See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/309180696

Review on Developing Corpora for Sentiment Analysis Using Plutchik’s Wheel of

Emotions with Fuzzy Logic

Article · January 2014

CITATIONS READS

28 16,014

2 authors, including:

Amit Pimpalkar
Sathyabama Institute of Science and Technology
49 PUBLICATIONS 244 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A Review Paper on Hindi Language Graphical User Interface to Relational Database using NLP View project

Political Comments Analysis Using Text Mining View project

All content following this page was uploaded by Amit Pimpalkar on 02 November 2018.

The user has requested enhancement of the downloaded file.

Sentiment Analysis on Product ABHIYANTRIKI Chafale & Pimpalkar
An International Journal of Engineering & Technology
(AIJET)
Vol. 1, No. 2 (December, 2014) http://www.aijet.in/ eISSN: 2394-627X

Sentiment Analysis on Product Reviews Using Plutchik’s

Wheel of Emotions with Fuzzy Logic

Dhanashri Chafale* Amit Pimpalkar

Department of Computer Science & Engineering Department of Computer Science & Engineering,
G.H.R.A.E.T, Nagpur University, India G.H.R.A.E.T, Nagpur University, India.

Abstract
Internet is the day by day increasing global system. In recent years number of
efforts were devoted for mining opinions and sentiments automatically from
natural language in social media messages, commercial product reviews, news
and movie reviews. This task includes understanding explicit and implicit
information conveyed by the language deeply. Most of these approaches refer to
annotated corpora. The opinion mining is used for identification and extraction
the information, which is in the subjective form and collected from the internet.
This can be done by making use of data required for processing. The methods
such as natural language processing, text analysis etc are used. Sentiments can
also be extracted from the feedbacks. Feedback is important for purchasing or
selling any product. While shopping whenever one wants to choose a product, the
opinion of others may be useful to him/her to choose the best product. But in case
of customers, usually it is a difficult task to read thousands of reviews at a time
and also it is likely to be a time consuming process. It also creates confusion.
Therefore as a solution for this, some data mining techniques must be applied.
One more advantage of sentiment analysis is it helps in identifying the attitude of
the person. In our work, we present a system which develops a corpus for
opinion and sentiment analysis. We will collect the product reviews from any of
the available product review website and classify them as positive, negative and
neutral sentiments. The system will further classify the positive and negative
sentiments into emotions for which it uses the concept of Plutchik’s wheel of
emotions and makes a dictionary. It uses fuzzy logic approach for prediction and
generates output.

Keywords: Opinion Mining, Sentiment Classification, Sentiments, Corpus for

Sentiment Analysis.

*Author for correspondence [email protected]

1. Introduction
The task of mining sentiments and opinions from natural language is difficult one. It involves an
intense understanding of most of the implicit and explicit information which is conveyed by
structure of language. The availability of a dynamic corpus contains the user generated data, such
as reviews for products or polling data. The large and growing amount of information which is

ABHIYANTRIKI: An International Journal of Engineering & Technology 1

Volume 1, Number 2, December, 2014 (1-8)
Sentiment Analysis on Product Chafale & Pimpalkar

available in the Social Web fosters the proliferation of business and research activities around the
relatively new fields of sentiment analysis and opinion mining. Big data is the large amount of
easily available data on web, Social media, remote sensing data, etc. in form of structured data,
semi-structured or unstructured data. We can use this large data for sentiment analysis.

Sentiment analysis is the opinion mining used on the web for identifying the text. It is nothing
but to get the real voice of people for specific product, services, movies, news, issues etc.
Sentiment analysis can be done at various levels as, sentence level, document level and entity or
attribute level. The attitude of a particular person may be his/her judgment for the particular
product. Opinion or Feedback is very important for consumer as well as producer because most
of the people sale or purchase the products online. Individual consumers may want the opinions
of already existing users for the product before purchasing it. There are many web sites giving
information about product reviews. But, for customers it is somewhat a difficult task to read
these huge numbers of comments at a time. This creates confusion in mind, due to which
comparatively more time is required to take decision. So, the only option is using Data Mining
which mines opinion and performs sentimental analysis on this large data.

