A Survey On Sentiment Analysis Of Movie Reviews

© 2014 IJIRT | Volume 1 Issue 7 | ISSN: 2349-6002
IJIRT 101388 INTERNATIONAL JOURNAL OF INNOVATIVE RESEARCH IN TECHNOLOGY 36
A SURVEY ON SENTIMENT ANALYSIS OF
MOVIE REVIEWS
Neha Nehra
M.E (CSE), L.J Institute of Engineering and Technology,
Gujarat Technological University, Gujarat, India
Abstract- With the help of technology, the internet
becomes a valuable place for exchanging ideas, online
learning, reviews for a product or service or movies.
It makes hard to record and understand the user
emotion because reviews over the internet are
available for millions for a product or services.
Sentiment analysis is an emerging area for research
to collect the subjective information in source
material by applying Natural Language processing,
Computational Linguistics and text analytics and
categorised the polarity of the opinion or sentiment.
In simple words we say that sentiment analysis is
important for decision making process. This paper
provides an overall survey about sentiment analysis
or opinion mining related to movie reviews.
Index Terms- Opinion Mining, Movie Reviews,
Sentiment Analysis, tokenization.
I. INTRODUCTION
Sentiments are nothing but emotions of the user. It
may be good, excellent, bad or neutral. Analysis of
such emotions is known as sentiment analysis. In
other words we can say that, it is language
processing task that uses computational approach to
identify the opinion of user and classify it as
negative, positive or neutral. Web contains the
unstructured textual information that often carries
opinion or sentiments of user. The analysis of
sentiment tries to identify the mood of writers and
expressions of opinion. A simple method of
sentiment analysis categories reviews of user’s as
positive, negative or neutral. When review of user
expresses a positive opinion than it is denoted by
positive label and in similar way if review
expresses a negative opinion than it comes under
negative label.
Simple sentiment analysis methods are used to
classify a document as positive or negative, based
on document opinion that is expressed in it. For
example D is given set of documents and d is
document present in D, i.e. d belongs to D,
sentiment analysis method categories each d
document into three classes, positive, negative and
neutral. The methods or algorithm that identifies
sentiments at sentence-level and feature-level or
identity-level are sophisticated one. There are three
levels where sentiment analysis is performed and
they are:
a. Document level
b. Sentence level
c. Entity or feature or aspect level
Document level: For a product or service, the
entire document opinion is classified into a
positive, negative or neutral sentiment and this is
document level sentiment analysis.
Sentence level: For a product or service, to
determine whether each sentence expresses a
positive, negative or neutral opinion and this is
sentence level sentiment analysis. This type is used
with reviews and comments that contain one
sentences and written by user.
This is performed by two tasks: subjective or
Objective. Objective: I purchase XYZ mobile few
days ago. Subjective: It is such a perfect phone.
Entity or feature or aspect level: The opinion
mining and summarization based on feature is also
known as Aspect level. This type is used when we
need sentiments about desired aspect/feature in a
review.
II. LITERATURE SURVEY
Shravan Vishwanathan et al., [1] proposed Reviews
of rotten tomato is collected from the one of the
database. Than on each review tokenization is
done, filter the tokens by the length. After that
stemming is performed and then remove tokens
which are not required for the sentiment analysis.
Multiply operator is used which compare each
token with the positive word dictionary and
negative word dictionary. If given token matches
with any of the word dictionary than token is
categorized into that category. After that sum all
the occurrence at both positive database and
negative database. Apply join operator which
subtract the positive sum and negative sum and
generate the class label of review and display it to
the user

