Telematics and Informatics Reports 11 (2023) 100094

Contents lists available at ScienceDirect

Telematics and Informatics Reports

journal homepage: www.elsevier.com/locate/teler

Marketing insight from consumer reviews: Creating brand position through

opinion mining approach
Mustafa Kemal Yılmaz a, Hale Tuğçe Altunay b, *
a
University of Samsun, Faculty of Economics, Administrative and Social Science, Department of International Trade and Business, Turkey
b
Isparta University of Applied Sciences, Sütçüler Prof. Dr. Hasan Gürbüz Vocational School Management and Organization Department Logistics Program, Turkey

A R T I C L E I N F O A B S T R A C T

Keywords: Purpose: Brand managers can determine the ideas, perceptions, or experiences of consumers about their brands
Brand positioning and their competitors through traditional marketing research techniques such as focus group discussions, sur­
Perceptual maps veys, and in-depth interviews. Through these methods, brands can observe the visual expression of consumer
Text mining
perceptions concretely by using perception maps or radar chart techniques. However, traditional methods have
Opinion mining
Global brands
certain limitations. The validity of brand positioning in perception maps may be questioned since consumers’
answers to survey questions can be misleading, consumers may express opinions about a product that they have
not purchased yet, and other methods may divert the opinions of consumers at a specific stage.
In this study, global brand positions in the minds of consumers were concretized through perception maps by
analyzing product comments on e-commerce sites through the use of opinion mining/sentiment analysis, one of
the text mining methods. Moreover, perception maps and radar charts were created for the selected brands by
forming the dataset from consumers’ reviews about brands and products in online environments. perception map
errors caused by traditional methods have been prevented by using opinion mining/sentiment analysis, which
constitutes the originality of the research.
Design/methodology/approach: In this study, the text mining method, which includes opinion mining and senti­
ment analysis techniques, has been used to create perceptual maps from consumer reviews about products on e-
commerce platforms. The TF*IDF method is used to identify the features to be compared in the perceptual map,
which selects the features with weighted importance among the features mentioned by consumers in their
comments. In this study, a lexicon-based opinion mining technique has been used to assign sentiment scores to
consumers’ comments on e-commerce websites for each product and its specified features.
Findings: The findings of the study have illustrated the areas in which the brands are evaluated as “insufficient” or
“successful” by consumers within the framework of the characteristics selected in the study. It has been revealed
that the method used in the research can provide accurate results for brand managers to make positioning de­
cisions and develop effective marketing strategies.
Research limitations/implications: The research has been conducted by analyzing product reviews about selected
brands on e-commerce platforms. The comments selected within the scope of the research include only those
written in the Turkish language, and the evaluations have been carried out on the comments made until a specific
date.
Originality/value: In this study, the analyses of the comments on the selected brands have ensured that the
businesses are informed of the level of perception about their products in the market and that the research is
based on a holistic approach that helps other consumers to develop a perspective on a specific product based on
the comments of a consumer who shares his/her experience about the aforesaid product.

Introduction the world has become a single global village [1], which results in an
increase in numerous factors such as the number of global businesses
With the acceleration of the concept of globalization since the 1980s, and product diversity (Bakırtaş et al., 2009). The innovation brought by

* Corresponding author.
E-mail addresses: [email protected] (M.K. Yılmaz), [email protected] (H.T. Altunay).

https://doi.org/10.1016/j.teler.2023.100094
Received 5 April 2023; Received in revised form 17 August 2023; Accepted 20 August 2023
Available online 22 August 2023
2772-5030/© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

globalization and the resulting challenges has led businesses to try to Concepts and theoretical background
keep up with the increasing competition in the world [2]. The efforts of
businesses to resist competition have led to the production of numerous Sentiment analysis/Opinion mining
alternatives for a single product and to the continuous exposure of
consumers to countless stimuli in environments where intense commu­ Ideas are often based on subjective judgments as a result of emotions
nication is experienced every day [3]. As a result, businesses have shaped around a problem, issue, fact, or thought. Expressing a personal
sought to gain a different position in the minds of potential consumers opinion about a situation is one of the most common uses of oral and
compared to their competitors [4]. Creating strong brand equity and a written communication [13]. Every behavior in our daily lives is the
transparent and consistent brand image for consumers are among the result of ideas; this is why ideas play an important role in the interaction
important conditions for businesses to be in the “right” position in the between individuals. For this reason, when we want to buy a product or
minds of consumers ([5]: 79). The businesses should first determine the learn about a product we do not yet know about, we first seek the
perception of their brands and products in the minds of consumers in opinions of others [14]. In the past years, individuals used to commu­
order to address consumer perceptions “correctly” [6]. For this purpose, nicate with the people in their immediate surroundings to fill this in­
perceptual maps are the tool that provides the highest value to the formation gap, but today, thanks to the restructured updates of blogs,
business in positioning strategies. The advantage of perceptual maps for social media, and e-commerce platforms, they collect opinions from
marketing managers derives from the concrete presentation of the per­ these platforms [15].
ceptions of consumers in the market [7]. For example, with the emergence of global services such as e-Bay,
Traditional marketing research methods (surveys, interviews, etc.) TripAdvisor, and Amazon, most web users have gone from being mere
are generally used to create perceptual maps (Karataş and Altunışık, users of web applications to actively becoming the center of these ap­
2015). However, the reliability of consumers’ answers to survey ques­ plications [16]. Therefore, the ideas that consumers share about prod­
tions is often a matter of debate. Expressing his/her opinion even though ucts and brands on these platforms every day have accumulated and led
he/she has no knowledge about the brand he/she is supposed to to the emergence of collective intelligence [17]. However, the huge
comment on can be given as an example for this issue of reliability. Due amount of unstructured content created by consumers makes it difficult
to the limitations arising from traditional marketing research methods to be analyzed manually [18]. In order to solve this problem, the concept
[8], it is aimed to solve the problem that arises in the creation of of sentiment analysis, which is based on the automatic analysis of tex­
perceptual maps by taking into consideration the consumer product tual data in the web environment, was first introduced by Tetsuya and
reviews that have accumulated on e-commerce platforms in recent Jeonghee in 2003, and in the same year, Dave et al. started to use it as
years. opinion mining [19].
According to the “Digital 2021: Global Overview” Report, 66.6% of Opinion mining studies are basically analyzed at three levels: (1)
the world’s population of 7.83 billion at the beginning of 2021 uses a Document-Based, (2) Sentence-Based, and (3) Aspect-Based [20]. In
mobile phone, while 59.5% uses the internet. Access the report here: document-based mining studies, the document to be analyzed is
https://datareportal.com/reports/digital-2021-global-overview-report considered as a whole without being reduced to the sentence level and
[9]. The intensive use of the Internet in the world has reduced the dis­ the opinions in the whole document are dealt with whether they are
tance between the buyer and the seller, and shopping can be completed positive, negative, or neutral [12,21]. However, it is quite unreliable to
in a shorter time through online sources [10]. Moreover, e-commerce draw a pure conclusion from these sources due to the large dataset of
has become a growing area due to the advantages of e-shopping such as documents written in natural language text [22]. In sentence-based
the ability to do e-shopping at on-demand hours and the wide variety of studies, it is decided to perform sentiment analysis on each of the sen­
products offered to consumers [11]. Thanks to the innovations brought tences in the document rather than finding a single sentiment to
to e-commerce platforms, consumers have been able to share their ex­ represent the entire document. However, reviewers may write “Excel­
periences about the product they have purchased with other consumers, lent” about one feature of the product and “Very bad” about another
and other individuals who do not have information about the product feature. Therefore, this level of opinion mining, which does not reflect
have started to be influenced by each other in terms of purchasing de­ the opinions expressed in different ways for product features, cannot
cisions. The growing number of consumer reviews on different channels provide deep information to the researcher [23]. At the level of
such as forums, blogs, and e-commerce websites makes this data worth aspect-based opinion mining, also known as goal-based opinion mining,
examining for businesses [12]. However, the manual processing of huge the goal is to convey to the researcher directly the sentiment of each
amounts of textual data, also called “Big Data”, is almost impossible due feature that the consumer writes about the product [20] and to deter­
to time and cost constraints for researchers to extract meaningful in­ mine the sentiment direction of each feature in the sentences that make
formation from this database. Therefore, our study is based on the use of up the entire document in the summary extracted from big data [24]. Hu
opinion mining/sentiment analysis algorithms, which include text and Liu [25] proposed aspect-based sentiment analysis for a more
mining and natural language processing techniques, to solve the complete analysis to identify the positive or negative state of emotions
above-mentioned problem. for each feature.
In this context, the researchers have collected comments on 6 mobile Both in the aspect-based technique and in other opinion mining
phone brands selected from e-commerce platforms where consumers in techniques, there are different methods for determining the sentiment of
Türkiye share their opinions after the purchase, they have used aspect- consumer opinions in analyzed documents [26]. These methods are
based opinion mining methods to determine the product features that categorized under two headings: (1) Machine Learning-based methods
help to compare brands in perceptual maps, and they have calculated and (2) Lexicon-based methods [27]. It is observed that researchers
sentiment scores for each comment using lexicon-based sentiment mostly use Naive Bayes and Support Vector Machines algorithms among
analysis approach. At the end of this process, they have positioned the machine learning methods in opinion mining studies [26].
scores obtained from the analyses to the coordinates of the relevant In this study, consumer comments on mobile phone brands were
brand on the map in the perceptual maps created according to the examined at the level of aspect-based opinion mining, and the sentiment
features. expression of the comments made for each feature was embodied in
perception maps by calculating the sentiment score by means of the
lexicon-based method. For this reason, Section 2.2. will discuss the
lexicon-based sentiment analysis approach.

