Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Sentiment Analysis Techniques: A Review

Akansha Srivastava Ravindra Gupta
Research Scholar Assistant professor
Rkdf Institute of Science and Technology Rkdf Institute of Science and Technology
Bhopal (M.P.) Bhopal (M.P.)

Abstract:- Sentiments are the attitude, opinions, opinions. A subjective sentence may include numerous
thoughts, beliefs or feelings of the writer towards opinions and subjective and truthful parts. There are several
something, such as people, artifacts, company or prime data mining strategies applied to extract facts and
location. Sentiment analysis intends to conclude the information. Figure 1 shows the techniques of Opinion
judgment of a presenter or an author apropos to some Mining. A lot of steps are included in the whole process.
subject matter or on the whole relative polarity of the These steps include online cleaning of text, removing white
manuscript. The outlook could be the perception or spaces, amplifying acronym, stemming, stop word
assessment, emotional condition, or the projected elimination, refusal managing and lastly feature selection.
poignant message of the person behind.
Later, opinions are classified as positive, negative and
Keywords:- Sentiment Analysis, Machine learning, neutral using classification approaches.
Classification.

I. INTRODUCTION

Sentiments are the attitude, opinions, thoughts, beliefs

or feelings of the writer towards something, such as people,
artifacts, company or location. Sentiment analysis intends
to conclude the judgment of a presenter or an author
apropos to some subject matter or on the whole relative
polarity of the manuscript. The outlook could be the
perception or assessment, emotional condition, or the
projected poignant message of the person behind. Opinions
are decisive influencer of our behavior. Our views and Fig 1:- Techniques of Sentiment Analysis
insights of veracity are conditioned on how others perceive
the world. The rudimentary job in opinion mining deals Machine learning is completely based on machine
with deducing the inclusive polarity of the document on learning approaches. These approaches provide solution of
some specific subject matter. Sentiment analysis is a sentence level classification issue. Also, these approaches
‘suitcase’ field of research that contains numerous diverse make the decree of syntactic features. Machine learning
disciplines, not just associated to computer science but also approaches are of two types namely supervised learning
to social disciplines, such as psychology, philosophy, and and unsupervised learning. Machine learning is expected to
ethics [6]. Sentence level classification involves two tasks. allow machines to adjust their interior configuration in such
The purpose of primary task is to verify the nature of a way that they can predict the upcoming performance
statement i.e. subjective or objective. Subjective means boost.
individual’s own interpretation and objective opinion
means that you are looking as an outsider or another  Supervised Learning: Supervised learning considers
person. The main aim of second task is to verify if the classification issues. The general purpose is to obtain
subjective sentence is positive, negative, or neutral. There the workstation to discover a classification scheme that
are mainly two steps included in this process: we have formed. Digit recognition, once again, is a well
 Subjective classification of a sentence into one of two known example of classification learning. More widely,
categories i.e. objective and subjective classification learning is appropriate for any problem
 Sentiment classification of subjective sentences into two where classification learning is valuable and
categories i.e. positive and negative classification detection is easy. In some cases, it might
not be compulsory to give programmed classifications
Generally, truthful information is presented by an to every occurrence of a problem if the method can
objective sentence whereas a subjective sentence articulates itself perform classification.
individual feelings, views, sentiments, or values. There are  Unsupervised Learning: Without referring any labeled
several techniques using which subjective sentence can be results, the patterns of any dataset are assumed through
identified e.g. Naïve Bayesian classifier. Nevertheless, it is these types of algorithms. In contrast to supervised
merely not sufficient to know whether the sentence contain machine learning, it is not possible to apply
a positive or negative opinion. This is an intermediary step unsupervised machine learning techniques to a
that provides support in filtering out sentences having no regression or a classification problem. This makes the
training of algorithm complicated in normal way. In its

