International Journal of Computer Applications (0975 8887)

Volume 149 No.10, September 2016

A Critical Review of Sentiment Analysis

Punjabi University Regional Centre for Information
Technology and Management,
Mohali, Punjab, India

Fatehjeet Kaur Chopra

Rekha Bhatia, PhD

Punjabi University Regional Centre for Information
Technology and Management,
Mohali, Punjab, India

ABSTRACT

1.1 Document-level

Sentiment analysis is the process which helps in recognizing

peoples perspective and emotional conditions. Sentiment
analysis appeared in the roots of the disciplines of
psychology, sociology and anthropology. Sentiment analysis
takes place from the hypothesis of emotive attitude and
assessment hypothesis. It emphasis on sensation in forming
perceptions. Feelings that are generated from both conscious
and unconscious processing are called emotions. The feelings
of the people can be expressed in positive or negative ways.
Mostly, parts of speech are used as feature to extract the
sentiment of the text. Sentiment analysis is an evolving field
with a variety of use applications. Further, the evaluation of
the accuracy of the existing systems, from which it is
analyzed that the result can be improved by calculating the
sentiments of word instead of calculating sentiment of
complete sentence or paragraph.

Instructing of an evaluation report as conveying a positive or

negative emotion or slant is expected by Sentiment Analysis.
A fundamental data unit is considered as an entire archive.

Keywords
Sentiment Analysis, Opinion mining, NLP, Linguistic
Resources, Web data Analysis, Information Retrieval

1. INTRODUCTION
Sentiment Analysis administers inspecting feelings,
sentiments, and the mentality of a speaker or an author from a
given portion of text. "Sentiment analysis (opinion mining)
refers the utilization of custom unrefined (natural) language
processing, computational semantics and satisfied inquiring to
differentiate and withdraw internal facts in reference matters".
Sentiment analysis (opinion mining) is a multidisciplinary and
multifaceted artificial intelligence issue. Its point is to
minimize the hole in the middle of human and PC. In this
way, it is gathering of human insight and electronic
knowledge for mining the content and arranging client
notions, likes, despises and wishes [1].
Sentiment analysis(opinion mining) includes categorizing
beliefs in words into classes same as "positive" or "negative"
frequently accompanied by an implied class of "neutral".
Opinion mining or power of speech of the customer is other
names of Sentiment Analysis. Viewpoint(s) unexpressed by a
text span is attempted to be identified by Sentiment analysis;
for example an application that classifies a movie review as
"thumbs up" or "thumbs down". A novel machine-learning
method is proposed that applies text-categorization methods
to just the subjective portions of the document for determining
the sentiment polarity(for classification).Techniques for
finding minimum cuts in graphs can be used to implement
extraction of these portions and by this incorporation of crosssentence contextual constraints is facilitated greatly.
Three main classification levels in Sentiment Analysis which
are document-level, sentence-level, and aspect-level Opinion
mining(Sentiment Analysis).

1.2 Sentence-level
Group conclusion to be communicated in every sentence is
intended by Sentiment Analysis. Distinguishing whether the
sentence is subjective or objective is the principle step. When
the sentence is subjective, if the sentence communicates
positive or negative assessments is figured out by Sentencelevel SA. Number of applications and upgrades on SA
calculations were proposed in the most recent couple of years.
A more critical look on these improvements is intended to be
given by this overview and outlined and arranged a few
articles displayed in this field which are indicated by the
different SA strategies. The writers have gathered fifty-four
articles which exhibited critical upgrades to the SA field
recently. A wide mixed bag of Sentiment Analysis fields is
covered by these articles.
Market research well knows Sentiment Analysis. Utilizing out
of date consumer remark cards, researches, evaluations and
center sections, views or opinions have been examined
through the market. For proceeding benefits of the internet
mutual habitat some tools can be adapted, but only researcher
presence and small sample sizes can use them. By systematic
collection and analyzation of online sentiments through a
large sample of customers in real time, these problems can be
addressed by Sentiment analysis. Online sentiments are
conceptualized as human convictions or emotions expressed
on the internet. With the most popular social network sites ,
attitude is usually expressed from various media alternatives
towards a situation, event or object.
Online sentiments examples are:
I actually love my new Smart phone
"This picture is the ultimate picture I have ever seen
Taste of bagels is worse in town
So systematic analysis of online expressions is known as
Sentiment analysis. Sentiment analysis (opinion mining)
emphasis on viewpoint assessment and views of beliefs on a
subject of regard utilizing mechanism studying approaches
particularly.