2
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

Lexicon-based sentiment analysis the most effective way and to get the information needed by businesses
when making brand positioning decisions in the context of strategic
The lexicon-based approach is a method that involves the compu­ marketing [30]. Brand positioning techniques, referred to as perceptual
tation of the semantic aspect in words or sentences in a document [28]. maps or mind maps, are schematic tools for marketers to visualize
The sentiment lexicon is the element needed for computation in this competitive gaps in the market or consumer opinions about rival firms.
approach. The main task of the sentiment lexicon, which is also defined Methods such as surveys, interviews, or text mining help to create and
as an opinion lexicon, is to find the dictionary equivalents of the words sustain a competitive advantage by obtaining customers’ opinions about
in the document, classify them as positive, negative, and neutral, and the brand [33]. They provide businesses using perceptual maps with a
assign a sentiment score to these words [25]. Since the beginning of consumer-oriented perspective in achieving the ideal market position
sentiment analysis studies, many researchers have developed sentiment and enable them to make positioning decisions that differentiate them
lexicons and made them available to users [29]. The most popular of from their competitors [34].
these lexicons in the world is SentiWordNet, which mostly shows words
with their synonyms and antonyms in lexicon-based studies on the En­ Method description and opinion mining construction
glish language. In this platform, a score is given for each word by
checking whether the words used in the comments for all the features of The framework building
the product have a dictionary equivalent, and at the end of the process,
the sentiment pole and total score of the relevant sentence are deter­ The figure below shows the steps involved in the design of the
mined by summing up all the sentiment scores given to the words in the research:
sentence. For example, Lee et al. [30] used a goal-based approach to Fig. 1 shows the general workflow to be followed in the research. The
collect reviews about selected cell phone models from Amazon and research suggests creating perceptual maps based on lexicon-based
PhonoArena websites and assigned sentiment scores to consumer re­ sentiment analysis from consumer reviews of the pioneer/flagship
views written in English using the WordNet dictionary. Kızılkaya [31], models that mobile phone brands offer for sale every year. In this study
on the other hand, used sentiment analysis to interpret public opinions focusing on a systematic analysis, the six stages in the figure are divided
about political parties and their representatives in Türkiye by collecting into two basic stages in order to create perceptual maps: The first stage is
tweets from the Twitter database and calculating the sentiment score of to collect data on post-purchase experiences about the product and
each tweet using the sentiment lexicon he created. analyze them with text mining methods, and the second stage is to
identify the product features that consumers mention most frequently
Perceptual mapping with the help of a sentiment lexicon, score them and place them in their
positions in perceptual maps Figs. 2 and 3.
In a competitive business environment, global brands need infor­
mation from consumers to continue to operate and increase profits due Research scope
to ever-changing external environmental factors related to customers,
competitors, regulations, or countries [32]. In this sense, perceptual In the past, many products such as cell phones were mostly catego­
maps are visual diagrams that enable to illustrate where and how the rized in the group of “luxuries”; however, today, there has been a great
brand and its competitors are positioned in the minds of consumers in increase in the number of businesses operating globally and of products

Fig. 1. Steps of research design.