IJISRT20SEP653 www.ijisrt.com 913

Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
place, unsupervised learning can be utilized for that they computed the page rank separately for positive
discovering the underlying data structure. Unsupervised synset and negative synset (starting the entire graph from
machine learning is used to reveal earlier unknown data scratch for each case). Also, it is noticed that effectiveness
patterns [21]. is much better with positive terms. This means that
classifying negative terms is a harder task. As a conclusion,
II. LITERATURE REVIEW they see that this type of model can be applied to other
cases related with semantic properties of words.
Hu and Liu (2004) [7] performed opinion mining of
online product reviews in 3 steps: (1) features of goods K. Cai et al. (2008) [19] explained sentiment analysis
which have been remarked on by users are taken out first; which included a classification method along with an
(2) opinion sentences are discovered in each review and opinion based approach. The opinion classification element
then decision is taken whether each opinion is positive or differentiated the comparative sentiment expressed by the
negative (3) demonstrating the result. They collected terms in all fragments and then partitioned the fragments
reviews of numbers of products sold online like MP3 into positive, negative, and neutral groups. The sentiment
Players, DVD’s, digital camera and mobile phones from subject recognition module identifies the important areas
Amazon.com and CNN.net. Opinion word extraction and implied beyond every sentiment group by word support
aggregation is the main technique used by them and metrics.
features are preferred on the basis of opinion words itself.
Their contribution resulted in efficient performance as M. Eirinakiet al. (2012) [33] proposed an opinion
compared to opinion sentences extraction for DVD-73%, search engine scheme. The proposed approach integrated
and MP3-93%. The overall accuracy of five products is the pair of opinion mining algorithms. The outlooks are
achieved from 64% to 84%. based on features and the position of these outlooks is also
substantially built on the features as a substitute of an
Godbole et al. (2007) [13] proposed classification in object as a whole. Inhabitants appear to dislike a precise
a lexicon obtained from Word Net. They designed different object as of several features allied with the result. Their
lexicons for each topic. So, lexicon for politics is totally primary experimental assessment on numerous patron
different from that for health. From an initial lexicon, they review data sets has exposed that their findings achieved
designed a graph model to expand polarities to other words. extremely high level of accuracy.
For instance, if the word “good” is marked as positive, all
synonyms of “good” are marked as positive and all Karamibekr and Ghorbani (2012) [32] firstly
antonyms of “good” are marked as negative. Then, a new carried out an arithmetical exploration on the divergence
iteration is performed for next level (with the synonyms of among sentiment analysis of products and social issue.
the synonyms and the antonyms of the antonyms) and so Then, on the basis of some conclusions, they proposed a
on. Depending on the distance, the polarity score is scheme to consider the part of verb as the most imperative
different. Applying the formula 1/cd where c > 1 and d is expression in conveying opinions concerning the societal
the number of nodes away. With this kind of formulation, matters. Statistical and experimental fallouts confirm that
the system ends up with polarities defined for all the words. making an allowance for verbs not merely is essential and
After getting score for all words, we can calculate polarity definite, other than that they also augment the concert of
scores of each text by dividing the sum of all polarity sentiments analysis. They collected their data from
scores in a text between numbers of total words. The score Procon.org, yahoo and CNN answers. Features are picked
was tested against names of celebrities i.e. Maria on the hinge of opinion directories and opinion structure.
Sharapova got the best score. Formed on verb-oriented method result are calculated as
65% for social issues and 62.5% for car models.
Esuli and Sebastiani (2007) [14] presented an
extremely interesting scheme that applied page rank K. Ghag and K. Shah (2013) [35] surveyed that
algorithm to determine term polarities. For this purpose, Sentiment Analyzers are based on language. Various
they used extended WordNet to build a graph where each practices used a dictionary to collect opinion. Few
synset has certain polarity depending on the polarity of its techniques used training set while others used both training
members. The main hypothesis is that there won’t be huge set and dictionary. No existing method is widespread
variations and each synset will have a similar degree of sufficiently to be language independent. This clearly stated
negativity. This will produce a graph of relation between the necessity of hard work to demonstrate Sentiment
different synsets that will transfer its polarity properties to Analyzer without utilizing training dataset.
its neighbors. One interesting point of this experiment is

IJISRT20SEP653 www.ijisrt.com 914

Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Author Year Description Outcome
Hu and Liu 2004 They collected reviews of numbers of Their contribution resulted in efficient
products sold online like MP3 Players, performance as compared to opinion
DVD’s, digital camera and mobile phones sentences extraction for DVD-73%, and
from Amazon.com and CNN.net. Opinion MP3-93%. The overall accuracy of five
word extraction and aggregation is the main products is achieved from 64% to 84%.
technique used by them and features are
preferred on the basis of opinion words itself.

Godbole et 2007 They designed different lexicons for each After getting score for all words, we can
al topic. So, lexicon for politics is totally calculate polarity scores of each text by
different from that for health. From an initial dividing the sum of all polarity scores in a
lexicon, they designed a graph model to text between numbers of total words. The
expand polarities to other words. score was tested against names of celebrities
i.e. Maria Sharapova got the best score.

Esuli and 2007 For this purpose, they used extended WordNet This means that classifying negative terms is
Sebastiani to build a graph where each synset has certain a harder task. As a conclusion, they see that
polarity depending on the polarity of its this type of model can be applied to other
members. The main hypothesis is that there cases related with semantic properties of
won’t be huge variations and each synset will words.
have a similar degree of negativity. This will
produce a graph of relation between different
synsets that will transfer its polarity properties
to its neighbors.