Santanu Modak et al., [2] in this paper studied is
done on different approaches for sentiment
classification. So that information is used for the
future research. Fuzzy Sets or fuzzy [3]
classification method is used for Opinion Mining or
sentiment analysis. In this method fuzzy set is
prepared which is used to calculate the degree of
positive and negative of sentiment words.
Su, Qi, et al. proposed mutual reinforcement
approach to deal with the feature-level opinion
mining problem [4]. Clustering was done on
product features and opinion words simultaneously
and iteratively by fusing both their content
information and sentiment link information. They
constructed the sentiment association set between
the two groups of data objects by identifying their
strongest n sentiment links. POS tagger used to
detect sentiment word and product features. Using
sentiment word and product features they derived
association rule to detect hidden sentiment. Finally
sentiment scoring was done.
This study concludes that, if Sentiment Analysis is
a regression type problem, then we can choose
fuzzy set, which is one of the best techniques for
this purpose. If we consider Sentiment Analysis is a
classification type problem then we can choose
semi-supervised learning or supervised machine
learning approach. Small dataset is used for
training in semi-supervised approach. Classifier is
used for supervised machine learning approach.
Out of all classifiers, Maximum Entropy Classifier
produces overall good result, but Support Vector
Machine (SVM) produce best result all time.
Vivek Kumar Singh et al., [5] in this paper
evaluation of sentiment analysis method is done
for both supervised classifiers of machine learning
and lexicon based unsupervised. This comparison
is performed at document level. The result shows
that SVM and NB are better than SentiWordNet
approaches. Performance is calculated based on the
three factors: accuracy, F- measure and entropy.
I found that machine learning classifiers are
suitable method for aspect level sentiment analysis
and reason is there is lack of availability of training
data. That why, it is better to use unsupervised
SentiWordNet approach. This method uses two
different ways for the sentiment analysis. The first
one is combination of adverb and adjective, and
adverbs are consider as modifiers. The second one
is combination of adverb adjective and adverb verb.
For aspect level first method of SentiWordNet is
used and aspect level is used for detail
classification. Example are, music of movie, story
of movie, acting of actors and direction.
Khin Phyu Phyu Shein et al., [6] on the Internet
there are lots of content that opinion or sentiments
about an object such as review about music, movie,
software, product and books etc. The aim of
sentiment classification is to extract the feature on
which reviewer express their emotion or feeling
and identify them as positive, negative or neutral.
In this paper, proposed model is the combination of
Support Vector Machine with Natural Language
Processing techniques, ontology based on Formal
Concept Analysis design for classifying the
software reviews are negative, positive or neutral.
In it’s proposed model main focus is on feature
level sentiment classification. the three main parts
in this approach are: assigning the POS tags,
identifying domain related features and classifying
the sentiment words. They use Part Of Speech
(POS) tagger to assign
V.K. Singh et al., [7] used a SentiWordNet based
approach with two different linguistic feature. The
SentiWordNet approach for document level
sentiment classification of movie reviews and blog
posts is implemented in this paper and performance
evaluation is also made. With different variations
of linguistic features, aggregation threshold and
scoring schemes, SentiWordNet is implemented.
Here it uses two approaches: i) SVM (Support
Vector Machine) and Naïve Bayes for classification
of sentiments.
In this we have taken four SentiWordNet
approaches for two blog posts and two movie
reviews datasets. Where SWN-1 (first method of
simple aggregation of adjective scores in
SentiWordNet), SWN-2 (second method of
aggregation of adverb adjective scores in
SentiWordNet), SWN(VS)- Variable scoring and
SWN(ASP)- Adjective priority scoring method.
1400 reviews for first movie,2000 reviews for
second movie, 1486 blog posts on Libyan
Revolution and 807 blog posts on Tunisian
revolution. Result shows that machine learning
algorithms are better than SentiWordNet
approaches.
V.K. Singh et al., [8] presents new feature based
domain specific heuristic for aspect level sentiment
classification for reviews of movie. In this method
it analyzes the movie textual reviews and then on
each aspect assigns its sentiment label. After that it
aggregate the scores on each aspect from various
reviews and on all parameters net sentiment profile