2. LITERATURE REVIEW
(Jeonghee Yi, 2003) [2] has stated that extraction of sentiment
(or opinion) about a subject from online text documents is
done by sentiment Analyzer (SA). All references to the given
subject is detected by SA, and sentiment in each references
using natural language processing (NLP) techniques is also
determined by SA, instead of classifying the sentiment of an
entire document about a subject. In this sentiment analysis has
1) A topic specific feature term extraction, 2) Sentiment
extraction, and 3) Subject, sentiment association by
37

International Journal of Computer Applications (0975 8887)

Volume 149 No.10, September 2016
relationship analysis. Sentiment Analysis for the analysis
utilizes two linguistic resources :the sentiment lexicon and the
sentiment pattern database. Online product review articles
(digital camera and music evaluation(reviews)), and
general documents (web pages and news article also included
) verifies the performance of the algorithms.

Mikalai, 2012)[9], (B L. , 2012)[10], (Pang B, Opinion mining

and sentiment analysis, 2008)[3] , (Cambria E, 2013)[11],
(Feldman, 2013)[12] and (Montoyo Andre s, 2012;)[13].
The issue of Sentiment Analysis (SA) from the execution
viewpoint not from the Sentiment Analysis(SA) methods
viewpoint have been reviewed by those surveys.

(Pang B, Opinion mining and sentiment analysis., 2008) [3]

have analyzed concept in (Liu, 2004); the academics have
undertaken a diverse range of related research, due to its
practicality in opinion monitoring and business competitive
intelligence. Sentiment analysis on online reviews has become
increasingly popular. A multidisciplinary research field in
nature, various areas like natural language processing (NLP),
computational language rules, information recovery, machine
intelligence and artificial intelligence etc are included in
sentiment analysis. No research on the commercial value of
online product reviews.

(Yu Liang-Chih, 2013) [5] Has shown their research that

extraction of peoples opinions on features of an entity is the
important task of opinion mining. There is a need to assemble
these words and phrases, which are domain synonyms, into the
same feature group to produce a useful synopsis, . Also, there
is a complication in the sentiment relation of the features and
opinions .To deal with the feature-level opinion mining
problems, a novel method is proposed.

(Wilson T, 2005) [4] have pointed out that nature of sentiment

expressions are not necessarily subjective. As sentences are
just short documents so no basic distinction in the middle of
document and sentence level classes is there (B L., 2012).
Document level or sentence level classes of wordings does not
supply or give the required piece of information needed; so
there occurs a necessity to go to the aspect level. Categorizing
the view or emotion with respect to the particular features of
bodies is the focus or point of the Aspect-level SA.
Recognizing the bodies and their features is the main first step.
Dissimilar beliefs for dissimilar features of the identical body
can be stated by belief possessors same as sentence The
battery life is extensive, but voice quality of this phone is not
good. The first two types of Sentiment Analysis can be
approached by this observation. A chief matter in this area is
the data sets that are used in SA. From product reviews the
main sources of data can be taken. Sentiment Analysis (SA) is
not only implemented on product assessment but can also be
implemented on stock markets (Yu Liang-Chih, 2013)[5] and
(Michael Hagenau, 2013)[6], news reports(articles), (Tao Xu,
2012) [7] or political discussions (Maks Isa, 2012)[8]. For
example, peoples beliefs on a definite election applicants or
political parties could be figured out in political debates.
Prediction of election results can also be done from political
posts. In the SA process, they are also used as data sources.

2) Vague opinion words and clear opinion words are the

categorized divisions of opinion words, which aim to discover
the implicit features and clutch the features. There are three
aspects on which clutching of features depend: The correlated
opinion words, the resemblance of the features and the
formation of the features.

As most famous SA techniques and applications are covered in

one research paper, so in this field for new comer researchers
this survey can be useful. A purified classification to the
different Sentiment Analysis (SA) methods which is not
originated in other observations is given uniquely by this
survey. New associated areas in Sentiment Analysis (SA)
which have allured the researchers recently and their correlated
papers(reports) are discussed. In these areas Emotion Detection
(ED), Building Resources (BR) and Transfer Learning (TL)
are incorporated .Extraction and analyzation of emotions is the
goal of emotion detection, while in the sentences the feelings
could be unreserved (direct) or indirect. Transfer studying (or
learning) or Cross-Domain categorization is concerned with
inspecting statistics from one area and then utilizing the
outcomes in a objective field. .Creating lexica, corpora are the
goal of building resources in which opinion expressions are
elucidated in proportion to their polarity, and sometimes
glossaries. Every year in the SA fields great number of articles
are introduced. Along years number of articles is growing. By
this the requirement to have observation documents have been
created that abstract the latest analysis styles and ways of
Sentiment Analysis(SA).Some worldly-wise and complete
surveys can be found by the reader including (Tsytsarau

1) The explicit features and the implicit features are

contemplated in the proposed method.