3
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

produced by these businesses, and the products that are considered Hepsiburada, N11, GittiGidiyor, Amazon Türkiye) until the date of April
luxury have turned into the ones that can be easily purchased by 23, 2021.
everyone together with a constantly developing and self-renewing world The data was collected with the “Data Miner” program, a Google
understanding [35]. As a result, mobile phones have become a phe­ Chrome plugin. Data Miner is a Google Chrome extension that allows
nomenon that integrates with our daily lives [36]. you to extract all comments (data) found on websites or social media
The advantages of smartphones, such as offering a large number of platforms. The consumer comments were collected from 5 different e-
product models that appeal to everyone, have driven global brands commerce websites with the Data Miner program between the date
operating in this sector into intense competition with each other. By the when the selected products were listed on the websites and April 23,
end of 2020, the global smartphone penetration rate was 46.5%, and the 2021. In this context, a total of 6081 comments were collected for the
number of smartphones sold worldwide reached 1 billion 380 million. research and saved as CSV files in the Excel database. In the next step,
Access the report here: https://www.statista.com/statistics/263437 6081 comments on e-commerce websites were put through the text
/global-smartphone-sales-to-end-users-since-2007/ [37]. Based on all preprocessing process, which will be explained in the following section,
these data, in our study, we will carry out a sentiment analysis for the to make them ready for analysis. Except for the extraction of data from
comments on the selected brands and we will visualize the perception of the relevant web addresses, all stages were carried out in R software
the customers about the smartphone brands with the help of perceptual language.
maps for the use of brand managers.
Since the opinion mining method will elaborate on the comments
written in Turkish, the study included the brands with the highest Text preprocessing
market share in the Turkish market. According to Statcounter
(https://datareportal.com/reports/digital-2021-global-overview-report Text preprocessing is the next step after the creation of the text
(date of access: 17.04.2021) The cell phone brands with the highest collection. Proses are the main subjects of text mining, and they can be
percentage of market share in Türkiye are as follows: written in colloquial language. This reality has posed the biggest chal­
The reviews written about the mobile phone brands listed above will lenge due to the unstructured nature of text mining data compared to the
be collected from e-commerce websites operating in Türkiye. Trendyol, structured data of data mining. Therefore, the texts have to go through a
Hepsiburada, N11, Gittigidiyor, and Amazon Türkiye were selected as long pre-processing stage to extract information patterns from the data
the largest e-commerce websites in Türkiye within the scope of this collected in the previous phase. This stage should inevitably be carried
research (Access the report here: https://www.eticaret.tv/h out with precision by the researchers for the output to be accurate. In
aberler/icerik/turkiyedeki-en-buyuk-e-ticaret-siteleri/). this way, the text data will be cleaned and made more suitable for
analysis.
The comments contain many spelling mistakes, emojis, irrelevant
Determination of competitive brands
sentences, etc. that should be excluded from the dataset. The irrelevant
data in the data set of this study consists of the contents that do not
In addition to the main model determined by brands in the produc­
include any comment/opinion about any feature of the selected product
tion and sale of phones, many different side models are offered for sale to
in consumer reviews. In some websites where consumer reviews were
satisfy the demands and needs of different market segments. This study
obtained, the comments were only about product sellers, cargo companies,
focused on the leading models, colloquially referred to as “flagship”
fast or slow delivery of the product, or incomplete/incorrect delivery.
models that brands put on sale during the year. Since the products to be
Therefore, at this stage, comments directly about the features of the
positioned in the perceptual maps at the end of the study should be up-
product were selected, and “irrelevant” ones were excluded from the
to-date, we have selected the products that have recently entered the
data set. As a result, the number of data to be analyzed decreased from
market and we have taken into consideration whether the number of
6081 to 3212. After this stage, the data became suitable for opinion
comments about these products is sufficient. As a result of the evalua­
mining methods.
tions, the phone models to be analyzed with the opinion mining method
Although the text preprocessing stage is a long process that requires
are listed below:
time, it is an important process for the research to be conducted
correctly. In this study, the R software program was used to prepare the
Collection of comments from websites data for analysis. The text preprocessing stage consists of subjecting the
texts to different processes such as correcting spelling mistakes, sepa­
All consumer reviews written about the above-mentioned products rating words into their roots, removing punctuation marks and stop
were collected from the selected e-commerce websites (Trendyol, words, and Turkishization (Oğuzlar, 2011: 31).

Table 1 • Removal of Stop Words;

Market shares of mobile phone brands in the Turkish market (%).
Samsung Apple Xiaomi Huawei Oppo General Mobile In the comments, there are words such as “for, and, but, however”,
43,46 15,54 14,13 13,23 2,65 2,5
prepositions and conjunctions that do not express any consumer opinion
in sentiment analysis. At this stage, the comments in the “stopwords”

Table 2
The information of competing products.
Product/Info Samsung Galaxy s20 iPhone 11 Xiaomi Mi Note 10 Huawei P30 Oppo Reno 4 GM 20 Pro

Total reviews 403 2934 273 397 784 1290

Time to market Feb. 2020 Oct. 2019 Nov. 2019 March 2019 June 2020 August 2020
Period of reviews until 23/04/2021 until 23/04/2021 until 23/04/2021 until 23/04/2021 until 23/04/2021 until 23/04/2021
Price (TL) 6000 8000 6000 6500 4000 2050
Rear camera 64 mp 12 mp 108 mp 40 mp 48 mp 48 mp
Front camera 10 mp 12 mp 32 mp 32 mp 32 mp 16 mp
Memory 128 gb 128 gb 128 gb 128 gb 128 gb 128 gb
Battery 4000 mAh 3110 mAh 5260 mAh 3650 mAh 4015 mAh 4050 mAh

4
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

general corpus. The image below shows the representation of the stop­
word list in the R:

• Text Normalization;

In the text normalization step, which is one of the text preprocessing

stages, spelling mistakes in the Turkish comments written in colloquial
language were corrected. Since the comments written by consumers on
e-commerce sites are often written in colloquial language, they are not
written correctly. For this reason, spelling mistakes in the comments or
keyboard-related errors such as using “i”, “u”, “c” instead of “i”, “ü”, “ç”
were corrected in the text normalization step of the text preprocessing
stage to carry out the analysis efficiently. To sum up, comments written
in colloquial Turkish were converted into official Turkish to ensure the
accuracy of the analysis.
This process was implemented through the use of the Istanbul
Technical University Turkish NLP Web Service API. ITU NLP (Istanbul
Technical University Natural Language Processing) is a research group
established at Istanbul Technical University for different interdisci­
plinary studies that include text mining, data mining, natural language
processing, and knowledge extraction. They develop software tools that
can be used by researchers in the field of computational linguistics,
including Turkish language and other languages (Eryiğit, 2014: 1).
Using the Normalization method of the ITU NLP Web Service tool, errors
in the data set that would affect the validity of the analysis, such as
misspellings, were corrected and the text was normalized for all
comments.
Using the Normalization method of the ITU NLP Web Service tool,
errors in the data set that would affect the quality of the analysis, such as
spelling mistakes, were corrected and the text was normalized for all
comments. As a result of the observations made, the words/sentences
that the ITU NLP tool could not provide 100% correction in the texts
were edited manually.
After the Normalization method was applied to the entire dataset,
the text was checked and errors were found that the service could not
correct; in addition, some words could not be corrected and other words
Fig. 2. Stopwords list view in R program. transformed into different words after the system was run. As a result of
the examinations, the words that could not be corrected and trans­
data set, which do not have any meaning for analysis in the sentiment formed were edited manually. The table below shows example words
analysis/opinion mining study to be conducted with the help of senti­ that could not be corrected in the comments collected for the cell phone
ment dictionary, were cleaned with the R software. The stopword list and the words that the system transformed:
was introduced to R and the words in this list were extracted from the

Fig. 3. Text normalization.