is generated. In this author also uses the
SentiWordNet with different linguistic features
such as adverbs, adjectives and verbs.
SentiWordNet is used with document level
sentiment classification with two linguistic
features. SentiWordNet is publicly available library
that contain scores of each word and based on the
score we classify the reviews as positive, negative
or neutral opinion. The two linguistic features are:
i) combination of adverb and adjective and ii)
combination of adjective adverb and adverb verb.
This is used for producing better results. Aspect
level is used when we consider specific feature of
movie like, direction, acting, music, etc.
Using combination of adverb+adjective gives better
result as compare to using only adjectives, because
adverbs increase the score and we can say that they
play the role modifier. When we combine scores of
adverb verb combination with scores of adjective
adverb combination than it improves the accuracy
of sentiment classification. The more accurate or
focused sentiment summary of particular movie is
produce by aspect-level sentiment profile.
Limitation of aspect level sentiment classification
is that it is domain specific. In aspect vector if
make small changes, than it can be used in different
domain.
Kang Wu et al., [9] focus on sentiment analysis of
topical Chinese microblogs. In this paper most
popular microblog of China is taken i.e. Sina
WeiBo. User of WeiBo writes their messages that
contain usually various sentences, messages length
is up to 140 Chinese Microblog contain several
sentences, which allow users to share their opinion.
Study shows that Chinese people express their
sentiments in indirect way. For classification of
such sentiments we need more semantics. The
proposed model first, analysed the Chinese
Microblogs which express the opinion of user, and
analysis of features of single sentence. Second, to
optimize the result of sentiment classification we
use sentence relationship.
Asha S Manek et al., [10] proposed a model for
detecting spamming activities such as writing fake
reviews about a product to mislead the users. This
model uses efficient Repetitive Pre-processing
(SentReP) which is based on focused pre-
processing and tested parameters for categorizing
the reviews. To obtain “list-of-words” movie
reviews are pre-processed. After that each review
under go the following steps: tokenization, case
transformation, porter and snowball stemming
process and then stop words are removed. After
pre-processing cross validation is performed which
consist of two steps: i) each attribute weight
calculation and ii) by weight select top K attributes.
Mostafa Karamibekr et al., [11] Sentiment analysis
have done only for product, services or movies not
for the social issues. For government work, it is
necessary to know the public opinions regarding
social issues. So, first we should know how social
issues are different from product and services. The
difference is that it is easy to define features for
product, but not for social issues. In social area,
verb play vital role to express opinions of user. In
sentiment analysis of social issues first, from each
sentence we collect the opinions, construct opinion
structures, and then their orientations are
determined regarding social issues.
Martin Wollmer et al., [12] proposed method
performer sentiment classification for audio plus
video reviews of user. Review for a movie is given
in 2 minute YouTube video. For sentiment
classification of such reviews method use
automatic speech recognition system and video
recognition system. For better classification of
reviews vocal and face expression play vital role.
Richard Socher et al., [13] shows that semantic
word space are very useful but they can’t used with
long sentences. That why, Sentiment Treebank was
introduced. This Treebank consist of various parse
trees to classify the sentence into the one of classes
of sentiments. Recursive Neural Tensor network is
the example of such method.
One example is taken, to understand how this
method works. Example review is “This film
doesn’t care about cleverness, wit or any other kind
of intelligent humours. It divide the sentence into
token and make tree structure which divide the
comment into one of the class label. sentence is
taken and then using Treebank concept it is
accurately classify into one of five classes. Five
class labels are very negative (- - ), negative (-),
neutral (0), positive (+), and very positive (+ +).
The figure 1, showing one example of recursive
neural network. In this figure we can easily
understand the how this sentiment Treebank
works.