K. Cai 2008 The opinion classification element The sentiment subject recognition module
differentiated the comparative sentiment identifies the important areas implied beyond
expressed by the terms in all fragments and every sentiment group by word support
then partitioned the fragments into positive, metrics.
negative, and neutral groups
M. 2012 The proposed approach integrated the pair of Their primary experimental assessment on
Eirinakiet opinion mining algorithms. The outlooks are numerous patron review data sets has
based on features and the position of these exposed that their findings achieved
outlooks is also substantially built on the extremely high level of accuracy.
features as a substitute of an object as a whole.
Inhabitants appear to dislike a precise object
as of several features allied with the result.
Table 1:- Table of Comparison

III. CONCLUSION REFERENCES

Sentiment analysis is a ‘suitcase’ field of research that [1]. J. M. Wiebe, R. F. Bruce, and T. P. O’Hara,
contains numerous diverse disciplines, not just associated “Development and use of a Gold-standard Data Set
to computer science but also to communal disciplines, such for Subjectivity Classification.” Proceeding of the
as psychology, philosophy, and ethics. The sentiment 37th Annual Meeting of the Association for
analysis methods which are proposed so far have various Computational Linguistics on Computational
steps. In the pre-processing stage, the missing and Linguistics, USA, pp. 246-253, 1999.
redundant values are removed from the dataset. The feature [2]. D. Pelleg and A. Moore, “X-means: Extending K-
extraction method established relationship between means with Efficient Estimation of the Number of
attribute and target set. In the last step of classification, the Clusters,” in Proc. of the 17th Int. Conference on
classification method is enforced which can categorize data Machine Learning, San Francisco, USA, pp. 727-734,
into certain classes like positive, negative and neutral. 2000.
[3]. B. Pang, L. Lee, and S. Vaithyanathan, “Thumbs up?
Sentiment Classification using Machine Learning
Techniques,” Conference on Empirical Methods in
Natural Language Processing, USA, pp. 79-86,2002.

