Content uploaded by Srishti Vashishtha
Author content
All content in this area was uploaded by Srishti Vashishtha on Nov 22, 2020
Content may be subject to copyright.
Fuzzy Interpretation of Word Polarity Scores for
Unsupervised Sentiment Analysis
Srishti Vashishtha
Information Technology
Department
Delhi Technological University
Delhi, India
srishtidtu@gmail.com
Seba Susan
Information Technology
Department
Delhi Technological University
Delhi, India
seba_406@yahoo.in
Abstract— Sentiment or Opinion Mining aims to determine
the polarity of people’s opinions, feeling towards any product,
service, event or any individual. One of the most popular
technique applied in sentiment analysis of textual content is
natural language processing. Sentiment can be evaluated using
numerous methodologies like machine learning algorithms and
statistical tools but the use of the fuzzy concept is not common. In
this paper, we analyze the effect of fuzzification of word polarity
sentiment scores. These word scores are obtained by deploying
two lexicons: SentiWordNet and AFINN. Experiments are
conducted on three benchmark datasets: polarity movie dataset
by Pang-Lee, IMDB and hotel reviews dataset. The key
highlights are: i) proposed an unsupervised fuzzy logic-based
approach for sentiment analysis of textual reviews, ii) the
proposed model formulated fuzzy cardinality as the measure for
the evaluation of word polarity scores, iii) our model has two
versions based on the sentiment lexicon deployed in the model,
iv) comparison of our fuzzy cardinality approach with other non-
fuzzy state-of-the-art methods reveals the superiority of our
fuzzy approach.
Keywords—sentiment analysis, natural language processing,
fuzzy, sentiment score, SentiWordNet, AFINN.
I. INTRODUCTION
Sentiment Analysis (SA) is the interpretation and study of
people’s opinions, attitudes, and emotions toward an entity.
The entity can represent individuals, events, or topics. It is a
way to analyze written or spoken language to determine if the
expression is favorable, unfavorable, or neutral, and to what
extent [1]. Nowadays, SA is in demand in several areas of
application: marketing, e-commerce, movies [2], news [3],
politics [4], hotels [5-7], restaurants, social media platforms
[8], etc. In current days the internet has created a boom. The
digital universe is estimated to consist of 44 zettabytes of data
at the beginning of 2020. In 2019, Google processed 3.7
million queries, Facebook saw one million logins, and
YouTube recorded 4.5 million videos viewed every 60
seconds. The number of internet users has risen from 3.7
billion in 2018 to 4.5 billion in 2019. Data production is high;
it is being produced every minute by internet applications like
emails, google apps, WhatsApp, music apps, etc and social
websites like Facebook, Twitter, Instagram, etc. This vast
amount of data can be used as input to the SA process for
prediction, marketing, research purpose, data mining, and
many other purposes. In marketing, it is required to analyze the
mindset of customers; what the customers need from a product
or service can be analyzed by conducting a survey or by
collecting reviews about a product or service [9-11]. The
application areas of SA are shown in Fig.1. Online and social
media platforms are the reservoir of reviews. These reviews
can be fed as input to the SA process, the output is the
sentiment: positive or negative, for binary sentiment
classification. Public sentiments regarding any social issue can
be analyzed easily using SA. Basic steps of this process include
collection of data (input), text preparation (pre-processing),
feature extraction, feature selection, detection of sentiment,
classification of sentiment, and presentation of output.
Sentiment Analysis can be carried out using Natural
Language Processing (NLP). NLP is an area of research and
application that explores how, with the help of computers, we
can understand and manipulate natural language for further
analysis. The researchers use NLP to collect information and
knowledge on how humans understand and use language that
develop models, tools and techniques [12].
Most of the data in SA is related to real-world problems.
These problems are complex so they require smart and
intelligent systems to integrate knowledge, techniques and
methodologies from various sources. These systems are
supposed to have such power that they think like human, and
are expert within a specific domain; adapt themselves and learn
to do better things in changing environment [13]. To deal with
such real-world problems the concept of fuzziness was
evolved. A fuzzy set consists of such elements which don’t
have any crisp boundary [14].