5
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

Table 3 Table 4
Words edited by manual method. The exemplary price feature set.
words in the consumer ITU NLP normalization with manual method a competitive product in terms of price
comment result edited result
Price is affordable
Iphone abone iPhone Too expensive
Samsung samsun samsung Expensive compared to other phones on the market
camera şamara kamera You could get a better flagship model for this price
Iphone xr İphone zer iPhone XR

explored to create perceptual maps. The TF*IDF approach, which

Feature selection quantifies the importance of the terms in the collection, is preferred for
feature extraction. Accordingly, “price, design, battery, performance,
The aspect-based (goal-based) approach of Hu and Liu [25] was used sound, camera, processor, screen” were identified as the product fea­
to extract and cluster the product features mentioned in the comments tures that would reflect the comparative positions of brands in the
and to efficiently find useful information for the research. In the perceptual maps for mobile phones.
aspect-based sentiment analysis approach, the aim is to identify the
features of the product to be analyzed and to determine consumers’ Clustering
opinions about these features. Therefore, in aspect-based sentiment The main purpose of the feature extraction phase is to identify the
analysis research, it is necessary to determine the emotion-attributed features that will enable brand positioning on the perceptual maps. In
features of the product, which is considered as an “entity” and which this context, the features to be included on the horizontal and vertical
is identified as mobile phones in this study. axes of the map were determined as a result of the analyses conducted in
In this study, the aspect-based (opinion-based) opinion-mining the previous stage. The total number of comments on the 6 selected
approach proposed by Hu and Liu [25] was preferred to determine brands is 3212. It is not correct to say that consumers express opinions
consumers’ perceptions about the “product features” (price, perfor­ about only one feature of the product in the comments that constitute
mance, design, camera, etc.) of the 6 global mobile phone brands. the dataset of the study. For example;
Based on the view that the product features that consumers mention
the most in the consumer feedback on e-commerce platforms are the - “camera and processor and screen are really nice if you are a gamer you
most important aspects for them [8], we used the TF*IDF method, which need to charge it twice a day the price is good”
selects the most frequent and predominantly important features in the -“camera is great but looks a bit rough but still affordable”
comments and which is frequently mentioned in the literature [38,39],
to extract these features from the text document. Many features of the phone are mentioned in these comments, and
The purpose of preferring the TF*IDF technique in this study is not to the statements carry different emotional states for each feature. There­
analyze the words within each individual comment, each with different fore, in this stage, a manual analysis was carried out on which feature of
content, and extract features on a comment-by-comment basis; rather, it the product consumers expressed an opinion about, and the comment
aims to scan all comments within the entire document and discover was included in the scope of the relevant set among all the feature sets
multiple features by assigning a statistical weight to each word determined in the previous stage.
appearing in each comment (aspect extraction) in relation to the total At this point, an exemplary “iPhone 11′s Price Feature Set” will be
document. While BERT or another technique could be used to under­ depicted as follows:
stand the prominent feature within a single comment, this study aims to This stage was applied to all brand reviews. The comments for the 6
determine the most prominent features within a dataset that encom­ brands were read one by one, and the sentence in the comment was
passes all comments, rather than focusing on extracting features from a assigned to the feature of that brand, whichever feature was commented
single comment. TF*IDF is preferred over BERT because it statistically on in the comments. Considering that there are 8 feature sets for 6
weights the importance of words occurring within a document. In other brands determined in the grouping stage performed by manual process,
words, instead of reducing it to a single comment for feature extraction, a total of 48 feature sets were created. In the next stage, sentiment scores
the aim is to emphasize the importance of a term within both a single were given to these feature sets by using a lexicon-based sentiment
comment and the total number of comments. Additionally, the BERT analysis approach, and then the numerical values were placed on the
technique is oriented towards calculating the sentiment polarity of each coordinates in the perceptual maps created according to the features.
consumer comment within the document (positive, negative, or
neutral). This study goes beyond determining the sentiment polarity of Calculating opinion score for feature sets
consumer product reviews and assigns sentiment scores to each There are multiple sentiment lexicons developed for the sentiment
comment [40]. analysis approach in the literature. For example, Uçan’s sentiment
TF*IDF not only measures the number of times the term appears in lexicon which he translated from English to Turkish in 2014, is one of
the document, but also highlights the importance of the term for the the most basic lexicons in this field [42]. The present study was carried
document in numerical terms. That is, it emphasizes the importance of a out with the use of the sentiment lexicon consisting of 49,241 terms
term in a comment as well as in the total number of comments [41]. aimed by Sağlam et al. [43] to be used for general purposes. Sağlam et al.
The R program was used to create the TF*IDF table within the scope [43] first created the “SWNetTR-PLUS General Purpose Sentiment
of this study. After the TF*IDF list was generated, each term (74,254) Lexicon” with a volume of 37 K, including the common terms (4 K) that
presented by the algorithm was identified as a potential feature. These they identified by combining the dictionary covering 23 K terms by Uçan
terms were then ranked according to the TF*IDF score and the terms that [42] and the 10 K dictionary developed within the GDELT project. Later
could be the features for the aforesaid cell phone models were selected on, they needed to develop a new lexicon due to the lack of capacity of
according to the sample size and sentence length. For example, there are the 37 K lexicon and the low performance of the present lexicon on the
some terms that can or cannot be selected as features in reviews. While dataset. The current SWNetTR++ lexicon, which they created by adding
the terms “sound, storage, shoot, charge” can be features of a cell phone, words with synonyms and antonyms, contains 49 K terms. As a result of
words such as “good, product, telephone” do not have the quality to be the analyses, it was concluded that the current lexicon gave better re­
included in the feature category. sults compared to the previous one. Therefore, the SWNetTR++ lexicon
At this stage, key terms/features related to the cell phone were developed by Sağlam et al. [43] was used in this study considering that
its high volume would provide an advantage over other sentiment