Also, to strengthen the clutching in the procedure the context

information is used , which is shown to be useful in clustering
or clutching. Feature-level opinion mining including three
steps: (1) Extract the features and the corresponding opinions
(2) Cluster the features (3) Orient the opinions of the feature.
Feature extraction for entities is an important task for opinion
mining. This paper proposed a new method to deal with this
problem. The new method uses the corresponding opinion
words extracting the features, and according to mutual support
and confidence to filter the noise. It also identifies the implicit
features and clusters the features based on the knowledge of
the background which strengthen cluster results. Empirical
evaluation show the proposed method outperforms. However,
this method has some shortcomings. Small scale corpus cannot
(Yao & Chen, 2013) [14] perform well. And the structure of
the vague opinions dictionary and part-of-speech dictionary
increases the cost of the method. Next, the establishment of
two dictionaries by automatically and improve the precision
and recall for the small scale corpus have been studied.
(Singh & Piryani, 2013) [15] has stated an fact-finding effort
work on a new type of field particular feature-based
investigatory for aspect-level sentiment analysis of picture
analysis(reviews). A feature directed plan that examines the
textual assessment of a picture have been conceived and
sentiment tag on each point is allocated to it. The scores on
each side or point from various analyses are then summated
and a final description of the picture is created on all
boundaries. A Senti Word Net established plan with two
dissimilar semantic attribute choices consisting of adjectives,
adverbs and verbs and n-gram attribute withdrawal is utilized.
Also a Senti Word Net plan is utilized to calculate the
document-level sentiment for each picture inspected and
contrasted the outcomes with outcomes acquired utilizing
Alchemy API. The sentiment description of a picture is also
contrasted with the document-level sentiment outcome. The
outcomes acquired show that the plan provides a more precise
and concentrated sentiment description than the uncomplicated
document-level sentiment analysis.
There are mainly three kinds of viewpoints for sentiment
categorization of words:

38

(a) Machine learning based text classifier like Nave Bayes,

SVM or KNN- with satisfactory attribute selection plan are
used;
(b) Unrestricted semantic alignment plan of extricating
applicable n-grams of the words (text) are used and then
tagging them either as positive or negative and considerably
the report or document; and
(c) Senti Word Net rooted publicly obtainable library that gives
positive, negative and neutral scores for words or texts is used.
Some of the applicable previous works on sentiment
categorization can be originated in (K. Dave, 2003)[16],
(Turney P. D., 2002)[17], (F.Sebastiani, 2005)[18].
(Varghese, 2004) [19] has used the concept where the word
level extent attribute extraction is done utilizing Naive
Bayesian Classifier. The semantic alignment of the separate
sentences is recovered from the contextual data or information.
This machine learning viewpoint on average normally affirms
an precision rate of 83%. Another remarkable work is the
execution of both Natural Language perception and creation in
Sentiment analysis (SA) (Pang B, Opinion mining and
sentiment analysis, 2008).

3. OBJECTIVES
Sentiment analysis is an emerging research field and this task
is very important because peoples spent their most of the time
on the web. In this an approach is proposed to determine the
sentiment orientation i.e. polarity of the Punjabi general
evaluation (reviews)s by scoring procedure. Sentiment analysis
(SA) is needed to be performed in Punjabi language because of
the growth in Punjabi facts or figures on the web. Separate
positive and negative summarized results are created which is
useful for the user in making the decision. The objective of this
is

on fields or areas or topics. In time to come, more work is

required on furthermore bettering the performance
evaluations. Sentiment analysis can be tried or auditioned for
new applications. Even though the algorithms and methods
used for Sentiment analysis (SA) are proceeding fast, but, a
lot of issues in this area of study prevail unsolved. The main
demanding details exist in hold of other languages, trading
with negation proclamations(expressions); construct or yield a
summary of beliefs build on product features(attributes),
difficulty
of
sentence(document),
managing
of
implied(indirect) product attributes etc. More research could
be committed to these problems (challenges) in time to come.