In our work, fuzzy logic is applied to online reviews to
compute the fuzzy sentiment score. Sentiment lexicons-
SentiWordNet [15] and AFINN [16] are deployed to compute
the sentiment score of words. These are lexical resources that
contain a list of words and their polarity scores. The proposed
model uses fuzzy cardinality as the measure for the evaluation
of word polarity scores. Our model has two versions based on
the sentiment lexicon deployed in the model. The application
of fuzzy logic with NLP provides us results that match human
interpretation for sentiment analysis. The rest of the paper is
IEEE - 49239
11th ICCCNT 2020
July 1-3, 2020 - IIT - Kharagpur,
Kharagpur, India
Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 08:47:28 UTC from IEEE Xplore. Restrictions apply.
organized as follows. Section II discusses different papers on
SA and fuzzy applied on SA. The proposed fuzzy approach is
described in Section III. Section IV is about experimental setup
of our approach. The results are presented in Section V and the
conclusions are drawn in Section VI.
Fig.1. Application Areas of Sentiment Analysis
II. RELATED STUDIES
In this section we give an overview of past research works
on Sentiment Analysis and fuzzy logic-based SA approaches.
SA is one of the most recent and upcoming research areas.
The easiest way to assign sentiment is by classifying it as
either positive or negative. Several papers have discussed how
we can perform SA on different entities and classify them
according to our requirements. Users post their opinions about
different products on various online websites, these product
reviews are analyzed for SA [9-11]. In [9], the SA process has
been classified in five ways. First of all, the SA process can be
divided according to levels: Document Level, Sentence Level,
and Aspect Level. The second way is polarity-based sentiment
classification: Binary approach, Multi-level approach and
Contextual or Fuzzy approach. In [17], users of mobile
handsets give their thoughts about mobile handsets based on
mobile features. There are numerous features but only those
features that are crucial and essential for the performance of
mobile handset are selected for determining the sentiments.
Another paper discusses overall perspectives related to SA
based on text and emotions [18]. The number of hotel review
websites and blogs are increasing rapidly. These reviews help
people from these websites and blogs when they plan their
next vacation. A number of studies have proposed different
approaches to compute sentiment [5-7] and ratings [19] for
these hotel reviews. In a recent work, the reactions and
sentiments of people on a public platform during the natural
calamity like Kerala floods are analyzed using the Naïve
Bayes classifier [20]. In [21-22], the authors have explored the
effect of the demonetization on public and Indian financial
market using SA. The public opinions about demonetization
are collected from tweets across the whole country.
Many SA approaches use classifiers like Naïve Bayes
[2,19,23], Support Vector Machine (SVM) [3,23] and
Maximum Entropy [23]. But there has been little use of
classifiers based on fuzzy sets. The importance of fuzziness
comes into play while dealing with natural language due to the
presence of ambiguity in language. The concept of fuzzy sets
was formulated by Zadeh [14]. Fuzzy sets can be applied to
decide the degree of a positive or negative word, with the help
of fuzzy memberships, for evaluating sentiments [24]. In [25],
the proposed model for SA shows that not all positive or
negative words can be treated as equal; some words are more
positive or negative compared to other words. The concept of
fuzzy helps us to deal with real-world problems. SA is
performed for product reviews to classify them as positive,
negative or neutral with the help of a fuzzy model [10,11]. In
[10], intuitionistic fuzzy set theory is applied to convert the
sentiment orientations into fuzzy numbers. There are SA
approaches that incorporate the effect of different linguistic
hedges with fuzzy logic to compute the sentiment [8,10].
While the SentiWordNet lexicon is used in [8], Feature
Orientation dictionary is used in [10] to calculate the fuzzy
value of each word, and further, these values are used to
evaluate the results. Triangular fuzzy sets can be applied to
hotel reviews with the help of three quantifiers: “most”, “half
of” and “few”. Short sentiment summaries are created from
these fuzzy quantified sentences [5]. Fuzzy rule-based systems
for SA [26-29] have shown better results than commonly used
Naïve Bayes, Decision Trees and SVM. The aim is to provide
polarity classification degree values. There exist fuzzy logic-
based systems for sentiment classification of online customer
reviews [30,31] using fuzzy inference systems based on rules.