6
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

lexicons.
Each word in Sağlam et al.’s 49 K sentiment lexicon has an equiva­
lent sentiment score. After the text preprocessing stages, each word in
each comment in the data set consisting of 3212 consumer reviews was
matched with the scores in this sentiment lexicon, and then the senti­
ment scores of the words in that comment were averaged, resulting in a
total sentiment score for the relevant sentence/comment.
A sample diagram of how the sentiment score is assigned to each
comment and each word within a comment in the texts consisting of
consumer reviews is summarized as follows:
All comments in the feature sets of each brand were given a senti­
ment score, and then these scores were averaged and the result was the
final numerical value of the relevant set of that brand. Tables 1-5.
The R software program was used to assign sentiment scores to the
comments on each aspect/feature of the brands. The sentiment lexicon
of Sağlam et al. [43] was first read into the program, and then the file Fig. 4. Perceptual map for price and performance.
was called from the program to assign a sentiment score to the com­ While the horizontal axis shows how consumers position the brands in
ments in the relevant feature set. The same procedure was applied to 8 their minds in terms of the “Price” feature of the product, the vertical Y-
different feature sets of 6 brands. axis shows the perceptions on the “Performance” - also known as “Qual­
Accordingly, the final score of each brand for each feature is shown ity” for this study - of the phone models. In addition, the size of the circles
in Table 6. in which the brands are positioned on the perceptual map represents the
proportion of their share in the Turkish market. Fig. When the perceptual
Building perceptual maps map in Fig. 4 is analyzed, Huawei P30 and Oppo Reno 4 models with the
same sentiment score (0.33) are the brands that are in a more favorable
The final stage of the research is the creation of perceptual maps to position in terms of price compared to others. The phone models
see the position of brands in the minds of consumers. By applying perceived as more “expensive” than other brands are iPhone 11 (0.21) and
opinion mining methods to consumer comments about brands and their Xiaomi Mi Note 10 (0.25), respectively. In terms of the price feature of
products, an average sentiment score was obtained from the selected these models, the average price values were found to be lower because the
features of the brands. The purpose of this process is to determine how comments mentioned that these products were expensive more frequently
cell phone brands are positioned in the minds of consumers in terms of than other brands. In other words, those closest to the point of origin are
various variables, which is the main subject of this study. The brands more negatively positioned in the consumer’s mind than the others.
and their values obtained through the use of the sentiment lexicon will The performance feature on the vertical axis shows where consumers
be used to place the brands on the horizontal and vertical axis to be position the brands by taking the average sentiment score of the com­
positioned in the perceptual maps. ments where the product is described as “well-qualified” in the com­
In this context, the variables to be placed at both ends of the ments and in which the evaluations were made on the whole product,
perceptual maps were categorized into 2 groups and the perceptual that is, its performance, regardless of a specific feature.
maps were designed to show how consumers positioned the selected Accordingly, it can be said that Xiaomi Mi Note 10 is the brand with
brands in their minds according to the groupings created within the the highest performance compared to others (0.41) although consumers
scope of the research. perceive this brand as relatively expensive (0.25). In other words, this
In the perceptual map below, price-performance features are cate­ brand, which the consumers position in their minds as the most
gorized into 2 groups: expensive product after iPhone 11, is perceived as the brand with the
highest performance. Xiaomi Mi Note 10 is followed by Oppo Reno 4 and
Table 5
General Mobile GM 20 Pro. Based on consumer reviews, it can be stated
Sentiment score assignment scheme to comments.
that Oppo and General Mobile are neck and neck in terms of quality.
Consumer Reviews Score of Words Total Score of The
Compared to other brands, the Huawei P30 is the phone model that is
Comment
positioned in minds as less qualified. This model of Huawei is perceived
Telefon seri hızlı Telefon = 0,042,097,439 Seri = 0,221,915,789 to be cheaper than other brands in terms of price, while there are
0,077,151,742 Hızlı = 0,102,666,608
Batarya fazla Batarya = 0,067,396,234 Fazla = 0,573,449,232
negative opinions in terms of performance.
ısınıyor kamera 0,049,018,712 Isınıyor =
güzel 0,112,797,771 Kamera =
0,05,219,811 Güzel = 0,292,038,405

Table 6
Opinion scores.
Brands/ Samsung iPhone Xiaomi Huawei Oppo GM
Features Galaxy 11 Mi Note P30 Reno 20 Pro
S20 10 4

Performance 0,36 0,31 0,41 0,30 0.40 0.39

Price 0,30 0,21 0,25 0,33 0.33 0.29
Battery 0,01 0,12 0,16 0,20 0.15 − 0.08
Processor − 0,06 0,19 0,16 0,24 0.17 0.10
Screen 0,49 0,26 0,47 0,41 0.44 0.39
Camera 0,38 0,27 0,38 0,38 0.35 0.36
Sound 0,26 0,30 0,16 0.45 0.38 0.25
Design 0,47 0,30 0,32 0.41 0.48 0.44
Average 0,28 0,25 0,29 0.34 0.34 0.27
Score
Fig. 5. Perceptual map for processor and battery.

7
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

In the perceptual map above, there are brand positions created ac­ If many products were shown at the same time, the lines representing all
cording to consumer perceptions for two important features that are products would overlap and the graph would not be easily read. For this
effective both in determining the price/marketing policies of businesses reason, it is necessary to limit the number of products shown on the
in the process from the production of the mobile phone product to its radar charts. In addition to showing multiple features at the same time,
sale and in consumer purchasing decisions. radar charts differ from perceptual maps in terms of displaying only a
Processors are the brains that enable all electronic devices, especially limited number of products/brands on a chart.
smartphones, to perform their basic functions. For the mobile phone In Table 7, two radar charts were created by limiting the number of
models that are the subject of this study, the processor is the main reason products (brands) as the perception maps were incomplete to compare
why consumers sometimes experience problems such as slowing down brand positions in the context of multiple features. A clustering
and freezing in the system which makes the transition between the op­ approach based on the average of all feature sentiment scores was used
erations performed on the phone fast and easy. Therefore, as indicated in to determine the brands to be displayed in the same graph.
the comments, it has been observed that the “Processor” feature has an The researchers ranked the top three brands with the highest “nearly-
important place in the purchase decision of consumers, and at this point, the- same- score” (Huawei, Oppo and Xiaomi) among the 6 global
the position of the aforesaid feature in the consumer’s mind was shown brands’ average sentiment scores in Table 6 and the three brands with
through a perceptual map by grouping it with “Battery” feature. the lowest average (Samsung, Galaxy, iPhone) and they formed these
In terms of the processor feature, the Huawei P30 performed the best two clusters, and they obtained two scores by averaging the averages of
(0.24). When consumer evaluations have been analyzed, it is perceived these brands with the averages of the brands in both clusters. In the
that this product ensures a fast transition between transactions and has a radar graphs below, the clustered brands were compared in terms of all
processor that can be used easily by consumers who play games. iPhone features Fig. 5.
11 has been the second brand after Huawei in the processor ranking. The three brands in the first cluster have higher average scores than
Apart from the aforesaid two brands, Oppo and Xiaomi are respectively the brands in the second cluster in terms of all features.
positioned better than General Mobile in terms of processor speed. The The radar graph in Fig 6 visualizes the brands with the highest
Samsung Galaxy S20 is positioned on the left side of the map with a sentiment scores in terms of all features of the brands obtained as a result
sentiment score of − 0.06 for this feature, which is quite poor compared of lexicon-based sentiment analysis studies. Accordingly, Oppo Reno 4 is
to other brands’ processor scores. the brand that covers the widest area of the three lines in the radar
Based on the comments made according to how long the phone graph. This shows that the Oppo brand provides more favorable results
batteries last in daily use (used without charging for a long time), in the selected product features compared to the other two brands. Ac­
Huawei P30 is positioned on the perceptual map with the best battery cording to this graph, Oppo Reno 4 has dominated its competitors in
ranking (0.27). Xiaomi Mi Note 10 (0.16) and Oppo Reno 4 (0.15) are almost all features except the processor. Contrary to the positive situa­
the brands that are perceived to be in the best position in terms of tion of the Oppo Reno 4 model, the Xiaomi Mi Note 10 is the brand that
battery after Huawei. iPhone 11 is perceived better in terms of battery occupies the least space in the graph and is represented by the red line.
than Samsung and General Mobile models, but more negatively than the Except for the performance and display features of the Xiaomi Mi Note
previously-mentioned 3 brands. This situation shows that the iPhone 11 10, all other features were negatively commented on in consumer re­
is situated in the middle position in terms of the battery feature among views. This can be easily seen in the perceptual maps in Fig. 6 and Fig. 7.
all brands. When the comments about the battery of the iPhone 11 The radar graph in Fig. 7 visualizes the lowest average sentiment
model have been examined, it has been seen that the brand has scores of the brands obtained as a result of lexicon-based sentiment
continuously improved this feature in its new models over the years and analysis studies. Accordingly, Samsung, iPhone and General Mobile
that it performs well above the battery performance of its older models. models have lower scores than the brands in Fig 7. In the radar chart
The Samsung Galaxy S20 model has been positioned closer to the origin (aka spider web chart) above, the iPhone 11 is the brand with the
on the map compared to other brands, as in the processor feature, and
this has confirmed that the relevant brand has resulted in a weak
perception in the consumer mind in terms of both the battery and the Table 7
processor. Brand sets.
Brand sets Average sentiment
Radar charts score
Perceptual maps are widely-used marketing tools to show where First Cluster: Huawei P30, Oppo Reno 4, Xiaomi Mi 0.32
brands are positioned in the minds of consumers for specific aspects. In Note10
Second Cluster: Samsung Galaxy S20, GM20 PRO, iPhone 0.26
this study, the eight features determined to see where the selected
11
brands are positioned in the minds of consumers have been positioned
according to their sentiment scores in the perceptual maps by forming
groups of 2. The limitation of the method is that perceptual maps only
show the brand locations based on two features placed on the horizontal
and vertical coordinates of the map. In this study, radar plots will be
used to fill the gap created by the perceptual maps as they show multiple
product features at the same time. Both perceptual maps and radar
charts are important tools for businesses to make effective and
competitive decisions within the framework of strategic marketing
planning [30].
Table 7 below shows the two clusters calculated using the average
scores of all brands within the scope of the features selected according to
the sentiment scores. The brands were divided into two clusters and then
displayed comparatively in radar graphs because product features were
compared by being displayed only on two axes - x and y - in perception
maps, and therefore multiple perception map graphs emerged.
In this study, radar charts will be used to fill the gap created by the
detection maps as they show multiple product features at the same time. Fig. 6. Radar chart: First cluster.

