Aspect Based Sentiment Analysis on Online Review Data

to Predict Corporate Reputation

R. E. Loke

and W. Reitter

Centre for Market Insights, Amsterdam University of Applied Sciences, Amsterdam, The Netherlands

Keywords: Aspect Based Sentiment Analysis (ABSA), Machine Learning, Natural Language Processing, Scraping.

Abstract: Corporate reputation is an intangible resource that is closely tied to an organization’s success but measuring

it and to derive actions that can improve the reputations can be a long and expensive journey for an

organization. In the available literature, corporate reputation is primarily measured through surveys, which

can be time and cost intensive. This paper uses online reviews on the web as the source for a machine-learning

driven aspect-based sentiment analysis that can enable organizations to evaluate their corporate reputation on

a fine-grained level. The analysis is done unsupervised without organizations needing to manually label

datasets. Using the insights generated through the analysis, on one hand, organizations can save costs and

time to measure corporate reputation, and, on the other hand, it provides an in-depth analysis that splits the

overall reputation into multiple aspects, with which organizations can identify weaknesses and in turn improve

their corporate reputation. Therefore, this research is relevant for organizations aiming to understand and

improve their corporate reputation to achieve success, for example, in form of financial performance, or for

organizations that help and consult other organizations on their journeys to increased success. Our approach

is validated, evaluated and illustrated with Trustpilot review data.

1 INTRODUCTION

One of the major objectives of strategic business

management is understanding driving factors of

organizational performance (Crook et al., 2008). A

possibility to evaluate and understand the

heterogeneity of firms is the resource based view, that

analyzes organizational resources and capabilities

(Eloranta & Turunen, 2015). Those resources can be

tangible or intangible (Kamasak, 2017) and according

to researchers (Kor & Mesko, 2013; Molloy & Barney,

2015) intangible resources are considered as the most

likely sources to an organizations success. Resources

need to be valuable, rare, inimitable and not

substitutable (Y. Lin & Wu, 2014). As such, according

to Brahim & Arab (2011) intangible resources are most

difficult to imitate and substitute and it can be argued,

that these are the most valuable and rarest.

A main intangible resource of an organization that

is increasingly receiving attention is corporate

reputation (Wepener & Boshoff, 2015). According to

Schwaiger et al. (2011) corporate reputation is “the

ultimate determinant of competitiveness”. However,

measuring corporate reputation is a challenge for

https://orcid.org/0000-0002-7168-090X

researches and businesses alike. Corporate reputation

is seen as a factor that can explain the performance of

a business, with Firestein (2006) arguing that corporate

reputation is the strongest aspect of a company’s

sustainability. The relevance of corporate reputation

has been stated numerously throughout the years, as

such Abratt & Kleyn (2012) concluded that corporate

reputation is a strategic resource to create competitive

advantage. Furthermore, Vig et al. (2017) argue that

corporate reputation can be a significant factor to

predict financial performance. A reason for the

possible correlation of corporate reputation and

financial performance can be the effect of corporate

reputation on the customers and its purchase intention.

Keh & Xie (2009) observed a relation between

corporate reputation and customer trust, that leads to

increased purchase intention and willingness to pay.

The positive effect corporate reputation can have

on a customer is becoming increasingly relevant in an

environment where businesses need to become more

customer centric for long term business success and

profitability (Roy & Shekhar, 2010), making the

corporate reputation more important than ever. As

such, it can be said that understanding and improving

Loke, R. and Reitter, W.

Aspect Based Sentiment Analysis on Online Review Data to Predict Corporate Reputation.

DOI: 10.5220/0010607203430352

In Proceedings of the 10th International Conference on Data Science, Technology and Applications (DATA 2021), pages 343-352

ISBN: 978-989-758-521-0

 2021 by SCITEPRESS – Science and Technology Publications, Lda. All r ights reserved

343

corporate reputation is an essential part for an

organization’s success, but measuring it is a challenge

for organizations.

A method to measure and evaluate corporate

reputation is doing surveys, which is used by several

researchers (Cintamür & Yüksel, 2018; Fombrun et al.,

2015; Puncheva‐Michelotti & Michelotti, 2010;

Sequeira et al., 2015; Wepener & Boshoff, 2015) to

measure corporate reputation. It can be argued that in

order to receive the most reliable results possible, more

data is needed and albeit the method of conducting

surveys allows for targeted and in depth information, it

can be time consuming when trying to collect as much

data as possible. Furthermore, the relation between

time and cost of a project (Babu & Suresh, 1996) can

lead to higher cost involved with more time spent on

conducting the survey. Therefore, it can be said that the

main problem of organizations for evaluating and

measuring corporate reputation, which is said to impact

an organization’s success, is the time and cost intensive

methodology of collecting primary data through

surveys. As such, a goal of an organization is to find a

time and cost efficient way of measuring corporate

reputation that can provide them added value.