Machine learning methods [32-34] and Deep learning methods
[35,36] along with a fuzzy approach for SA on textual reviews
have proved to yield good results. In [36], fuzzy entropy
measure and k-means clustering are used to shortlist important
words from reviews that carry high sentiment quotient. The
fuzzy scores of these shortlisted words are computed using the
SentiWordNet lexicon [15]; further these scores are sent into
the LSTM neural network for sentiment classification. A
previous work by authors has used supervised machine
learning method- SVM for SA of video reviews posted on
social media [37]. This method uses a fusion of acoustic and
linguistic features for sentiment classification in multimodal
SA.
There are millions of online reviews on internet for various
topics, events, products or services. These reviews have been
analysed by numerous researchers by applying SA but they
have failed to unveil correct results for the uncertain or
ambiguous data present in the language. SA is the process of
determining and computing the opinions, attitudes, feelings
expressed using natural language by people. Words are the
fundamental building block of language. Every human
language spoken or written, is composed of words. Word-level
approaches to NLP are the first step towards understanding the
natural language. The computer needs to comprehend how
things work in real-world domain; this effort although very
progressive, has a limitation. There is an intelligence gap
between a human and a machine. Fuzzy logic can be used to
make the machine understand this intelligence gap in a better
way, because it deals with uncertainty, vagueness or
impreciseness factors that are present in human language.
IEEE - 49239
11th ICCCNT 2020
July 1-3, 2020 - IIT - Kharagpur,
Kharagpur, India
Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 08:47:28 UTC from IEEE Xplore. Restrictions apply.
Thus, this motivated us to incorporate fuzzy logic in SA to
improve classification. In this paper, we have applied a fuzzy
logic word-level approach for determining and classifying the
sentiment of online reviews. Most of the existing works in
field of SA employ supervised learning algorithms. We have
implemented an unsupervised approach because unlike
supervised learning, unsupervised learning processes do not
need any labeled data; hence training time and computational
complexity is saved. The next section describes the proposed
methodology.
III. PROPOSED APPROACH
Our proposed unsupervised approach for sentiment
analysis of textual reviews has four major steps. The steps
include tokenization, formulation of a bag of words model,
formulation of fuzzy sentiment score and assigning polarity.
We have two versions depending on the type of lexicon being
applied. The two lexicons are SentiWordNet [15] and AFINN
[16]. Following is the description of steps:
A. Tokenization & Lemmatization
Reviews are stored in a document. To work on documents,
we need to first break down the document into sentences.
Splitting up of paragraphs into sentences is termed as Sentence
Tokenization [38]. Tokenizing a sentence is a process of
splitting a sentence into a list of words. In other words, a
tokenizer parses a sentence into a list of tokens (words). The
output of the tokenization process will be stored in a dynamic
list. Each word of the sentence is lemmatized.
B. Bag of Words
Bag of words term in SA refers to those keywords which
are important for mining reviews, opinions, etc. Technically it
is a model that transforms documents into vector (numerical)
form, where each word in the document is assigned some
score in the range of 0 to 1. This can be interpreted as a fuzzy
membership pertaining to the fuzzy sets Pos and Neg. In our
proposed model we have used NLTK Part of Speech (POS)
Tagger [38] to extract words which are nouns, adjectives,
verbs or adverbs. A fuzzy set A can be represented as
{( , ( )},AA x x x U
=
where x is the element from the
universal set and
A
is the membership of element x. The
fuzzy sets Pos and Neg are represented as:
{( , ( )},Pos iPos a a a X
=
()
{( , ( )},Neg iNeg a a a X
=
()
where a is the word, Xi is ith set of Bag of words. If the total
number of reviews is n; then a bag of words is created for each
review. The membership functions
Pos
and
Neg
are
renamed as
swnPos
and
swnNeg
for SentiWordNet
lexicon[15] and;
afPos
and
afNeg
for AFINN lexicon[16].