8
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

the product positions in both indicators will provide brands with an

opportunity to evaluate their relationship with their competitors in
terms of their product features.

Contributions to management

Qualitative data obtained from reviews is considered a source of

customer insight and can enable better creation of brand positioning
strategies. For example, when a company wants to enter the market with
a new product, it can potentially benefit from brand reviews to under­
stand consumers’ perspectives on this product. These reviews can be
found not only on the platforms where the product is sold but also on
social media, forums, and blogs. By focusing on the perceptions of the
new product, marketers can better illustrate the effect of the positive
brand image in their advertising and communication efforts. With the
Fig. 7. Radar chart: Second cluster. help of automatic perceptual maps based on the comments, it can be
measured by using the previous data whether the position of the brand
smallest footprint. The Samsung and General Mobile brands are very for the consumer has changed and whether the positioning efforts have
close to each other in this sense. Accordingly, with the exception of the increased the brand image. Moreover, this method will create a great
processor and audio features, the iPhone model is in a poor position opportunity to see the gaps in the market and will also provide busi­
compared to its two competitors. Samsung Galaxy S 20 and GM 20 Pro- nesses with the opportunity to avoid the fields that are dangerous in the
models are very close to each other in terms of all other qualities except minds of consumers.
processor and battery. In the past, businesses used techniques such as traditional surveys or
focus groups to determine customers’ perceptions of their products and
Conclusion competitors. The difference between the method used in this study and
the traditional methods provides advantages to businesses in several
Opinion mining is the process of extracting consumer opinions from ways. First of all, consumers may not have any past experience or even
reviews in textual format on the Internet. Consumers’ expressing their any knowledge about the brand asked in the perceptual map developed
opinions about a product or a topic using colloquial language is the main with the survey method; this may be misleading for the position of the
subject of opinion-mining analyses. Consumers’ opinions about brands brand on the map. However, in a comment-based method, the consumer
and their products in specific categories have provided great opportu­ who comments is someone who has already used the product and shared
nities for businesses to recognize customer perceptions and feedback in a their experience. Therefore, any inaccuracy in the analysis on the part of
less time-consuming and cost-effective way, away from traditional and the reviewer/consumer is out of the question here. Online evaluations
time-consuming customer feedback surveys. Research based on online are written in the colloquial language of consumers, so they accurately
product reviews not only provides businesses with considerations for reflect their satisfaction, desires, and concerns about the product, unlike
product improvement, repricing, or repositioning but also provides a inaccurate and incomplete answers in the questionnaire. This is due to
reference for global brands to analyze consumer perceptions of post- the fact that there is user confidentiality in the comments and consumers
purchase experience. openly express what they think about the product to other consumers
In this study, the comments made by consumers on e-commerce sites who may buy the product.
about the mobile phone models selected within the scope of the study Using these new approaches, brand managers can use the percep­
have been examined with opinion mining techniques based on text tions collected in customer reviews as a valuable source of input for
mining, and the brand positions in the minds of consumers have been future positioning decisions. The most accurate and effective marketing
visualized with perceptual maps and radar graphs. The perspective of decisions can be made by means of the user experience and perception of
this study is to enable global brands to capture customer insights based the purchasing process shared by consumers on the relevant platforms.
on opinion mining to assist in customer-centric product development The new product, new price, new market, and new location created
strategies. This methodology, which combines opinion mining and through R&D strategies can be channeled into the brand mantra to
perceptual mapping techniques, provides businesses with accurate and create a stronger brand image and identity. In addition, identifying the
effective strategic marketing information to learn the position of their perceptions of consumers in different country markets about the prod­
own brand and competitors’ brands in the minds of consumers. ucts of brands competing in the global market can make important
Since the data language used in the study is Turkish, the mobile contributions to the development of positioning strategies.
phone brands with the highest market share in Türkiye are included in
the scope of the study. In addition to identifying competitive brands, a Contributions to consumer
lexicon-based sentiment analysis approach has been used to score con­
sumer comments according to their sentiment, and brands have been The first behavior that consumers exhibit before purchasing a
positioned on perceptual maps according to their scores. The method­ product, regardless of having any information about the product, is to go
ology used in the study has been applied to consumer reviews on the through the reviews of other consumers on these platforms. With digi­
websites of Trendyol, Hepsiburada, GittiGidiyor, N11, and Amazon talization, online reviews on websites have become the first source that
Türkiye. With the help of the TF*IDF algorithm, it has been discovered consumers consult before making a purchase decision. In these envi­
which features of the product consumers most frequently commented ronments, other consumers who share their experiences and opinions
on, and then these features have been placed in groups on the perceptual about the use of a specific product provide guidance for consumers who
map axes. are in the phase of purchase decision. For example, if the reviews about
The main motivation for using this method is to convert consumers’ the product are negative, the consumer may give up buying the product;
post-purchase experience from online qualitative text format to quan­ on the contrary, if the reviews are positive, the consumer may decide to
titative data. By sharing their experiences, consumers not only act as buy the product. In this sense, online reviews contribute to businesses
voluntary sales agents for cell phone brands but also provide informa­ obtaining the target market perception of the product, while providing
tion to reduce the decision-making risks for consumers. It is thought that information/advice to consumers in a way that will affect their