To find an efficient way to tackle this problem,

digital aspects should be highlighted, as according to

Deloitte (2013), especially in the digital age, customers

are increasingly more using digital touchpoints to

resolve issues or get in contact with businesses. Dang

& Pham (2020) further stress this relevancy by

concluding that focusing on customers online is an

essential part for businesses. As such, with the goal of

finding a time and cost efficient way to evaluate

corporate reputation and the increasing relevance of

online aspects, another source to evaluate and improve

corporate reputation is online reviews, where

customers leave information such as feedback, positive

or negative, recommendations or other valuable

insights. According to Mayzlin et al. (2014), user

generated online reviews are an important resource for

consumers in their purchase decision. The main

advantage of analyzing online reviews compared to

conducting surveys is, that the reviews are already

available online and no time and cost intensive surveys

need to be carried out to collect the data. Furthermore,

it can be argued that online reviews can offer more data

than surveys, as there might be more customers that

leave a review than customers who are ready to fill out

a survey. It can be said that online reviews are a

favorable way of measuring corporate reputation, but

the questions of how they are analyzed needs to be

answered as well.

To derive corporate reputation Chung et al. (2019)

used sentiment polarity analysis on social media tweets

about several companies. It can be argued that a tweet

about a company is the same as an online review about

a company. However, the main problem of sentiment

polarity or sentiment analysis is that these analyses are

done on an overall level and are not in depth enough to

conclude concrete recommendations. As such, even

when an organization is perceived overall positively

there can be aspects of the organization that are seen as

negatively that need to be improved. To tackle this

problem, Chung et al. (2019) suggest that their main

limitation, of their analysis not being in depth enough,

can be addressed by ABSA, which can predict the

sentiment of various aspects. The need and relevancy

of ABSA is also backed up by Jebbara & Cimiano

(2016), who argue that sentiment analysis needs to be

done on a more fine-grained level and stress the

importance of analyzing online reviews. The authors

propose an ABSA that can extract sentiments

expressed towards aspects from the text and thus, can

detect multiple opinions in a single review. However,

ABSA faces another time related problem, which is the

manual labeling, research that has provided great

ABSA results (Araque et al., 2017; Chen et al., 2017)

needed a significant amount of time for manually

labelling data. This will also limit the amount of data,

as there is only a limited amount of data a human can

label in a short time. As such, conducting ABSA on

online reviews is a way to evaluate corporate

reputation cost and time efficiently for an organization,

but it needs to be done in a way that does not need

manual labeling.

In the following sections, we describe relevant

conceptual models, give details in our processing

methodology, and conclude our work.

2 THEORETICAL FRAMEWORK

This section is intended to give an overview of the

current state of research that is needed to understand

corporate reputation.

2.1 Current State of Corporate

Branding

To understand corporate reputation, the origin needs to

be examined first. Corporate reputation is part of

corporate branding, which can be described as a

multidisciplinary field that has proved its usefulness in

scientific and business environments (Biraghi &

Gambetti, 2015). Corporate branding is currently

experiencing three major shifts: (1) Brand strategy

shifting its focus from products to an organizational

DATA 2021 - 10th International Conference on Data Science, Technology and Applications

344

perspective (Balmer, 2001); (2) Shifting from

marketing to corporate strategy (Abratt & Kleyn,

2012); (3) Focusing on a stakeholder-centric view

where the corporate brand is in an ongoing dialogue

with the stakeholders (Melewar et al., 2012).

According to Melewar et al. (2012) the third shift

is to move away from the traditional perspective and to

incorporate a dynamic view on corporate branding. In

this paper and for corporate reputation, the third shift is

most relevant. Here, corporate branding is seen as an

collaborative, relational and social process between the

company and its stakeholders (Cornelissen et al.,

2012). Melewar et al. (2012) further stress the fact that

this view of corporate branding sees the brand as a

vehicle that helps with the interaction between the

company and its environment. Koporcic & Halinen

(2018) assume that corporate brands are formed in the

minds of the individual people and that the image is

constantly being refined and changed. According to the

authors, the key concepts that are part of corporate

branding are corporate identity and corporate

reputation. According to Podnar (2015, p. 29)

corporate identity characteristics are real and constant

attributes, whereas corporate reputation can be

different, depending on the view of the observer.

2.2 Current Research and Definition on

Corporate Reputation

Research has underpinned the relevance of corporate

reputation and it is seen as a strategic resource to

create competitive advantage (Abratt & Kleyn, 2012).

Early on, corporate reputation was defined as a

concept, that is a personal impression of the perceived

entity (Shee & Abratt, 1989). In more recent

researches on corporate reputation, Agarwal et al.

(2015) affirm the view that corporate reputation is an

unobservable construct that only exists in the minds

of the stakeholders.

Definitions of corporate reputation can be found

in a growing body of literature. There is currently no

singular definition or adaptation of corporate

reputation and some interpretations overlap or

contradict. Table 1 displays several definitions of

corporate reputation before deriving the main

definition that will be used in this paper.

What is consistent throughout the definitions is

that corporate reputation is a perception of an

organization that a stakeholder has, that is based on

past impressions. As such, this paper’s definition of

corporate reputation is:

“A sociocognitive construct consisting of the

organization’s perception that a stakeholder has, that

is built through past impressions”

Table 1: Definitions of corporate reputation.

Literature Definition

Gotsi &

Wilson,

2001

Stakeholder’s overall evaluation of a

company based on experiences,

communication and symbols that provide

information about company’s action, over

time. The evaluation is also compared with

rivals.

Chun,

2005

Umbrella referring to the cumulative

impressions of the different stakeholders of

what the organization stands for and what it

is associated with.