5. REFERENCES
[1] Kaur, A., & Gupta, V. (Nov. 2013). A survey on
sentiment analysis and opinion mining techniques.
Proceedings of Journal of Emerging Technologies in
Web Intelligence , Vol.5, No. 4. 62.
[2] Jeonghee Yi, T. N. (2003). Sentiment Analyzer:
Extracting Sentiments about a Given Topic using Natural
Language Processing Techniques. Proceedings of the
Third IEEE International Conference on Data Mining .
IEEE.
[3] Pang B, L. L. (2008). Opinion mining and sentiment
analysis. Foundations and trends in information retrieval
, 1-2 (2), 1-135.
[4] Wilson T, W. J. (2005). Recognizing contextual polarity.
In: Proceedings of HLT/EMNLP.
[5] Yu Liang-Chih, W. J.-L.-C. (2013). Using a contextual
entropy model to expand emotion words and their
intensity for the sentiment classification of stock. KnowlBased Syst , (pp. 41:8997).

To Study the algorithm of existing sentiment

analysis in various non-identical languages like
Hindi, Punjabi and English.

[6] Michael Hagenau, M. L. (2013). Automated news

reading: stock price prediction based on financial news
using context-capturing features. Decis Supp Syst.

To make an application that maintains the directory

of positive, negative and neutral words
corresponding to their values.

[7] Tao Xu, P. Q. (2012). Identifying the semantic

orientation of terms using S-HAL for sentiment analysis.
(pp. 35:27989). Knowl-Based Syst .

To implement an algorithm that analyzes the

sentiments of entered Punjabi paragraph and
highlights the words according to their sentiment.

[8] Maks Isa, V. P. (2012). A lexicon model for deep

sentiment analysis and opinion mining applications. (pp.
53:6808). Decis Support Syst.

Algorithm calculates the result according to value of

highlighted words.

[9] Tsytsarau Mikalai, P. T. (2012). Survey on mining

subjective data on the web. (pp. 24:478514). Data Min
Knowl Discov. 64.

To find the results of proposed algorithm with the

help of negative, positive and neutral words
database maintained in the application.

[10] B, L. (2012). Sentiment analysis and opinion mining.

Synth Lect.

4. CONCLUSION
Sentiment analysis appeal is tremendously developing. The
researchers can be faced with problems, oppositions and
challenges; though semantic analysis is emerging but this field
is still new to them .Problem related to the nature of
classification is a possible challenge. Classification techniques
that create two or three groups at most, there is a limit in the
extraction of groups and subgroups. Text based data that is
valid in specific places at specific times are usually context
specific and domain dependent. Sentiment observation has a
broad type of approaches in information systems, which
includes categorizing reviews, encapsulating analysis (review)
and different real time applications. There are probably to be
various distinct applications particularly not examined. It is
constituted that sentiment words (classifiers) are acutely based

[11] Cambria E, S. B. (2013). New avenues in opinion mining

and sentiment analysis. IEEE Intell Syst (pp. 28:1521).
IEEE.
[12] Feldman, R. (2013). Techniques and applications for
sentiment analysis. Commun ACM , (pp. 56:829).
[13] Montoyo Andre s, M.-B. P. (2012;). Subjectivity and
sentiment analysis: an overview of the current state of
the area and envisaged developments. (pp. 53:6759).
Decis Support Syst.
[14] Yao, R., & Chen, J. (2013). Predicting Movie Sales
Revenue using Online Reviews. International Conference
on Granular Computing (GrC). IEEE.

39

International Journal of Computer Applications (0975 8887)

Volume 149 No.10, September 2016
[15] Singh, V., & Piryani, R. (2013). Sentiment Analysis of
Movie Reviews A new feature based heuristic for
aspect-level sentiment classification. IEEE.
[16] K. Dave, S. L. (2003). Mining the Peanut GalleryOpinion Extraction and Semantic Classification of
Product Reviews. Proceedings of the 12th International
World Wide Web Conference, (pp. 519-528).
[17] Turney, P. D. (2002). Thumbs up or thumbs down?
Semantic
orientation applied
to
unsupervised
classification of reviews. Proc. 40th Annual Meeting on
Association for Computational Linguistics (ACL),
Philadelphia , US.

IJCATM : www.ijcaonline.org

[18] F.Sebastiani, A. E. (2005). Determining the Semantic

Orientation of terms through gloss analysis.
Proceedings of CIKM-05, 14th ACM International
Conference on Information and Knowledge Management
(pp. 617-624,). Bremen, DE, CIKM.
[19] Varghese, R. (2004). Opinion Mining Based on FeatureLevel Aspect Based Sentiment Analysis using Support
Vector Machine Classifier. International Conference on
Advances in Computing, Communications and
Informatics
(ICACCI),ICACCI.Heba
Khudhair
Abass(2013), A Study of Digital Image Fusion
Techniques Based on Contrast and Correlation
Measures.

40