, In Sixth IEEE International
Conference on Data Mining-Workshops (ICDMW'06,
490-496, (2006).
[17] Abhishek, V., & Hosanagar, K., “Keyword generation for
search engine advertising using semantic similarity
between terms.”, In Proceedings of the ninth
international conference on Electronic commerce, 89-
94, (2007).
[18] Sordoni, A., Bengio, Y., Vahabi, H., Lioma, C., Grue
Simonsen, J., & Nie, J. Y., “A hierarchical recurrent
encoder-decoder for generative context-aware query
suggestion.” In Proceedings of the 24th ACM
International on Conference on Information and
Knowledge Management, 553-562, (2015).
[19] Hong, Y., Vaidya, J., Lu, H., & Liu, W. M. Accurate and
efficient query clustering via top ranked search results. In
Web Intelligence, 14(2): 119-138, IOS Press, (2016).
[20] Süzek, T. Ö. Using latent semantic analysis for automated
keyword extraction from large document corpora.
Turkish Journal of Electrical Engineering & Computer
Sciences, 25(3): 1784-1794, (2017).
[21] Varçın, F., Erbay, H., & Horasan, F., “Latent semantic
analysis via truncated ULV decomposition.”, In 2016
24th Signal Processing and Communication
Application Conference (SIU), 1333-1336,. IEEE,
(2016).
[22] Horasan, F., Erbay, H., Varçın, F., & Deniz, E. “Alternate
Low-Rank Matrix Approximation in Latent Semantic
Analysis.”, Scientific Programming, (2019).
[23] Martin, D. I., & Berry, M. W., “Mathematical foundations
behind latent semantic analysis.”, Handbook of latent
semantic analysis, 35-55, (2007).
[24] Berry, M. W., & Fierro, R. D., “Low‐rank Orthogonal
Decompositions for Information Retrieval
Applications.”, Numerical linear algebra with
applications, 3(4): 301-327, (1996).
[25] Duman E., Erbay H., “Latent semantic analysis approach
for automatic classification of web pages contents.”,
Master Thesis, (2013).
[26] Shima, K., Todoriki, M., & Suzuki, A., “SVM-based
feature selection of latent semantic features.”, Pattern
Recognition Letters, 25(9): 1051-1057, (2004).
[27] Uysal, A. K., & Gunal, S. Text classification using genetic
algorithm oriented latent semantic features. Expert
Systems with Applications, 41(13): 5938-5947, (2014).
[28] Jessup, E. R., & Martin, J. H.,“Taking a new look at the
latent semantic analysis approach to information
retrieval.”, Computational information retrieval, 121-
144, (2001)