Barnett et

al., 2006

Collective judgement of observers of a

corporation, based on assessments of

financial, social and environmental impacts,

over time.

Walker,

2010

Relatively stable, issue specific

representation of a company’s past actions

and future prospects compared to the

standard. Takes time to build and can remain

stable once built.

Rindova et

al., 2010

Sociocognitive construct that is

characterized by quality and prominence to

determine value as an intangible asset

contributin

to com

etitive advanta

Lange et

al., 2011

Objective reality for the organization that is

sub

ectivel

created b

outside observers.

The goal of this paper is to measure corporate

reputation and several researchers have tried to

measure corporate reputation and find relevant

dimensions. Table 2 shows the proposed variables of

researchers to measure corporate reputation.

Table 2: Dimensions of corporate reputation.

Literature Dimensions Methodolo

Wepener &

Boshoff,

2015

Emotional appeal, social

engagement, corporate

performance, good

er, service

oints

Online

survey

Cintamür &

Yüksel,

2018

Financial performance,

customer orientation, social

and environmental

res

onsibilit

, trust

Face-to-face

survey

Fombrun et

al., 2015

Products, innovation,

workplace, governance,

citizenship, leadership,

erformance

Survey

Puncheva‐

Michelotti

Michelotti,

2010

Management excellence,

social responsibility,

customer value, economic

performance, patriotic

appeal, consumer impact,

emotional appeal, credibilit

Survey

Sequeira et

al., 2015

Enterprise agreeableness,

competence, commitment,

ruthlessness

Survey

Aspect Based Sentiment Analysis on Online Review Data to Predict Corporate Reputation

345

The problem of these dimensions is that they are

intended to be used through surveys and not ABSA

on online reviews. As such, emotional appeal,

management excellence, social responsibility and

other dimensions are not applicable to this research

where online reviews serve as the source of the

analysis. Those dimensions are more applicable to

surveys where detailed and targeted questions can be

formed, however online reviews are based on what

customers freely write. Therefore, it can be expected

that those detailed and specific information cannot be

found in online reviews, thus, most dimensions are

not applicable to the nature of this research.

This paper’s dimensions of corporate reputation

will be based on Fombrum et al. (2015), due to the

detailed explanations of the dimensions and attributes

such as quality or value that can be more applicable

to online reviews than the other dimensions shown in

Table 2. Furthermore, Fombrum et al. (2015) propose

two approaches for measuring corporate reputation,

thus there is a better option of validating the results

by comparing the results of dimensions 1 and 2; see

Figure 1. It can be compared with a recent framework

proposed by Wepener & Boshoff (2015). The

frameworks have similarities such as performance or

employee dimensions. However, the framework of

this paper has a two-way approach to measure

corporate reputation. Furthermore, the framework for

this paper has more dimensions/attributes to measure

corporate reputation on. This fact can be a bigger

advantage when using online reviews as compared to

surveys. In surveys, it can be said that the respondent

can lose attention and motivation when the survey is

longer, however when analyzing online reviews, this

hurdle is not given, and a more extensive approach

can be chosen. By having a framework to measure

corporate reputation on, this can be applied onto

ABSA. Where each dimension and attribute are an

aspect that gets analyzed on its corresponding

sentiment. This approach can be seen as an extension

to current literature and research (Caviggioli et al.,

2020; Chung et al., 2019) where corporate reputation

is increasingly more often measured through social

media or online reviews. However, those researchers

have not applied their analysis on frameworks such as

the proposed one of this paper. As such, this paper

aims to conduct ABSA on a theoretical framework

that is usually applied in surveys. This can lead to a

more reliable result of corporate reputation that is

backed up by a theoretical framework.

However, it can be seen that the dimensions of the

proposed framework are diverse in their nature, some

have a customer orientation, such as “products”

whereas others have an employee orientation such as

Figure 1: Framework of corporate reputation used in this

paper.

“workplace”. With online reviews being the source of

the analysis, it can be argued that a singular online

review source such as Trustpilot won’t provide

reviews that can cover all of the dimensions.

However, the preliminary results have shown that all

dimensions are covered and mentioned in reviews,

albeit to a varying degree from the online review

source Trustpilot (more in section 3.2). As such, all

dimensions of Fombrum et al. (2015) are kept in the

analysis, because having more aspects to analyze

won’t negatively affect the results and even less likely

to be mentioned dimensions are kept in to provide a

full picture. Furthermore, the attributes and

dimensions of the framework were adapted to the

customer language of online reviews and how a

customer would write these terms in a review. E.g.

“value” was translated to “price” as this word can be

seen as more likely to be used in a review or “Equal

DATA 2021 - 10th International Conference on Data Science, Technology and Applications

346

opportunities” changed to “equality”. This was done

based on the definitions Fombrum et al. (2015)

provided.

According to Fombrum et al. (2015), corporate

reputation can be measured through the extensive

approach of measuring all attributes related to the

dimensions of “dimensions 1”. However, corporate

reputation can also be measured alongside the

dimensions of “dimensions 2”. To evaluate validity

of the models, the result of each dimension of

“dimensions 1” was compared to the overall result of

each dimension. Furthermore, the overall results of

“dimensions 2” was compared to the overall result of

“dimensions 1”.