IJISRT20SEP653 www.ijisrt.com 915

Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[4]. E. Riloff and J. Wiebs, “Learning Extraction Patterns [18]. A. Beygelzimer, J. Langford, and B.Zadrozny,
for Subjective Expressions,” Conference on Empirical “Machine Learning Techniques- Reductions between
Methods in Natural Language Processing, Japan, Prediction Quality Metrics,” Performance Modeling
pp.105-112, 2003. and Engineering Springer, pp. 3-28, 2008.
[5]. T. Wilson and J. Wiebe, “Annotating opinions in the [19]. K Cai, S. Spangler, Y. Chen, L. Zhang, “Leveraging
world Press,” 4th SIG dial Workshop on Discourse Sentiment Analysis for Topic Detection,”
and Dialogue, Sapporo, Japan, pp. 13-22, 2003. International Conference on Web Intelligence and
[6]. K. Dave, S. Lawrence, and D. M. Pennock,“Mining Intelligent Agent Technology, pp. 265-271, 2008.
the peanut gallery: Opinion extraction and semantic [20]. M. Annett, G. Kondrak, “A comparison of sentiment
classification of product reviews,” Proceedings of analysis techniques: Polarizing movie Blogs”, In
WWW, 2003, pp. 519–528. Canadian Conference on AI, pp. 25–35, 2008.
[7]. Hu and Liu, “Mining and Summarizing Customer [21]. B. Pang and L. Lee, “Opinion mining and sentiment
Reviews,” in International Conference on Knowledge analysis,” Foundations and Trends in Information
Discovery and Data Mining, Seattle, USA, pp. 168- Retrieval 2(1-2), 2008, pp. 1–135.
177, 2004. [22]. A. Abbasi, H. Chen, and A. Salem, “Sentiment
[8]. B. Pang and L. Lee, “A sentimental education: analysis in multiple languages: Feature selection for
Sentimental analysis using Subjectivity opinion classification in web forums,” In ACM
Summarization based on Minimum cuts,” Proceeding Transactions on Information Systems, vol. 26 Issue 3,
of the 42nd Annual Meeting on Association for pp. 1-34, 2008.
Computational Linguistics, USA, pp. 271-278, 2004. [23]. A. Agrawal, F. Biadsy, and K.R. Mckeown,
[9]. S. M. Kim and E. Hovy, “Determining the Sentiment “Contextual Phrase-Level Polarity Analysis using
of Opinions.” Proceedings of the 20th International Lexical Affect Scoring and Syntactic n-grams,”
Conference on Computational Linguistics, USA, pp. Proceeding of the 12th Conference of the European
1367-1373, 2004. Chapter of the Association for Computational
[10]. A. M. Popescu and O. Etzioni, “Extracting Product Linguistics, Athens, Greece, pp. 24-32, 2009.
Features and Opinions from Reviews,” Conference on [24]. Q. Ye, Z. Zhang, and R. Law, "Sentiment
Human Language Technology and Empirical Methods classification of online reviews to travel destinations
in Natural Language Processing, British Columbia, by supervised machine learning approaches", Expert
pp. 339-346, 2005. Systems with Applications, vol. 36, pp. 6527-6535,
[11]. T. Wilson, J. Wiebe, and Paul Hoffmann, 2009
“Recognizing Contextual Polarity in Phrase-level [25]. B. Liu, “Handbook of Natural Language Processing,”
sentiment analysis,” Proceedings of the conference on Chapter Sentiment Analysis and Subjectivity, Second
human language technology and empirical methods in edition, ISBN 978-1420085921, pp. 1-38, 2010.
natural language processing, USA, pp. 347-354, 2005. [26]. Bing Liu, “Sentiment Analysis: A Multi-Faceted
[12]. M. Chau and J. Xu, “Mining communities and their Problem,” Journal of IEEE Intelligent Systems,
Relationships in Blogs: A study of online hate vol.25, issues 3, pp. 76-80, 2010.
groups,” International Journal of Human – Computer [27]. S. Li, H. Zhang, W. Xu, G. Chen and Jun Guo,
Studies, vol. 65, issue 1, pp. 57-70, 2007. “Exploiting Combined Multi-level Model for
[13]. N. Godbole, M. Srinivasaiah, and S. Skiena, “Large- Document Sentiment Analysis,” International
Scale Sentiment Analysis for News and Blogs,” Conference on Pattern Recognition, pp. 4141-4144,
International Conference on Weblogs and social 2010.
Media, USA, pp.21-24, 2007. [28]. K. Xu, S. S. Liao, J. Li, Y. Song, “Mining
[14]. A. Esuli and F. Sebastiani, “PageRanking WordNet comparative opinions from customer reviews for
Synsets: An Application to Opinion Mining,” 45th Competitive Intelligence,” Decision Support Systems,
Annual Meeting-Association for Computational vol. 50, issue 4 , pp. 743–754, 2011.
linguistics, Prague, Czech Republic Vol.45, pp. 424- [29]. R. Lau, W. Zhang, P. Bruza and K. Wong, “Learning
431, 2007. Domain-specific Sentiment Lexicons for Predicting
[15]. B. Pang and L. Lee, “Opinion Mining and Sentimental Product Sales,” IEEE International Conference on e-
Analysis,” Foundations and Trends in Information Business Engineering, pp. 131-138, 2011.
Retrieval, USA, vol.2, issue 1-2, pp. 1-135, 2008. [30]. L. Zhang, R. Ghosh, M. Dekhil, M. Hsu, and B. Liu,
[16]. Liu B., “Opinion Mining and Summarization,” World “Combining Lexicon-based and Learning-based
Wide Web Conference Beijing, China, 2008, Methods for Twitter Sentiment Analysis”, Technical
Downloaded from: report, HP Laboratories, 2011.
https://www.cs.uic.edu/~liub/FBS/opinion-mining- [31]. Ji Fang and Bi Chen, “Incorporating Lexicon
sentiment-analysis.pdf [21st June 2016] Knowledge into SVM Learning to Improve Sentiment
[17]. P. Turney, “Thumbs Up or Thumbs Down? Semantic Classication”, In Proceedings of the Workshop on
Orientation Applied to Unsupervised Classification of Sentiment Analysis where AI meets Psychology
Reviews,” Proceedings of the 40th Annual Meeting on (SAAIP), pages 94–100, 2011.
Association for Computational Linguistics, USA, pp.
417-424, 2008.

IJISRT20SEP653 www.ijisrt.com 916

Volume 5, Issue 9, September – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[32]. M. Karqmibekr and A.A. ghorbani, “Sentiment
Analysis of a Social Issues,” International Conference
on a Social Informatics, USA, pp. 215-221, 2012.
[33]. M. Eirinaki, S. Pisal, J. Singh, “Feature-based opinion
mining and ranking,” Journal of Computer and
System Sciences, vol. 78, issue 4, pp. 1175- 1184,
2012.
[34]. E. Haddia, X. Liua and Y. Shib, “The Role of Text
Pre-processing in Sentiment Analysis,” Information
Technology and Quantitative Management, vol. 17,
pp. 26-32, 2013.
[35]. K. Ghag and K. Shah, “Comparative Analysis of the
Techniques for Sentiment Analysis,” International
Conference on Advances in Technology and
Engineering (ICATE), pp. 1-7, 2013

IJISRT20SEP653 www.ijisrt.com 917