Figure 1: Example of Recursive Neural Network
[13]
III. CONCULSION
In this literature survey it is seen that for decision
making process about product, service, movie,
social issues, sentiment analysis or opinion mining
play very important role. Opinion mining is not
only consisting of the concepts of text mining but
also the concepts of information retrieval. For good
classification, feature weighting which play crucial
role is one of the major challenges in opinion
mining. Life without opinion is like an empty
vessel.
Over the Web, social media is one of the major
parts of it. Calculation says that every 9 user out of
10 uses one form social media. Now a day’s user
over internet creates large amount of data. So, for
web content users become co-creators. Over social
media, user contribution ranges from photo and
video uploads, reviews, blog posts and tweets. On
internet the data that is available is unstructured
text.
Over social media, views or opinion are
expressed through user reviews or posts. With
demand growth about the accessibility of opinion
resources such blog reviews, movie reviews,
social network tweets, product reviews, results in
new challenge is to mining large volume of
data/text and required suitable algorithm for
analysis of opinion of others. This is important
for the organizations because these help them to
improve their services or goods and it also help
them for making decision for future.
Understanding opinion of user about goods or
service is not only helpful for companies but also
helpful for users. For example, review about
restaurant in a city help user to find good food in
city. Similarly, hotel review help user to find best
hotel in a city.
IV. REFERENCES
[1] Shravan Vishwanathan, “Sentiment Analysis
of Movie Reviews”, Proceedings of 3rd IRF
International Conference, 10th May-2014,
Goa, India.
[2] Santanu Modak, Abhoy Chand Mondal, “A
Study on Sentiment Analysis”, International
Journal of Advanced Research in Computer
Science and Technology, Volume 2, Issue: 2,
Version 2, April-June 2014.
[3] Jusoh, S. Alfawareh, H.M., “Applying fuzzy
sets for opinion mining”, Computer
Application Technology (ICCAT), 2013
International Conference on, vol., no.,
pp.1,5,20-22 Jan, 2013 doi:
10.1109/ICCAT.2013.6521965.
[4] Su, Qi,et al, “Hidden Sentiment Association
in Chinese web Opinion mining”, Proceeding
of the 17th
International Conference on World
Wide Web.ACM,2008.
[5] Vivek Kumar Singh, Rajesh Piryani, Pranav
Waila and Madhavi Devaraj, “Computing
Sentiment Polarity of Texts at Document and
Aspect levels”, ECTI Transactions on
Computer and Information Technology Vol.8,
NO.1 May 2014.
[6] Khin Phyu Phyu Shein and Thi Thi Soe Nyunt,
“Sentiment Classification based on ontology
and SVM classifier”, 2010 Second
International Conference on Communication
Software and Networks, IEEE, 2010, DOI
10.1109/ICCSN.2010.35.
[7] V.K. Singh, R.Priyani, A. Uddin and P.Waila,
“Sentiment Analysis of Movie Reviews and
Blog Posts Evaluating SentiWordNet with
different Linguistic Features and scoring
schemes”, 3rd
IEEE International Advance
Computing Conference (IACC), 978-1-4673-
4529-3/12,2012.
[8] V.K. Singh, R.Priyani, A. Uddin and P.Waila,
“Sentiment analysis of Movie Reviews a new
Feature-based Heuristic for Aspect-level
Sentiment Classification”,978-1-4673-5090-
7/13 IEEE, 2013
[9] Kang Wu, Bofeng Zhang, Jianxing Zheng and
Haidong Yao, “Sentiment Classification for
Topical Chinese Microblog Based on
Sentences Relations”, IEEE International
Conference on Green Computing and
Communication and IEEE Internet of Things

and IEEE Cyber, Physical and Social
Computing, 2013.
[10] Asha S Manek, Pallavi R P, Veena H
Bhat, P Deepa Shenoy, M Chandra Mohan,
Veenugopal K R and L M Patnaik, “SentReP:
Sentiment Classification of Movie Reviews
using Efficient Repetitive Pre-
Processing”,978-1-4799-2827-9/13 IEEE,
2013.
[11] Mostafa Karamibekr and Ali A Ghorbani,
“Sentiment analysis of Social Issues”, ASE
International Conference on Social
Informatics,978-0-7695-5038-2/12 IEEE,
2012.
[12] Martin Wollmer, Felix Weninger, Tobias
Knaup, Bjorn Schuller, Congkai Sun, Kenji
Sagae and Louis-Philippe Morency, “YouTube
Movie Reviews: Sentiment Analysis in an
Audio-Visual Context”, IEEE Computer
Society,1541-1672,2013.
[13] Richard Socher, Alex Perelygin, Jean Y.
Wu, Jason Chuang, Christopher D. Manning,
Andrew Y. Ng and Christopher Potts,
“Recursive Deep Models for Semantic
Compositionality Over a Sentiment Treebank”,
Stanford University, Stanford, CA 94305,
USA.

A Survey On Sentiment Analysis Of Movie Reviews

More Related Content

Similar to A Survey On Sentiment Analysis Of Movie Reviews (20)

More from Shannon Green (20)

Recently uploaded (20)

A Survey On Sentiment Analysis Of Movie Reviews