It can be concluded that corporate reputation and

identity both stem from corporate branding, with the

identity being a real and constant construct whereas

reputation is subjective and depends on the view of

the observer. To measure this view of corporate

reputation, it will be measured alongside the 34

attributes and dimensions of this paper’s framework

(see Figure 1).

3 METHODOLOGY

This research is focusing on a mixed-method research

as described by Saunders et al. (2015), which will be

explained in the following.

3.1 Primary Research

For this paper, online reviews needed to be collected

that can be analyzed in the later stages using a

machine learning model that conducts ABSA.

Karalevicius et al. (2018) who carried out sentiment

analysis tackled the primary research by using a

scraper that can collect social media and online

review data. For this research, a Python scraper from

CMI HVA was provided in order to collect the data.

For a platform to scrape the reviews from and to

analyze using sentiment analysis Vankka et al. (2019)

opted for Trustpilot, because of the accumulation of

companies with their respective reviews. As such

Trustpilot was chosen for scraping the online reviews

from. Finally, for this paper, the companies to extract

the reviews from and to conduct ABSA on are Nike

and Adidas, which amounts to over 2.000 scraped

online reviews. Those companies are chosen, because

they are prominent direct competitors and this allows

to compare the results of the analysis and derive

conclusions, which can also be relevant for

organizations that want to use the proposed work to

compare with their competition. Furthermore, these

companies have both a balanced amount of positive

and negative reviews, which allows to detect

weaknesses, where reviewers are unsatisfied and

clear improvement points can be deduced.

Figure 2: Starting dataset.

3.2 Secondary Research

Following the scraping of the review data, the online

reviews were analyzed using secondary research. All

the analyses were conducted using Python. The

process is according to de Kok et al. (2018) who

conduct ABSA and follow the same steps, with the

exception that they used a pre annotated dataset

where the aspects where defined in the dataset, for

this paper, the additional step of aspect term

extraction has to be added. De Kok et al. (2018) used

readily available restaurant review data that they did

not have to scrape and conducted ABSA on this data.

With their result they could evaluate which aspect

(e.g. food, ambience, service) the reviewee is

mentioning and if they have a positive or negative

sentiment towards the aspect.

Figure 3: Cleaned dataset.

1 Cleaning Data:

First the dataset was loaded; Figure 2 shows a

snapshot.

In the following step, the column “stars” was

stripped of all unnecessary characters until only the

number of stars awarded by the reviewer was left.

Then, the columns title and text were combined and

these reviews were cleaned and lemmatized using

NLTK, which is also according to de Kok et al.

(2018). NLTK is one of the most popular and widely

used libraries in the field of natural language

processing (NLP), because of its simplicity and

effectiveness (Hardeniya, 2015, p. 3). The end result

is a dataset with a text that was stripped of all non-

letters, words with not much value such as “the” and

the number of stars awarded (Figure 3).

Aspect Based Sentiment Analysis on Online Review Data to Predict Corporate Reputation

347

2 Aspect Term Extraction:

Figure 4: Example dependency parsing.

In this part, only the aspects were extracted. This is

an extra step that de Kok et al. (2018) did not go

through as their dataset was pre labeled, thus the

aspects were already extracted beforehand. For

example, in a review such as “The delivery went

smooth” the aspect is delivery. These aspects need to

be extracted to assign them the corresponding

sentiment at the later stages. This was done using the

dependency parser spaCy (Figure 4), which is

regarded as the industrial strength natural language

processing (Srinivasa-Desikan, 2018, p. 33) and

researchers such as Bandhakavi et al. (2018) rely on

spaCy for aspect extraction. Poria et al. (2016)

propose a deep convolutional neural network,

however supervised models are not fitting for this

paper, because manually labelling the dataset is too

time costly.

Using spaCy it is possible to extract the aspect and

the corresponding sentiment, which is mostly an

adjective. Figure 5 illustrates how the aspects are

extracted.

Even though spaCy is an often used and efficient

way of extracting aspects, the major downside is that

is only sees nouns as an aspect. As such, in a sentence

like “It was delivered quickly” the reviewer is mainly

talking about the delivery, but because it is expressed

as a verb, spaCy won’t detect it as an aspect.

However, it can be said that most aspects the reviewer

is talking about are articulated as a noun and therefore

this drawback won’t affect the results too negatively.

Figure 5: Example aspect extraction.

3 Training the Model:

This part is needed for the following sentiment

analysis and is mostly used through Scikit-learn. This

module is regarded as the state-of-the-art machine

learning algorithm for supervised and unsupervised

problems (Pedregosa et al., 2011). Using the machine

learning module Scikit-learn, a Support Vector

Machine (SVM) can be trained. De Kok et al. (2018)

trained and used an SVM for aspect detection. This

trained model creates a feature vector of values for

every instance that will be classified and the model is

taught using training data to interpret these values. In

the case of this paper, the accuracy of the SVM was

compared to a Naïve Bayes model, with the latter

performing better, due to an SVM generally needing

more data to train. This trained Naïve Bayes can take

in each review and assign each sentence to one or

more aspects that were extracted with spaCy.

Therefore, spaCy lists all the aspects that are available

and the Naïve Bayes classifies each review sentence

into one or more of those extracted aspects. E.g.