[ . _ ()]
() ()
synsets
swnPos
syn pos score
alength synsets
=
()
[ . _ ()]
() ()
synsets
swnNeg
syn neg score
alength synsets
=
()
Eqs. (3) and (4) represent the fuzzy membership functions
of a word for SentiWordNet lexicon where syn.pos_score()
and syn.neg_score () are the scores obtained from
SentiWordNet; synsets is the set of synonyms of each word
present in SentiWordNet. The second version uses AFINN
lexicon. In eq. (5), the score, µaf, of each word is computed
using AFINN and divided by five because the score by
AFINN is between -5 and +5. Then the range of score is
checked, if it is greater than or equal to zero it is a positive
score, µafPos, otherwise it is a negative score, µafNeg. In eq. (6)
and eq. (7) the computation of fuzzy membership functions of
a word for AFINN lexicon are represented.
. ( )
() 5
af af score a
a
=
()
( ( ) 0) ( ( ) ( ))af afPos afif a then a a
= =
()
( ( ) 0) ( ( ) ( )af afNeg afif a then a a
= −
()
C. Formulation of proposed Fuzzy Cardinality measure
The fuzzy interpretation of word polarity scores for textual
reviews using both lexicons is computed in the last section. We
have proposed the Fuzzy Cardinality [13] measure for
evaluating the sentiment score of each review. The fuzzy
sets Pos and Neg contains the positive and negative score of
words in a review. The positive cardinality of the Pos set and
negative cardinality of Neg set is calculated by summing all the
elements in respective sets. This measure shows the strength of
fuzzy sets. Following are the definitions of positive and
negative Cardinality:
1
_ ( ),
l
Pos i
j
Pos cardinality a a X
=
=
()
1
_ ( ),
l
Neg i
j
Neg cardinality a a X
=
=
()
where l is the length of a review, a is the word, X is the set of
Bag of Words and it belongs to the ith Bag of words.
D. Assigning Polarity
The binary polarity classification of a review is either
positive or negative. Comparison of positive and negative
IEEE - 49239
11th ICCCNT 2020
July 1-3, 2020 - IIT - Kharagpur,
Kharagpur, India
Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 08:47:28 UTC from IEEE Xplore. Restrictions apply.
cardinality (eq. (8) and (9)) is performed to assign polarity to
each textual review. If positive cardinality of a review is
greater than or equal to negative cardinality, it implies it is a
positive review; otherwise it is a negative review.
, _ _
() ,
P Pos Cardinality Neg Cardinality
Polarity i N otherwise
=
()
In eq. (10), Polarity(i) is the polarity of ith review and the
labels for positive and negative polarity is P and N. Thereby
using above comparison all the reviews are divided into two
classes: Positive (P) and Negative (N). The process flow of
proposed fuzzy approach is depicted in Fig.2.
Fig.2. Process Flow of Proposed Fuzzy Approach
IV. EXPERIMENTAL SETUP AND IMPLEMENTATION
The proposed unsupervised methodology for SA has been
implemented in Python. The experiments are conducted on
three datasets that contain online reviews by users. There are
two movie datasets: polarity dataset v2.0 by Pang and Lee1
[39] and IMDB2. The third dataset provides reviews of a single
hotel3. These reviews were provided by the travellers who
stayed in this hotel. All the datasets are freely accessible via the
internet. The first dataset contains 2000 reviews, IMDB dataset
contains 50,000 reviews and the hotel review dataset has 38932
reviews. The dataset distribution of all the datasets according to
each sentiment class is represented in Table I.
In this paper, two sentiment lexicons: SentiWordNet [15]
and AFINN [16] are applied to compute the word polarity
scores. Table II shows the positive and negative score of few
words using the SentiWordNet lexicon. The positive words
have a positive score higher than the negative score and it is
vice-versa for negative words. There are some words that have
zero value for both positive and negative scores, these are
neutral words containing no sentiment, for example: “Hotel”
and “Staff”. “Nice”, “Helpful”, “Clean” and “Beautiful” are
1 http://www.cs.cornell.edu/people/pabo/movie-review-data/
2 http://www.imdb.com/
3 http://www.kaggle.com/harmanpreet93/hotelreviews
positive words while “Filthy”, “Difficult”, “Disgusting” and
“Accident” are negative words.
TABLE I. DATASET DISTRIBUTION FOR EACH CLASS.