9
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

purchasing decisions. Especially in technological products where risk Declaration of Competing Interest
perception is higher, user experiences help consumers to make the right
decision. Within the scope of the study, 6081 product reviews have been I declare that there is no conflict of interest in relation to this
collected from the selected e-commerce sites, and the dataset has been research article. I have no financial or personal relationships with other
reduced to 3112 by deleting irrelevant comments during the clustering people or organizations that could inappropriately influence this
process. The sentiment scores of the comments found with the opinion research.
mining method have been placed on the perceptual maps in terms of the
selected features, and the brand positions in the consumer mind have Data availability
been visualized. As a result, the contribution of the study to consumers is
thought to minimize the risk that may be experienced in the purchase Data will be made available on request.
decision process by analyzing product reviews based on scientific
methods and providing the information needed by consumers who will
choose between these brands. In addition, it is expected that the effec­ References
tiveness of consumers who comment on e-commerce platforms to reflect
mass market behavior will motivate other consumers to share more in­ [1] M. McLuhan, Gutenberg Galaksisi, Yapıkredi Press, İstanbul, 2014.
[2] O.J.R. Walker, J.W. Mullins, Marketing Strategy A Decision- Focused Approach,
sights by encouraging other consumers. In this way, customer perceived McGraw-Hill Irwin, New York, 2014.
value in the purchasing processes of technological products, where the [3] A. Ries, J. Trout, Positioning: The Battle for Your Mind, McGraw-Hill Education on
customer perceived risk is quite high, are positive. Brilliance Audio, New York, 2013.
[4] K. Kotler, K.L. Keller, Pazarlama Yönetimi, Beta Press, İstanbul, 2018.
[5] P. Kotler, Pazarlama 4.0, Optimist Press, İstanbul, 2019.
Limitations and directions for future research [6] J.P. Peter, J.C. Olson, Consumer Behaviour and Marketing Strategy, Irwin McGraw-
Hill, Boston, 1999.
[7] M. Baş, Marka Yönetimi, Detay Press, 2015.
This study has some limitations as in other studies in the literature. [8] A. Ekhlassi, F. Reshadi, A. Wan, Brand perceptual mapping by text mining online
First of all, this study is based on aspect-based sentiment analysis about product reviews, International Journal of Marketing and Business Communication
specific features of mobile phones. 8 features were identified for the 5 (3) (2016) 26–36.
[9] Datereportal. (2021). Smartphone sales and data. Retrieved from https://datarepo
brands analyzed in the study, and the comments were categorized ac­ rtal.com/reports/digital-2021-global-overview-report Accessed April 17, 2021.
cording to these features. However, it was observed that some comments [10] A. Lubis, U.S. Utara, Evaluating the customer preferences of online shopping:
were also made about the features not covered in this study although demographic factors and online shop applıcatıon issu, Academy of Strategic
Management Journal 17 (2) (2018) 1–13.
most of them included the selected features. In addition, the comment
[11] T.P. Monsuwé, B.G. Dellaert, K. De Ruyter, What drives consumers to shop online?
reviews were scored by the lexicon-based sentiment analysis method A literature review, International Journal of Service Industry Management 15 (1)
with the help of the sentiment lexicon created by Sağlam et al. [43]. (2004) 102–121.
[12] B. Özyurt, M.A. Akcayol, Fikir madenciliği ve duygu analizi, yaklaşımlar,
Although the scope of the sentiment lexicon used in the study was large
yöntemler üzerine bir araştırma, Selcuk University Journal of Engineering Science
(49 K), if a word mentioned in the comments was not included in the and Technology 6 (4) (2018) 668–693.
lexicon, no sentiment score could be assigned to that word. [13] E. Damer, Attacking Faulty Reasoning: A Practical Guide To Fallacy-Free
To summarize, in lexicon-based opinion mining, the researcher Arguments, Cengage Learning Press, Belmont, 2008.
[14] B. Liu, Sentiment analysis and opinion mining, Synthesis Lectures on Human
needs a sentiment lexicon that contains a high proportion of the words in Language Technologies 5 (1) (2012) 1–167.
the database and provides the mathematical equivalent of these words. [15] Ş. Karaca, N.G. Gümüş, Tüketicilerin online yorum ve değerlendirme puanlarına
However, the biggest limitation of this method is that the sentiment yönelik tutumlarının online satın alma davranışlarına etkisi, Sakarya Economics
Journal 9 (1) (2020) 52–69.
lexicon is created for general purposes and some of the words in the [16] G. Lackermair, D. Kailer, K. Kanmaz, Importance of online product reviews from a
content of the study do not have any equivalent in the sentiment lexicon. consumer’s perspective, Advances in Economics and Business 1 (1) (2013) 1–5.
In such a case, that word cannot be given a sentiment score and the [17] R. Filieri, M. Mariani, The role of cultural values in consumers’ evaluation of online
review helpfulness: a big data approach, International Marketing Review 38 (6)
accuracy rate of the study may decrease. For this reason, it is thought (2021) 1267–1288.
that in future studies, different results may be obtained through the [18] E. Cambria, R. Speer, C. Havasi, A. Hussain, Senticnet: a publicly available
analysis conducted by choosing any of the other sentiment analysis semantic resource for opinion mining, in: paper presented at the AAAI Fall
Symposium, 11-13 November, Virginia, USA, 2010. Retrieved from, https://www.
techniques in the literature.
bibsonomy.org/bibtex/174f773a5e6c365323ea7c25f60876f1c?lang=en. Accessed
Although the methodology is based on consumer reviews, in prac­ May 2, 2021.
tice, the sample preferred for creating perceptual maps or radar charts [19] K. Dave, S. Lawrence, D. Pennock, Mining the peanut gallery: opinion extraction
can easily be used to analyze reviews of other products. Since it is not and semantic classification of product reviews, in: paper presented Proceedings of
the 12th International Conference on World Wide Web, May, 2003. Retrieved
clear who made the comments collected in this study, no information from, https://www.kushaldave.com/p451-dave.pdf. Accessed April 20, 2021.
could be obtained about the demographic data of the commenters. It is [20] V.M. Pradhan, J. Vala, P. Balani, A survey on sentiment analysis algorithms for
thought that a broad perspective on brand positions can be developed opinion mining, Int. J. Comput. Appl. 133 (9) (2016) 7–11.
[21] Can, B. Alatas, Duygu analizi ve fikir madenciliği algoritmalarının incelenmesi,
based on demographic data in future studies. Besides, this study exam­ International Journal of Pure and Applied Sciences 3 (1) (2017) 75–111.
ines the phone brands with the highest market share in the Turkish [22] F.S. Çetin, G. Eryiğit, Türkçe hedef tabanlı duygu analizi için alt görevlerin
market. Therefore, the data language was Turkish, and foreign language incelenmesi – hedef terim, hedef kategori ve duygu sınıfı belirleme, Journal of
Information Technologies 11 (1) (2018) 43–56.
comments about the same brand were excluded from the scope of the [23] A. Özen, Müşteri şikayetlerinde fikir madenciliği: antalya helal oteller üzerine bir
study. It is considered that taking into account the interpretations made araştırma, in: Paper Presented 2nd International Halal Tourism Congress, 4-6 April,
in different languages in future studies will make significant contribu­ Antalya, Turkey, 2019. Retrieved from, https://www.halaltourismcongress.com/2
IHTC/wp-content/uploads/2019/05/2IHTC_Proceedings_Books.pdf. Accessed
tions to the literature. Also, the perceptual map created within the scope April 15, 2021.
of the study only reflected the perceptions found in the comments made [24] X. Xu, T. Meng, X. Cheng, Aspect-based extractive summarization of online
up to April 23, 2021; therefore, it is not indefinite. For this reason, it reviews, paper presented, in: Proceedings of the 2011 ACM Symposium on Applied
Computing, 21 March, Taiwan, 2011. Retrieved from, https://dl.acm.org/doi/
means that future changes in consumer perceptions about the brand will
proceedings/10.1145/1982185?tocHeading=heading40. Accessed May 13, 2021.
also alter the values in perception maps and radar charts. [25] M. Hu, B. Liu, Mining and Summarizing Customer Reviews paper presented, in:
As a final point, the cell phone models selected for the study are the Proceedings of the tenth ACM SIGKDD International Conference on Knowledge
“flagship” models that brands launch simultaneously with their com­ Discovery and Data Mining, August, New York, USA, 2004. Retrieved from, https
://www.cs.uic.edu/~liub/publications/kdd04-revSummary.pdf. Accessed
petitors every year. However, global brands offer many other cell phone Febryary 25, 2021.
models throughout the year. The analysis of consumer comments on
these different models would also be useful to businesses.