“Delivery was fast. Unfortunately, the product was

broken” the model will now assign the first sentence

to “delivery” and the second sentence to “product”. In

the later stages, the model will look for the sentiment

within each assigned sentence.

4 Sentiment Analysis:

For this part of analysis, first a few modules were

loaded that are according to Marrese-Taylor et al.

(2014), Ding et al. (2008), Rantanen et al. (2019) and

Urologin (2018).

First, opinion lexicon by Hu & Liu (2004). These

researchers defined a list of around 6.800 of positive

and negative words designed especially for sentiment

analysis and this lexicon is widely adopted in the NLP

community. The researchers defined the lexicon by

mining and summarizing customer reviews, then

extracting the opinion sentences and predicting

whether the opinion is positive or negative. Using this

lexicon, the code can detect if a word that an aspect is

associated with is either positive or negative. This

lexicon was used in researches such as of Marrese-

Taylor et al. (2014) or Ding et al. (2008) where it was

needed for aspect based opinion mining. Another

possible opinion lexicon could be SentiWordNet that

automatically assigns a score to a term, however

according to Na et al. (2009), this lexicon does not

handle the opinion scoring problem and therefore, the

lexicon of Hu & Liu was chosen for its simplicity.

Figure 6: Example replacing pronoun.

Second, Google’s pretrained word2vec model.

This model was published by researchers of Google

DATA 2021 - 10th International Conference on Data Science, Technology and Applications

348

(Mikolov et al., 2013) that trained the model on 1.6

billion words and it can recognize similarities of

words. Thus, it is able to find words that are similar

to the aspects and group them. This model is also

heavily utilized in textual analysis with researchers

such as Rantanen et al. (2019) using it to compute

corporate reputation from social media comments.

Another state of the art word embedding method is

Global Vectors for Word Representation (GloVe)

which was published by researchers of Stanford

(Pennington et al., 2014). However, according to the

study of Lin et al. (2015) where both wor2vec and

GloVe were compared, word2vec performed better.

Third, neural coreference model is used to

identify pronouns and replace them, as pronouns do

not add any sentiment or weight. This pre trained

neural network model can detect dependencies within

a sentence and for example understand which terms

refer to each other. According to Lee et al.(2017), this

model outperforms all previously related work.

Replacing pronouns is an important step according to

Urologin (2018), as pronouns are just placeholders

for proper nouns and this could affect the scoring of

the sentiments. An example would be “The product is

great, however it is smelling bad”. The pronoun

would be “it” which refers to “product”, therefore the

sentiment “bad” would be assigned to the term “it”

and not “product” and using the neural coreference

model, “it” will be replaced with “product”. Figure 6

is an example of how a pronoun is replaced.

In the next steps the functions are defined that will

make up the analysis. The functions consist of first

checking for similarities of the words. The maximum

similarity is 1, meaning it is the same word and it will

be checked if the term is similar to the aspect. The

threshold is set at 0.3, which garnered results where

perceived similar words are detected where the

penalty is not too high nor too low. E.g. “product”

and “brand” have a similarity of 0.50 thus brand is

similar to the aspect product and will be later assigned

to that aspect. Now all the predefined aspects of

Figure 1 will receive sentiment scores. First, it will be

checked if the sentiment word like “great” is in the

opinion lexicon and a sentiment score of +/- 1 will be

given out accordingly. If there is an adjective

modifier such as “incredibly”, the sentiment will

receive a greater weight of +/- 1.5. Negations such as

“not good” will have the score flipped. Figure 7 gives

an insight to the scoring.

Now the scores for all the aspects of Figure 1 will

be accumulated and the scores of the synonyms will

also be calculated. For example if the aspect

“product” has been mentioned positively 10 times it

will have 10 points, if the synonym “brand” has been

mentioned positively 5 times the 5 points will go to

the aspect “product”, because according to word2vec

the similarity is above the threshold of 0.3 and the

aspect “product” will have overall 15 points.

Figure 7: Example sentiment scoring.

Figure 8: ABSA output for corporate reputation.

5 Visualization and Evaluation:

The last step is visualizing the results using the results

garnered from previous step. Using tables, bar graphs

or pie charts, the result of the sentiment analysis can

be plotted to give an overview of the results.

Matplotlib and Plotly were used of this step. Figures

8-10 are example output of the analysis applied to

Adidas and Nike. Note that all aspects and scores

directly relate to the relevant variables and

dimensions in the corporate reputation model of

Fombrun et al. (2015) that was shown in Figure 1.

Figure 9: Aspects mentioned for corporate reputation.