Dataset
Positive
Negative
Total
Pang-Lee Movie
1000
1000
2000
IMDB Movie
25000
25000
50000
Hotel Reviews
26521
12411
38932
TABLE II. SENTIWORDNET SCORES
TABLE III. AFINN SCORES
Table III depicts the scores of these same words computed
using the AFINN lexicon. The words “Hotel” and “Staff” are
neutral words in both the lexicons. Since the scores of “Nice”,
“Helpful”, “Clean” and “Beautiful” are greater than zero their
polarity is positive. Whereas the words “Filthy”, “Difficult”,
“Disgusting” and “Accident” are negative words because their
scores are less than zero. The positive and negative words have
different scores in both lexicons, for example, the word “Nice”
has a 0.15 positive score and 0 negative scores in the
SentiWordNet lexicon while in the AFINN lexicon it has 3.0
score. Hence, both lexicons can be interpreted in different ways
to compute the polarity of a word and the score for the same
word is different.
Word
Pos Score
Neg Score
Hotel
0.0
0.0
Staff
0.0
0.0
Nice
0.15
0.0
Helpful
0.25
0.0
Clean
0.0278
0.0
Beautiful
0.3125
0.0
Filthy
0.0417
0.25
Difficult
0.0
0.3125
Disgusting
0.0625
0.3125
Accident
0.0
0.125
Word
Score
Score/5
Pos
Score
Neg
Score
Hotel
0.0
0.0
0.0
-
Staff
0.0
0.0
0.0
-
Nice
3.0
0.6
0.6
-
Helpful
2.0
0.4
0.4
-
Clean
2.0
0.4
0.4
-
Beautiful
3.0
0.6
0.6
-
Filthy
-2.0
-0.4
-
0.4
Difficult
-1.0
-0.2
-
0.2
Disgusting
-3.0
-0.6
-
0.6
Accident
-2.0
-0.4
-
0.4
IEEE - 49239
11th ICCCNT 2020
July 1-3, 2020 - IIT - Kharagpur,
Kharagpur, India
Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 08:47:28 UTC from IEEE Xplore. Restrictions apply.
We have compared our two versions of fuzzy approach
with two non-fuzzy approaches for SA. The first comparison
method is the Cavalcanti et al. approach [40], where the
sentiment score of each review is calculated by summing up
the sentiment score of each word, a, and dividing it by length
of each review, l, as shown in eq. (11). Here the sentiment
score of each word is the difference of positive and negative
score of each word computed using the SentiWordNet lexicon,
as given in eq. (3) and (4).
[ ( ) ( )]
( ) ,
Pos Neg
i
ii
aX
aa
score X a X
l
−
=
()
In Cavalcanti et al. approach, the scores greater than or equal
to zero denote the positive sentiment (P) and negative scores
denote the negative sentiment (N). The second comparison
method is Gilbert et al. approach [41], this approach had
created a VADER lexicon and its tool. This method computes
various scores for the given input textual sentence. We apply
the VADER approach to a review; the compound score is
generated. The range of compound score is checked for
different polarity classes: positive and negative. The
implementation of our proposed approach is available here4.
V. RESULTS AND DISCUSSION
An unsupervised fuzzy approach has been presented to
evaluate the sentiment of online textual reviews. There are two
versions of our fuzzy cardinality approach: SentiWordNet and
AFINN. These versions are compared with two unsupervised
non-fuzzy approaches: Cavalcanti et al. approach [40] and
Gilbert et al. approach [41]. All the experiments are conducted
on the three datasets of online reviews. The comparison of our
versions of fuzzy cardinality versions with other methods, in
terms of accuracy, are presented in Table IV.
TABLE IV. COMPARISION OF OUR VERSIONS OF FUZZY CARDINALITY
APPROACH WITH OTHER METHODS.
Dataset
Accuracy
Cavalcanti
et al.
Approach
[40]
Gilbert et
al.