The main task of this system is gathering the reviews in large number that are available on
different online websites. The websites are available for online products selling, like Amazon,
flipkart etc. Gathering overall detail for the particular product, the polarity of the given text is
checked at the document level sentences. The result gives confirmation, about the contents of the
documents whether it is positive, negative or neutral. Then it uses a technique called “Plutchik’s
Wheel of Emotions” to categories sentiments further into eight basic emotions: i.e. joy, disgust,
trust, surprise, anticipation, fear, anger and sadness. This wheel of emotions is first invented by
the scientist Robert Plutchik. The basic emotions can be divided into two polarities i.e. joy
opposites to sadness, anger opposites to fear, trust opposites to disgust, and surprise opposites to
anticipation. Then each emotion can be further divided into three degrees, for example, serenity
is a lesser degree of joy and ecstasy is a more intense degree of joy. The eight basic emotions can
combine to form a new emotion. For example, joy and trust can be combined to form love
emotion. However, joy, trust, and anger are combined to form a new emotion jealousy.

The aim of the system is analysis of the sentiments for the online available reviews given on
product from online shopping websites. The input data is collected as reviews from the online
shopping websites because the comments for products are posted there. It compares between
products and identifies the best product.

2. Related Work
Basically, Sentiment Analysis is used to express individual person’s sentiment. According to
current state of the art sentiment analysis is used to classify sentiments into two categories
positive and negative. Some works classified them into as positive, negative and also in one
more category as objective (or neutral).

Pimpalkar et al. [1] developed a system that shows the comments and feedbacks/reviews for
products. They determined the polarity of sentiments for the comments of the person. After this
the comparison between two different products was done using comments which were identified
from the online resources. This comparison leads to find the best product. They used

ABHIYANTRIKI: An International Journal of Engineering & Technology 2

Volume 1, Number 2, December, 2014 (1-8)
Sentiment Analysis on Product Chafale & Pimpalkar

Sentiwordnet and smiley’s dictionary for determining the scores of words present in the
comment. Classification of Sentiments of words was done in three categorize as, positive,
negative and objective. The rule based and fuzzy logic approach was used to give the output.
Lertsuksakda et al. [2] developed a model, Hourglass of emotions to tag Thai stories using
Plutchik’s wheel of emotions. They reviewed the adopted computational representation of
emotions the so-called Hourglass of Emotion. They also proposed a construction of Thai
sentiment resource based on such representation for Thai sentiment term tagging. The Hourglass
of emotions improved upon Plutchik’s wheel of emotions where Plutchik introduced eight basic
emotions: i.e. joy, disgust, surprise, anticipation, trust, anger, fear and sadness. There are also
three degrees of each emotion. Lizhen et al. [3] proposed a feature-based vector model and a
novel weighting algorithm for sentiment analysis of Chinese product reviews. The model
considered modifying the relationships between words and contained rich sentiment strength
descriptions, represented by both adverbs of degree and punctuations. Feature vectors were
calculated by using dependency parsing. A novel feature weighting algorithm was proposed for
supervised sentiment classification. The experimental results were used to demonstrate the
effectiveness of the proposed method compared with a state of the art method using term level
weighting algorithms. Bosco et al. [4] works on development of a corpus for opinion and
sentiment analysis and presented as a case study Senti-TUT, an ongoing project for Italian aimed
at investigating sentiment and irony about politics in social media. They developed the two
corpora for twitter, namely TWNEWS and TWSPINO using political tweets. TWNEWS corpus
had been extracted by applying filters based on time and metadata, aimed at selecting posts
where a variety of opinions about politics is represented. TWSPINO is composed of 1,159
messages from the Twitter section of Spinoza a very popular Italian blog of posts with sharp
satire on politics. They extracted posts published from July 2009 to February 2012 and removed
advertising (1.5%). Gupta et al. [5] developed a system in which is useful information is
collected from the twitter website and sentiment analysis is performed on the tweets regarding
the smart phone war. The developed system uses the efficient scoring system which is used for
predicting the user’s age. A well trained Naive Bayes Classifier is used to predict user gender.
Tweet were labeled with a sentiment using Sentiment Classifier Model which helped in
analyzing the data which is based on various consumer parameters such as gender, age group and
location. Hemalatha et al. [6] developed a system in which pre-processing and machine learning
techniques combined to collects tweets from social networking sites. The noise in the data was
removed using preprocessing techniques. Machine learning techniques were applied on those
tweets which improve business intelligence by providing some prediction for decision making.
Classification of results of specific issue analysis was done as Positive, Negative and Neutral.
They studied three machine learning algorithms and developed a machine learning tool for
sentiment analysis. They also compared the size of file before and after applying the tool. Modha
et al. [7] discussed about exiting approaches, methods etc. for performing sentimental analysis on
unstructured data available on web. Previously, Sentiment Analysis concentrated for subjective
statements or on subjectivity and it just overlooked objective statements which carry
sentiment(s). They proposed a new approach which classifies and handles not only subjective but
also objective statements for sentimental analysis. They used the four steps for classification in
which first documents are categories in opinionated and non-opinionated sentences and then
opinionated sentences were taken to further divide them as subjected and objective. After that
both subjective and objective were divided as positive, negative and neutral in separate steps.