The pos/neg ratio of aspects was calculated as follows

in order to measure corporate reputation overall as

well as within its dimensions and variables:

𝑃𝑜𝑠 𝑟𝑎𝑡𝑖𝑜  𝑆𝑢𝑚 𝑝𝑜𝑠 𝑝𝑜𝑖𝑛𝑡𝑠 𝑝𝑒𝑟 𝑎𝑠𝑝𝑒𝑐𝑡

/𝑆𝑢𝑚 𝑝𝑜𝑠

 𝑛𝑒𝑔 𝑝𝑜𝑖𝑛𝑡𝑠 𝑝𝑒𝑟 𝑎𝑠𝑝𝑒𝑐𝑡

(1)

Aspect Based Sentiment Analysis on Online Review Data to Predict Corporate Reputation

349

𝑁𝑒𝑔 𝑟𝑎𝑡𝑖𝑜  𝑆𝑢𝑚 𝑛𝑒𝑔 𝑝𝑜𝑖𝑛𝑡𝑠 𝑝𝑒𝑟 𝑎𝑠𝑝𝑒𝑐𝑡

/𝑆𝑢𝑚 𝑝𝑜𝑠

 𝑛𝑒𝑔 𝑝𝑜𝑖𝑛𝑡𝑠 𝑝𝑒𝑟 𝑎𝑠𝑝𝑒𝑐𝑡

(2)

The overall relative positive sentiment score on

Nike’s corporate reputation was higher than that of

Adidas on this dataset. We carefully checked on this

for evaluation purposes, and it was in line with the

average star rating of Trustpilot reviewers that was

also available in our scraped dataset and that is

publicly visible on the website.

Figure 10: Aspect comparison for corporate reputation.

3.3 Validity

In order to validate our approach, we compared the

overall scores obtained on both sides in the corporate

reputation model (see dimensions 1 and dimensions 2

in Figure 1). On the left side in the corporate

reputation model (dimensions 1 In Figure 1), we also

compared, for each dimension, the average score on

all attributes per dimension with the direct score on

the dimension; e.g. the average score on “High

quality”, “Good value”, “Stands behind” and “Meets

customer needs” is compared with that on “Products”.

For Nike and Adidas together, the difference in score

between dimensions 1 and dimensions 2 is 4%. The

difference in score between averaged attribute scores

and direct dimension scores is 7%. Despite that the

scraped dataset is not extremely large, these low

scores seem to imply the internal validity of our

approach. Of course, such validity was already

reported by Fombrun et al. (2015) based on their

research with surveys. However, the important

contribution that we aim to make in the corporate

reputation literature is to be able to claim it for online

review data on the public web.

4 CONCLUSION

In summary, with our approach, companies are

enabled to assign to each aspect, that can be flexibly

defined to corporate repution or to another relevant

construct, a sentiment score derived from online

reviews on the web.

ACKNOWLEDGEMENTS

This paper has been inspired on the MSc master

project of Wolfgang Reiter who was involved via the

master Digital Driven Business at HvA. Thanks go to

Jesse Weltevreden for providing some useful

suggestions to an initial version of this manuscript.

Rob Loke is assistant professor data science at

CMIHvA.

REFERENCES

Abratt, R., & Kleyn, N. (2012). Corporate identity,

corporate branding and corporate reputations:

Reconciliation and integration. European Journal of

Marketing, 46(7/8), 1048–1063.

Agarwal, J., Osiyevskyy, O., & Feldman, P. M. (2015).

Corporate Reputation Measurement: Alternative Factor

Structures, Nomological Validity, and Organizational

Outcomes. Journal of Business Ethics, 130(2), 507–

507.

Araque, O., Corcuera-Platas, I., Sánchez-Rada, J. F., &

Iglesias, C. A. (2017). Enhancing deep learning

sentiment analysis with ensemble techniques in social

applications. Expert Systems with Applications, 77,

236–246.

Babu, A. J. G., & Suresh, N. (1996). Project management

with time, cost, and quality considerations. European

Journal of Operational Research, 88(2), 320–327.

Balmer, J. M. T. (2001). Corporate identity, corporate

branding and corporate marketing ‐ Seeing through the

fog. European Journal of Marketing, 35(3/4), 248–291.

Bandhakavi, A., Wiratunga, N., Massie, S., & Luhar, R.

(2018). Context Extraction for Aspect-Based Sentiment

Analytics: Combining Syntactic, Lexical and Sentiment

Knowledge. In M. Bramer & M. Petridis (Eds.),

Artificial Intelligence XXXV (Vol. 11311, pp. 357–

371). Springer International Publishing.

Barnett, M. L., Jermier, J. M., & Lafferty, B. A. (2006).

Corporate Reputation: The Definitional Landscape.

Corporate Reputation Review, 9(1), 26–38.

Biraghi, S., & Gambetti, R. C. (2015). Corporate branding:

Where are we? A systematic communication-based

inquiry. Journal of Marketing Communications, 21(4),

260–283.

Brahim, H. B., & Arab, M. B. (2011). The effect of

intangible resources on the economic performance of

the firm. Journal of Business Studies Quarterly, 3(1),

36–59.

Caviggioli, F., Lamberti, L., Landoni, P., & Meola, P.

(2020). Technology adoption news and corporate

reputation: Sentiment analysis about the introduction of

DATA 2021 - 10th International Conference on Data Science, Technology and Applications

350

Bitcoin. Journal of Product & Brand Management.

Chen, T., Xu, R., He, Y., & Wang, X. (2017). Improving

sentiment analysis via sentence type classification

using BiLSTM-CRF and CNN. Expert Systems with

Applications, 72, 221–230.

Chun, R. (2005). Corporate reputation: Meaning and

measurement. International Journal of Management

Reviews, 7(2), 91–109.

Chung, S., Chong, M., Chua, J. S., & Na, J. C. (2019).

Evolution of corporate reputation during an evolving

controversy. Journal of Communication Management,

23(1), 52–71.