Approach
VADER
[41]
Our Fuzzy
Cardinality
SentiWordNet
Approach
Our Fuzzy
Cardinality
AFINN
Approach
Pang-Lee
Movie
54.8%
63%,
63.5%
65.45%
IMDB
Movie
52.87%
69.43%
64.13%
70.06%
Hotel
Reviews
64.54%
76.1%
72.74%
76.2%
Experiments on Pang-Lee movie datasets reveal that our
fuzzy cardinality AFINN version has achieved the highest
accuracy of 65.45%, followed by SentiWordNet version
4https://www.github.com/SrishtiVashishtha/Fuzzy-Interpretation-of-Word-
Polarity-Scores-for-Unsupervised-Sentiment-Analysis
63.5%; comparable accuracy of 63% is gained by Gilbert et
al.’ s approach [41] and Cavalcanti et al.’ s approach [40]
has acquired the least accuracy of 54.8%. For the IMDB
movie dataset our fuzzy cardinality AFINN version has
achieved the highest accuracy of 70.06% and Gilbert et
al.’s approach achieved the second-highest accuracy of
69.43%. Our fuzzy cardinality AFINN version has gained
the highest accuracy of 76.2% in the Hotel reviews dataset,
which is comparable to that of Gilbert et al.’s approach.
Our fuzzy cardinality SentiWordNet version has scored
higher accuracy compared to Cavalcanti et al.’ s approach
in all datasets.
From the results in Table IV, we can conclude that our
fuzzy approach based on AFINN lexicon has scored the
highest accuracy in all the datasets, Cavalcanti et al.’ s
approach has gained the lowest accuracy in all the datasets.
Whereas the results of Gilbert et al.’s approach are
comparable to our fuzzy cardinality versions.
VI. CONCLUSION
In this research work, a fuzzy logic-based technique
is applied to online reviews to compute the fuzzy sentiment
score. Two sentiment lexicons- SentiWordNet and AFINN
are used to compute the sentiment score of words. The key
highlights are: i) proposed an unsupervised approach based
on fuzzy logic for sentiment analysis of textual reviews, ii)
the proposed model uses fuzzy cardinality as the measure
for the evaluation of word polarity scores, iii) our model
has two versions based on the sentiment lexicon deployed
in the model, iv) our fuzzy cardinality approach is
compared to non-fuzzy state-of-the-art methods.
Our proposed fuzzy methodology is better than non-
fuzzy methods. This is because of fuzzy deals with
ambiguity in real-world problems. Our approach calculates
the strength of average positive and negative scores of each
word in each review and these scores are fuzzy. Thus, the
strength of fuzzy sets gives better results than simple
average scores. The application of fuzzy logic with NLP
provides us results that match human interpretation for
sentiment analysis. Our approach can be applied to any
textual dataset that is based on online or social media
content, like twitter datasets, product reviews dataset, any
other customer review datasets, etc. The limitation of our
work is that the scores of words are dependent on lexicons;
some words which do not exist in lexicons cannot be
processed further. In this work, we have applied word-level
fuzzy logic approach; we can extend our work by working
on phrases in future.
REFERENCES
[1] B. Pang and Lillian Lee. "Opinion mining and sentiment
analysis." Foundations and trends in information retrieval 2, no. 1-2, 1-
135, 2008.
[2] P. Bhoir and S. Kolte. "Sentiment analysis of movie reviews using
lexicon approach." In 2015 IEEE International Conference on
Computational Intelligence and Computing Research (ICCIC), pp. 1-6.
IEEE, 2015.
IEEE - 49239
11th ICCCNT 2020
July 1-3, 2020 - IIT - Kharagpur,
Kharagpur, India
Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 08:47:28 UTC from IEEE Xplore. Restrictions apply.
[3] X. Li, H. Xie, L. Chen, J. Wang, and X. Deng. "News impact on stock
price return via sentiment analysis." Knowledge-Based Systems 69, 14-
23, 2014.
[4] R. Bose, R.K. Dey, S. Roy, and D. Sarddar. "Analyzing Political
Sentiment Using Twitter Data." In Information and Communication
Technology for Intelligent Systems, pp. 427-436. Springer, Singapore,
2019.
[5] B. Dundar, S. Ozdemir and D. Akay. "Opinion mining and fuzzy
quantification in hotel reviews." In 2016 International Symposium on
Networks, Computers and Communications (ISNCC), pp. 1-4. IEEE,
2016.
[6] E. Park, J. Kang, D. Choi, and J. Han. "Understanding customers' hotel
revisiting behaviour: a sentiment analysis of online feedback
reviews." Current Issues in Tourism 23, no. 5, 605-611, 2020.