ABHIYANTRIKI: An International Journal of Engineering & Technology 3

Volume 1, Number 2, December, 2014 (1-8)
Sentiment Analysis on Product Chafale & Pimpalkar

They have evaluated their experimental results by using information Retrieval matrices such as
precision, recall, f-measure and accuracy.

Mudinas et al. [8] developed the system in which lexicon and learning based approach combined
for concept-level sentiment analysis. Vinodhini et al. [9] presented a survey which covered the
methods and techniques in sentiment analysis including challenges appeared in the field. They
compared the various techniques for sentiment classification. The techniques compared were
machine learning technique using supervised and unsupervised with different algorithms and the
feature based sentiment classification etc. They displayed a graphical result showing comparison
of these techniques. Mukherjee et al. [10] presented a novel approach which identified feature
specific expressions of opinion in product reviews with different features and mixed emotions.
They developed a system that extracts potential features from a review and clusters opinion
expressions describing each of the features. It finally retrieved the opinion expression which
describes the user specified feature. Their developed system showed improved accuracy over the
naïve baseline. They also showed that using supervised classification, the system outperforms the
naïve baseline by a huge margin. Nitin et al. [11] studied the problem of identifying comparative
sentences in text documents. A supervised learning approach is used for identifying comparative
sentences from text documents.

3. Challenges in Sentiment Analysis

Sentiment Analysis or Opinion Mining is the recent subtask of NLP (Natural Language
processing). Sentiment analysis field are having few challenges. The first is that, a word
containing opinion which is considered positive in one situation is not necessarily positive
everywhere as it can be considered negative in some other situations. Ex.: take the word “long”,
suppose a customer says “The battery life of laptop is long”. This statement indicates a positive
opinion for that laptop. But, in some other situation, the customer said that, start-up time of
laptop was long, and then this is a negative opinion.

Second thing is that none can always express opinions in the similar way. Almost all of the
traditional text processing makes use of the approach that relies on the fact that minute
differences between pieces of text not necessarily change their meaning very much. Ex.: “The
product is good” is so much different from “The product is not good”.

Sometimes, people may use contradictory statements. Some reviews will have both positive as
well as negative comments. For example: “The movie bombed even though the lead actor rocked
it”. This is not so difficult for a human being to understand it. But for a computer/machine it is
not so easy to parse. A well known drawback of the sentiment analysis using combination of
lexicon based and learning based approaches at document level is, the reviews with a lot of noise
are often assigned a neutral score. The reason for this is that the method fails to detect any
sentiment.

4. Proposed System
The focus of the system is on providing analysis of sentiments for product reviews and also
identifying the sentiment of the product efficiently for getting information of best product among
many products. First thing is it collects all the customer reviews for different products that
contain the facts and opinions. The basic classification subjective sentences is done into three

ABHIYANTRIKI: An International Journal of Engineering & Technology 4

Volume 1, Number 2, December, 2014 (1-8)
Sentiment Analysis on Product Chafale & Pimpalkar

categories as positive, negative and neutral by making use of the Hierarchical clustering. Then,
the Plutchik’s wheel of emotion is used. It further classifies the positive and negative sentences
into different Plutchik’s emotions. For this work machine learning based Neural network
technique is used. Along with this the corpus showing feedbacks classification is developed in
hierarchical form. The corpus will contain detailed classification. Finally the fuzzy logic is used
for prediction purpose and gives the best product.

Objectives of the proposed system are,

a) Collection of product reviews from customers/users which contains opinions.
b) Applying the filtration process and remove unwanted data (noise) from sentences using
stemming, stop word removal etc.
c) Classifying sentiments as positive, negative and neutral.
d) Again classifying positive and negative sentiments into various emotions using Plutchik’s
wheel of emotions.
e) Developing corpus in hierarchical format.