Cintamür, İ. G., & Yüksel, C. A. (2018). Measuring

customer based corporate reputation in banking

industry: Developing and validating an alternative

scale. International Journal of Bank Marketing, 36(7),

1414–1436.

Cornelissen, J., Thøger Christensen, L., & Kinuthia, K.

(2012). Corporate brands and identity: Developing

stronger theory and a call for shifting the debate.

European Journal of Marketing, 46(7/8), 1093–1102.

Crook, T. R., Ketchen, D. J., Combs, J. G., & Todd, S. Y.

(2008). Strategic resources and performance: A meta-

analysis. Strategic Management Journal, 29(11), 1141–

1154.

Dang, T. T., & Pham, A. D. (2020). What make banks’

front-line staff more customer oriented? The role of

interactional justice. International Journal of Bank

Marketing, ahead-of-print(ahead-of-print).

de Kok, S., Punt, L., van den Puttelaar, R., Ranta, K.,

Schouten, K., & Frasincar, F. (2018). Review-

aggregated aspect-based sentiment analysis with

ontology features. Progress in Artificial Intelligence,

7(4), 295–306.

Deloitte. (2013). The Digital transformation of customer

services: Our point of view.

https://www2.deloitte.com/content/dam/Deloitte/nl/Do

cuments/technology/deloitte-nl-paper-digital-

transformation-of-customer-services.pdf

Ding, X., Liu, B., & Yu, P. S. (2008). A holistic lexicon-

based approach to opinion mining. Proceedings of the

International Conference on Web Search and Web Data

Mining - WSDM ’08, 231.

Eloranta, V., & Turunen, T. (2015). Seeking competitive

advantage with service infusion: A systematic literature

review. Journal of Service Management, 26(3), 394–

425.

Firestein, P. J. (2006). Building and protecting corporate

reputation. Strategy & Leadership, 34(4), 25–31.

Fombrun, C. J., Ponzi, L. J., & Newburry, W. (2015).

Stakeholder Tracking and Analysis: The RepTrak®

System for Measuring Corporate Reputation.

Corporate Reputation Review, 18(1), 3–24.

Gotsi, M., & Wilson, A. M. (2001). Corporate reputation:

Seeking a definition. Corporate Communications: An

International Journal, 6(1), 24–30.

Hardeniya, N. (2015). NLTK Essentials. Packt Publishing

Ltd.

Hu, M., & Liu, B. (2004). Mining and summarizing

customer reviews. Proceedings of the 2004 ACM

SIGKDD International Conference on Knowledge

Discovery and Data Mining - KDD ’04, 168.

Jaidka, K., Khoo, C. S. G., & Na, J. (2013). Literature

review writing: How information is selected and

transformed. Aslib Proceedings, 65(3), 303–325.

Jebbara, S., & Cimiano, P. (2016). Aspect-Based Sentiment

Analysis Using a Two-Step Neural Network

Architecture. In H. Sack, S. Dietze, A. Tordai, & C.

Lange (Eds.), Semantic Web Challenges (Vol. 641, pp.

153–167). Springer International Publishing.

Kamasak, R. (2017). The contribution of tangible and

intangible resources, and capabilities to a firm’s

profitability and market performance. European

Journal of Management and Business Economics,

26(2), 252–275.

Karalevicius, V., Degrande, N., & De Weerdt, J. (2018).

Using sentiment analysis to predict interday Bitcoin

price movements. The Journal of Risk Finance, 19(1),

56–75.

Keh, H. T., & Xie, Y. (2009). Corporate reputation and

customer behavioral intentions: The roles of trust,

identification and commitment. Industrial Marketing

Management, 38(7), 732–742.

Koporcic, N., & Halinen, A. (2018). Interactive Network

Branding: Creating corporate identity and reputation

through interpersonal interaction. IMP Journal, 12(2),

392–408.

Kor, Y., & Mesko, A. (2013). Dynamic managerial

capabilities: Configuration and orchestration of top

executives’ capabilities and the firm’s dominant logic.

Strategic Management Journal, 34(2), 233–244.

Lange, D., Lee, P. M., & Dai, Y. (2011). Organizational

Reputation: A Review. Journal of Management, 37(1),

153–184. https://doi.org/10.1177/0149206310390963

Lee, K., He, L., Lewis, M., & Zettlemoyer, L. (2017). End-

to-end Neural Coreference Resolution. ArXiv E-Prints,

arXiv:1707.07045.

Lin, W., Dai, H., Jonnagaddala, J., Chang, N., Jue, T. R.,

Iqbal, U., Shao, J. Y., Chiang, I., & Li, Y. (2015).

Utilizing different word representation methods for

twitter data in adverse drug reactions extraction. 2015

Conference on Technologies and Applications of

Artificial Intelligence (TAAI), 260–265.

Lin, Y., & Wu, L.-Y. (2014). Exploring the role of dynamic

capabilities in firm performance under the resource-

based view framework. Journal of Business Research,

67(3), 407–413.

Marrese-Taylor, E., Velásquez, J. D., & Bravo-Marquez, F.

(2014). A novel deterministic approach for aspect-

based opinion mining in tourism products reviews.

Expert Systems with Applications, 41(17), 7764–7775.