[7] L. Mostafa. "Machine Learning-Based Sentiment Analysis for
Analyzing the Travelers Reviews on Egyptian Hotels." In Joint
European-US Workshop on Applications of Invariance in Computer
Vision, pp. 405-413. Springer, Cham, 2020.
[8] S. Vashishtha and S. Susan. "Fuzzy logic based dynamic plotting of
mood swings from tweets." In International Conference on Innovations
in Bio-Inspired Computing and Applications, pp. 129-139. Springer,
Cham, 2018.
[9] T.K. Shivaprasad and J. Shetty. "Sentiment analysis of product reviews:
a review." In 2017 International Conference on Inventive
Communication and Computational Technologies (ICICCT), pp. 298-
301. IEEE, 2017.
[10] Y. Liu, J.W. Bi, and Z.P. Fan. "Ranking products through online
reviews: A method based on sentiment analysis technique and
intuitionistic fuzzy set theory." Information Fusion 36, 149-161, 2017.
[11] K. Indhuja and RPC Reghu. "Fuzzy logic based sentiment analysis of
product review documents." In 2014 First International Conference on
Computational Systems and Communications (ICCSC), pp. 18-22. IEEE,
2014.
[12] G.G Chowdhury "Natural language processing." Annual review of
information science and technology 37, no. 1, 51-89, 2003.
[13] Jang, J. Sun. "R., Sun, CT., and Mizutani, E.(1997): Neuro-Fuzzy and
Soft Computing: A Computational Approach to Learning and Machine
Intelligence." Prentice Hall, Inc., Simon & Schuster/A Viacom
Company, Upper Saddle River, NJ 7458 (1997): 23.
[14] L.A. Zadeh. “Calculus of Fuzzy Restrictions.” , Fuzzy Sets and their
Applications to cognitive and decision processes, pages 1-40,1975.
[15] S. Baccianella, A. Esuli, and F. Sebastiani. "Sentiwordnet 3.0: an
enhanced lexical resource for sentiment analysis and opinion mining."
In Lrec, vol. 10, pp. 2200-2204, 2010.
[16] F.A. Nielsen. "Afinn." Richard Petersens Plads, Building 321, 2011.
[17] W. Singh. "Sentiment analysis of online mobile reviews." In 2017
International Conference on Inventive Communication and
Computational Technologies (ICICCT), pp. 20-25. IEEE, 2017.
[18] P. Yadav and D. Pandya. "SentiReview: Sentiment analysis based on
text and emoticons." In 2017 International Conference on Innovative
Mechanisms for Industry Applications (ICIMIA), pp. 467-472. IEEE,
2017.
[19] W. Songpan. "The analysis and prediction of customer review rating
using opinion mining." In 2017 IEEE 15th International Conference on
Software Engineering Research, Management and Applications (SERA),
pp. 71-77. IEEE, 2017.
[20] A. Dudani, V. Srividya, B. Sneha, and B. K. Tripathy. "Sentiment
Analysis on Kerala Floods." In International Conference on Innovative
Computing and Communications, pp. 107-124. Springer, Singapore,
2020.
[21] P. Singh, R.S. Sawhney and K.S. Kahlon. "Sentiment analysis of
demonetization of 500 & 1000 rupee banknotes by Indian
government." ICT Express 4, no. 3, 124-129, 2018.
[22] A. Kumar and J.P. Singh. "Demonetization in India: Good or Bad in
Context of Social Media." Available at SSRN 3349021, 2019.
[23] A. Tripathy, A. Agrawal, and S.K. Rath. "Classification of sentiment
reviews using n-gram machine learning approach." Expert Systems with
Applications 57, 117-126, 2016.
[24] C. Zhao, S. Wang and D. Li. "Fuzzy sentiment membership determining
for sentiment classification." In 2014 IEEE International Conference on
Data Mining Workshop, pp. 1191-1198. IEEE, 2014.
[25] S. Jusoh, and H. M. Alfawareh. "Applying fuzzy sets for opinion
mining." In 2013 International Conference on Computer Applications
Technology (ICCAT), pp. 1-5. IEEE, 2013.