A. Plutchik’s Wheel of Emotions

Robert Plutchik invented a wheel of emotions. He suggested eight primary emotions: joy
opposite to sadness. Similarly, anger opposite to fear; trust opposite to disgust and surprise
opposite to anticipation. These four opposite emotion pairs, show the 8 basic emotions.
Additionally, Plutchik’s model shows connections
between the ideas of circle of emotions using a color
wheel. Like the case of colors, primary emotions
can also be expressed at different degrees of their
intensities, for each emotion there are three degrees.
For example, serenity is a less intense degree of joy
and ecstasy is a more intense degree of joy.
Plutchik’s emotions can be mixed with one another
forming a new different emotion. For example,
combination of joy and trust resulted to form a new
emotion ‘love’. Likewise, joy, anger and trust are
combined and form jealousy.

Phases of the System

Fig. 1: Plutchik's Wheel of Emotions
1) Data collection: The process of collecting information from various resources is called as
data collection. Our proposed system is going to collect the data from the comments given
for products by the customers. We use online shopping websites like amazon, flipkart and
many more available there for collecting user’s comments. We can take different types of
product and even multiple companies for the same product and collect reviews for each of
them.

2) Pre-processing: The pre-processing means the process which removes noise that is an
unwanted data from the customer’s comments for product. The noise may be in the form of
articles, stop words, etc. Different procedures are used for stemming; by making use of them
the comments will be filtered. Short text classifier is used for classification and

ABHIYANTRIKI: An International Journal of Engineering & Technology 5

Volume 1, Number 2, December, 2014 (1-8)
Sentiment Analysis on Product Chafale & Pimpalkar

representation; because this technique particularly deals with the short text messages and the
feedbacks/comments are short text messages.

3) Clustering: To show the categorization of feedbacks of the users as positive, negative and
neutral, one of the clustering techniques will be used. Here for our project we are going to
use a clustering technique called hierarchical clustering. Hierarchical clustering is used to
build hierarchy of clusters and using this concept the corpus is developed according to
Plutchik’s wheel. Clustering starts with the positive tags, negative tags and neutral tags,
where tags refer to the particular group which contains sentiments of same type. For ex.
negative tag contains all -ve sentiments. For further processing we will consider only
negative and positive sentiments.

4) Sentiments classification: In this phase the positive and negative feedbacks will be classified
further into different emotions. We will make use of Plutchik’s wheel of emotions for this
classification. The corpus will be developed on the basis of this classification. For this
classification Neural network which is a machine learning based technique, will be used.
Here we have decided to use Artificial Neural Network which is a computational model
having capability of machine learning. We will have to process multiple inputs because there
are positive as well as negative tags and also we want to produce more than one output. For
this reason we are going to use neural network which accepts multiple inputs and produce
multiple outputs. It will classify positive and negative sentiments into emotions.

5) Prediction: Here in the last phase called prediction, the fuzzy logic will be used for the
prediction purpose that is to predict better product among various. Fuzzy logic basically, is a
many valued logic. Fuzzy logic deals with the approximate values instead of exact or fixed.
Therefore by making use of some of the fuzzy approximations we will be able to compare
products of different companies. From this process we will make the decision as which
product is better or best. It will give the output as if the given product is good or bad. And
also it will predict the best product which is our main aim.

The figure below shows the overall work flow of the proposed system. It gives the brief idea
about how the proposed system will work.

B. Implementation Details
The product reviews are collected from the online product review websites. For this we have
taken the dataset containing user’s feedback for various products. We requested for this data
from the Amazon, the online shopping website and stored in a database. We maintained a
sentiment words dictionary which contains positive words dictionary and negative words
dictionary. All the comments for the selected products are shown first and then we have to
choose a comment from the list to further process it. When any one comment is selected it starts
processing on that comment.