10
M.K. Yılmaz and H.T. Altunay Telematics and Informatics Reports 11 (2023) 100094

[26] T. Hardeniya, D.A. Borikar, Dictionary based approach to sentiment analysis-a [36] M. Polat, Y. Akan, Akıllı telefon piyasasında firmalar arasındaki rekabetin stratejik
review, International Journal of Advanced Engineering, Management and Science olarak incelenmesi: oyun teorisi kapsamında uygulamalı bir çalışma, Igdir
2 (5) (2016), 239438. University Journal of Social Sciences 4 (2020) 677–699.
[27] W. Medhat, A. Hassan, H. Korashy, Sentiment analysis algorithms and applications: [37] Statista (2021). Number of smartphones sold to end users worldwide from 2007 to
a survey, Ain Shams Engineering Journal 5 (2014) 1093–1113. 2021 Retrieved from https://www.statista.com/statistics/263437/global-smart
[28] M. Taboada, J. Brooke, M. Tofiloski, K. Voll, M. Stede, Lexicon-based methods for phone-sales-to-end-users-since-2007/ Accessed December 5, 2021.
sentiment analysis, Computational linguistics 37 (2) (2011) 267–307. [38] R. Ahuja, A. Chug, S. Kohli, S. Gupta, P. Ahuja, The impact of features extraction on
[29] Q. Rajput, S. Haider, S. Ghani, Lexicon-based sentiment analysis of teachers’ the sentiment analysis, Procedia Comput Sci 152 (2019) 341–348.
evaluation, Applied Computational Intelligence and Soft Computing 3 (2016) [39] Z. Yun-tao, G. Ling, W. Yong-cheng, An improved TF-IDF approach for text
1–12. classification, Journal of Zhejiang University-Science 6 (1) (2005) 49–55.
[30] A.J.T. Lee, F.C. Yang, C.H. Chen, C.S. Wang, C.Y. Sun, Mining perceptual maps [40] C. Sun, L. Huang, X. Qiu, Utilizing BERT for aspect-based sentiment analysis via
from consumer reviews, Decis Support Syst 82 (2015) 12–25. constructing auxiliary sentence, in: Proceedings of the 2019 Conference of the
[31] Y.M. Kızılkaya, Duygu Analizi ve Sosyal Medya Alanında Uygulama, Social North American Chapter of the Association for Computational Linguistics: Human
Sciences Institute, Bursa University, Bursa, 2018. Language Technologies, Volume 1 (Long and Short Papers), pages 380–385,
[32] A.S.A. Daabes, F.F. Kharbat, Customer-based perceptual map as a marketing Minneapolis, Minnesota, Association for Computational Linguistics, 2019.
intelligence source, International Journal of Economics and Business Research 13 [41] K. Srividya, A.M. Sowjanya, Aspect based sentiment analysis using pos tagging and
(4) (2017) 360–379. tfidf, International Journal of Engineering and Advanced Technology (IJEAT) 8 (6)
[33] K. Keller, Strategic Brand Management: Building, Measuring and Managing Brand (2019) 1960–1963.
Equity, Prentice-Hall, Upper Saddle River. NJ, 2008. [42] A. Uçan, Otomatik Duygu Sözlüğü Çevirimi ve Duygu Analizinde Kullanımı,
[34] A.C. Chang, C.V. Trappey, A.J. Trappey, L.W. Chen, Web mining customer Institute Of Science, Hacettepe University, Ankara, 2014.
perceptions to define product positions and design preferences, International [43] F. Sağlam, B. Genç, H. Sever, Extending a sentiment lexicon with synonym-
Journal on Semantic Web and Information Systems (IJSWIS) 16 (2) (2020) 42–58. antonym datasets: sWNetTR++, Turkish Journal of Electrical Engineering and
[35] B. Yazıcı, Yeni lüks kavramı bağlamında y kuşağı ile evrilen tüketim ve y kuşağının Computer Sciences 27 (2019) 1806–1820.
lüks kavramına bakışı, Kocaeli University Journal of Social Sciences 36 (2018)
95–112.

11