Mayzlin, D., Dover, Y., & Chevalier, J. (2014).

Promotional Reviews: An Empirical Investigation of

Online Review Manipulation. American Economic

Review, 104(8), 2421–2455.

Melewar, T. C., Gotsi, M., & Andriopoulos, C. (2012).

Shaping the research agenda for corporate branding:

Avenues for future research. European Journal of

Marketing, 46(5), 600–608.

Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., &

Aspect Based Sentiment Analysis on Online Review Data to Predict Corporate Reputation

351

Dean, J. (2013). Distributed Representations of Words

and Phrases and their Compositionality. In C. J. C.

Burges, L. Bottou, M. Welling, Z. Ghahramani, & K.

Q. Weinberger (Eds.), Advances in Neural Information

Processing Systems 26 (pp. 3111–3119). Curran

Associates, Inc.

Molloy, J. C., & Barney, J. B. (2015). Who Captures the

Value Created with Human Capital? A Market-Based

View. Academy of Management Perspectives, 29(3),

309–325.

Na, S.-H., Lee, Y., Nam, S.-H., & Lee, J.-H. (2009).

Improving Opinion Retrieval Based on Query-Specific

Sentiment Lexicon. In M. Boughanem, C. Berrut, J.

Mothe, & C. Soule-Dupuy (Eds.), Advances in

Information Retrieval (Vol. 5478, pp. 734–738).

Springer Berlin Heidelberg.

Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V.,

Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P.,

Weiss, R., Dubourg, V., Vanderplas, J., Passos, A.,

Cournapeau, D., Brucher, M., Perrot, M., & Duchesnay,

É. (2011). Scikit-learn: Machine Learning in Python.

Journal of Machine Learning Research, 12(85), 2825–

2830.

Pennington, J., Socher, R., & Manning, C. D. (2014).

GloVe: Global Vectors for Word Representation.

Empirical Methods in Natural Language Processing

(EMNLP), 1532–1543.

Podnar, K. (2015). Corporate communication, a marketing

viewpoint. Routledge.

Poria, S., Cambria, E., & Gelbukh, A. (2016). Aspect

extraction for opinion mining with a deep convolutional

neural network. Knowledge-Based Systems, 108, 42–

49.

Puncheva‐Michelotti, P., & Michelotti, M. (2010). The role

of the stakeholder perspective in measuring corporate

reputation. Marketing Intelligence & Planning, 28(3),

249–274.

Rantanen, A., Salminen, J., Ginter, F., & Jansen, B. J.

(2019). Classifying online corporate reputation with

machine learning: A study in the banking domain.

Internet Research, 30(1), 45–66.

Rindova, V. P., Williamson, I. O., & Petkova, A. P. (2010).

Reputation as an Intangible Asset: Reflections on

Theory and Methods in Two Empirical Studies of

Business School Reputations. Journal of Management,

36(3), 610–619.

Roy, S. K., & Shekhar, V. (2010). Dimensional hierarchy

of trustworthiness of financial service providers.

International Journal of Bank Marketing, 28(1), 47–64.

Saunders, M., Lewis, P., & Thornhill, A. (2015). Research

methods for business students (7th ed.).

Schwaiger, M., Raithel, S., Rinkenburger, R., Schloderer,

M., Burke, R. J., Martin, G., & Cooper, C. L. (2011).

Measuring the Impact of Corporate Reputations on

Stakeholder Behavior. In Corporate Reputation.

Managing Opportunities and Threats

(pp. 61–88).

Ashgate Publishing Limited.

Sequeira, N., da Silva, R. V., Ramos, M., & Alwi, S. F. S.

(2015). Measuring Corporate Reputation in B2B

Markets: The Corporate Personality Adapted Scale.

IUP Journal of Knowledge Management, 13(3), 31–63.

Business Premium Collection; Engineering Database.

Shee, P. S. B., & Abratt, R. (1989). A new approach to the

corporate image management process. Journal of

Marketing Management, 5(1), 63–76.

Srinivasa-Desikan, B. (2018). Natural Language

Processing and Computational Linguistics: A practical

guide to text analysis with Python, Gensim, spaCy, and

Keras. Packt Publishing Ltd.

Urologin, S. (2018). Sentiment Analysis, Visualization and

Classification of Summarized News Articles: A Novel

Approach. International Journal of Advanced

Computer Science and Applications, 9(8).

Vankka, J., Myllykoski, H., Peltonen, T., & Riippa, K.

(2019). Sentiment Analysis of Finnish Customer

Reviews. 2019 Sixth International Conference on

Social Networks Analysis, Management and Security

(SNAMS), 344–350.

Vig, S., Dumičić, K., & Klopotan, I. (2017). The Impact of

Reputation on Corporate Financial Performance:

Median Regression Approach. Business Systems

Research Journal, 8(2), 40–58.

Walker, K. (2010). A Systematic Review of the Corporate

Reputation Literature: Definition, Measurement, and

Theory. Corporate Reputation Review, 12(4), 357–387.

Wepener, M., & Boshoff, C. (2015). An instrument to

measure the customer-based corporate reputation of

large service organizations. Journal of Services

Marketing, 29(3), 163–172.

DATA 2021 - 10th International Conference on Data Science, Technology and Applications

352