First of all the stemming and stopword techniques are applied on it. This filtering removes
almost all unwanted noise from comment. The filtered comment is then split to get the separate
words for comparing. Then each single word is compared with the sentiment words dictionary. If
the word is matched with the positive or negative dictionary then it is placed in the

ABHIYANTRIKI: An International Journal of Engineering & Technology 6

Volume 1, Number 2, December, 2014 (1-8)
Sentiment Analysis on Product Chafale & Pimpalkar

corresponding box, that is positive word in positive words text and in the same way negative
words are placed. The comparison is done between number of positive word and number of
negative words in a given comment. The condition is checked whether the positive words are
more or negative and accordingly the comment is decided to be positive or negative. If both the
positive and negative words are same or if there are no positive or negative, the comment is
treated as neutral comment.

The implemented work is

described above briefly. Feedbacks of each
The total project consists of product

five modules. We have

developed the two modules Create Sentimental Dictionary
and described them here Hierarchical
shortly. The further Clustering
development is in progress.
Positive Negative Neutral
Until now we have done the Tag Tag Tag
data collection and
preprocessing on the Text
comments. And we are Representation
Filtration of Filtration of
selecting only a single sentiments user
comment at a time and Neural
comments
showed whether it is Network
positive, negative or neutral.
Now we will work in the
Positive Product1
same way for all comments Tag Corpus developed
for a particular product. In Plutchik’s wheel showing
of emotions Product2 classification in
future we will classify the hierarchical form
comments of users Negative
Tag Product3
according to Plutchik’s
wheel of emotions. And
also show the prediction Fuzzy
Prediction of product Logic
process that is compare the
product and give the best
Fig. 2: Workflow of system
product as a result.

5. Conclusion
The wide varieties of sentiment analysis applications are there in various systems that include
classifying and summarizing reviews. We used here a hierarchical clustering and showed the
hierarchy of sentiments in tree form. The sentiment analysis for the products review will help the
customer to choose the best product. Also it will help the developer or company to remove the
disadvantages of their product or services and re-design them according to customer’s need. The
use of Plutchik’s wheel of emotions will provide the real emotional view of comments.

ABHIYANTRIKI: An International Journal of Engineering & Technology 7

Volume 1, Number 2, December, 2014 (1-8)
 Sentiment Analysis on Product Chafale & Pimpalkar

References
[1] Pimpalkar, A., Wandhe, T., Kene, M., & Rao, M. S. (2014). Review of online product using rule
based and fuzzy logic with Smiley’s. International Journal of Computing and Technology. 1(1),
39-44.

[2] Lertsuksakda, R., Pasupa, K., & Netisopakul, P. (2014). Thai sentiment terms construction using
the Hourglass of emotions. In: Proceedings of 6th International Conference on Knowledge and
Smart Technology (KST, pp. 46-50).

[3] Lizhen, L., Wei, S., Chuchu, L., Hanshi, W., & Jingli, L. (2014). A novel feature-based method
for sentiment analysis of Chinese product reviews. In: Proceedings of ICT Management, China
Communication (ICTM-2014, pp. 154-164).

[4] Bosco, C., Bolioli, A., & Patti, V. (2013). Developing corpora for sentiment analysis and opinion
mining: the case of irony and Senti-TUT. IEEE Intelligent Systems.

[5] Gupta, A., Sondhi, K., Kumar, R., & Shivhre, N. (2013). Sentiment analysis for social media.
International Journal of Advanced Research in Computer Science and Software Engineering.
3(7), 216-221.

[6] Hemalatha, G. A., & Saradhi Varma, G. P. (2013). Sentiment analysis tool using machine
learning algorithms. International Journal of Emerging Trends & Technology in Computer
Science (IJETTCS). 2(2), 105-109.

[7] Modha, J. S., Modha, S. J., & Pandi, G. S. (2013). Automatic sentiment analysis for unstructured
data. 3(12), 91-97.

[8] Mudinas, A., Levene, M., & Zhang, D. (2012). Combining lexicon and learning based approaches
for concept-level sentiment analysis. In: Proceedings of the First International Workshop on
Issues of Sentiment Discovery and Opinion Mining (pp. 1-8).

[9] Vinodhini, G., & Chandrasekaran, R. M. (2012). Sentiment analysis and opinion mining: a
survey. International Journal of Advanced Research in Computer Science and Software
Engineering. 2(6), 281-292.

[10] Mukherjee, S., & Bhattacharyya, P. (2011). Feature specific sentiment analysis for product
reviews. Dept. of Computer Science and Engineering, IIT Bombay.

ABHIYANTRIKI: An International Journal of Engineering & Technology 8

Volume 1, Number 2, December, 2014 (1-8)

View publication stats