[26] C. Jefferson, H. Liu, and M. Cocea. "Fuzzy approach for sentiment
analysis." In 2017 IEEE international conference on fuzzy systems
(FUZZ-IEEE), pp. 1-6. IEEE, 2017.
[27] H. Liu and M. Cocea. "Fuzzy rule based systems for interpretable
sentiment analysis." In 2017 Ninth International Conference on
Advanced Computational Intelligence (ICACI), pp. 129-136. IEEE,
2017.
[28] S. Vashishtha and S. Susan. "Fuzzy rule based unsupervised sentiment
analysis from social media posts." Expert Systems with
Applications 138, 112834, 2019.
[29] J. R. Alharbi and W. S. Alhalabi. "Hybrid Approach for Sentiment
Analysis of Twitter Posts Using a Dictionary-based Approach and Fuzzy
Logic Methods: Study Case on Cloud Service Providers." International
Journal on Semantic Web and Information Systems (IJSWIS) 16, no. 1,
116-145, 2020.
[30] R.F. de Sousa, R.A.L. Rabêlo and R. S. Moura. "A fuzzy system-based
approach to estimate the importance of online customer reviews."
In 2015 IEEE International Conference on Fuzzy Systems (FUZZ-
IEEE), pp. 1-8. IEEE, 2015.
[31] S. Nadali, M.A.A. Murad and R.A. Kadir. "Sentiment classification of
customer reviews based on fuzzy logic." In 2010 International
Symposium on Information Technology, vol. 2, pp. 1037-1044. IEEE,
2010.
[32] B.V. Krishna, A.K. Pandey, and A.P.S. Kumar. "Feature based opinion
mining and sentiment analysis using fuzzy logic." In Cognitive science
and artificial intelligence, pp. 79-89. Springer, Singapore, 2018.
[33] H.T. Phan, N.T. Nguyen, T.V. Cuong, and D. Hwang. "A Method for
Detecting and Analyzing the Sentiment of Tweets Containing Fuzzy
Sentiment Phrases." In 2019 IEEE International Symposium on
INnovations in Intelligent SysTems and Applications (INISTA), pp. 1-6.
IEEE, 2019.
[34] H.T. Phan, V.C. Tran, N.T. Nguyen and D. Hwang. "Improving the
performance of sentiment analysis of tweets containing fuzzy sentiment
using the feature ensemble model." IEEE Access 8, 14630-14641, 2020.
[35] P. Bedi and P. Khurana. "Sentiment Analysis Using Fuzzy-Deep
Learning." In Proceedings of ICETIT 2019, pp. 246-257. Springer,
Cham, 2020.
[36] S. Vashishtha and S. Susan. "Sentiment Cognition from Words
Shortlisted by Fuzzy Entropy." IEEE Transactions on Cognitive and
Developmental Systems, 2019.
[37] S. Vashishtha and S. Susan. "Inferring Sentiments from Supervised
Classification of Text and Speech cues using Fuzzy Rules." Procedia
Computer Science 167, 1370-1379, 2020.
[38] E. Loper and S. Bird. "NLTK: the natural language toolkit." arXiv
preprint cs/0205028, 2002.
[39] Pang, Bo, and Lillian Lee. "A sentimental education: Sentiment analysis
using subjectivity summarization based on minimum cuts."
In Proceedings of the 42nd annual meeting on Association for
Computational Linguistics, p. 271. Association for Computational
Linguistics, 2004.
[40] Cavalcanti, Diana C., Ricardo BC Prudêncio, Shreyasee S. Pradhan,
Jatin Y. Shah, and Ricardo S. Pietrobon. "Good to be bad?
Distinguishing between positive and negative citations in scientific
impact." In 2011 IEEE 23rd International Conference on Tools with
Artificial Intelligence, pp. 156-162. IEEE, 2011.
[41] C. J. Hutto and E. Gilbert. "Vader: A parsimonious rule-based model for
sentiment analysis of social media text." In Eighth international AAAI
conference on weblogs and social media. 2014.
IEEE - 49239
11th ICCCNT 2020
July 1-3, 2020 - IIT - Kharagpur,
Kharagpur, India
Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 08:47:28 UTC from IEEE Xplore. Restrictions apply.