International Journal of Engineering and Technical Research (IJETR)

ISSN: 2321-0869 (O) 2454-4698 (P), Volume-6, Issue-1, September 2016

Sentiment Analysis on Unstructured Social Media

Data Compare with Different Classification
Algorithms
Vijay Kumar Mishra, Dr. Neelendra Badal

media to get and exchange information. Both buyers and

Abstract Presently, analysis of opinions from social media sellers can
(Twitter) has become very popular merely because such an
amount of views is difficult to extract through any other mutually create and promote brands to benefit one another.
existing custom means of collecting views like surveys, polls etc.
Social media has evolved to be the center for instant sharing
The analysis is interesting but at the same time challenging
because of a micro blog post generate information on Internet of information. The information can shared millions of
and number of opinions can be expressed which are usually people, who are connected through social media. The power
very short and colloquial and choosing the best opinion mining of its reach to large number of people immediately and the
algorithms is very difficult in such type of text. Therefore, we openness in sharing experiences, without fear, has emerged
propose a new system which automatically analyzes the the valuable suggestion of business and social entities.
sentiments of these types of messages using major classification Social media has today served as a catalyst for on-line chat
algorithms. Here we consider Twitter for the task of sentiment where individuals create content, share it, bookmark it and
analysis which can play significant role on the popularity of the network at a rapid rate. Social media is fast changing the
products and services. Hence, an accurate method for
public opinions in society and setting trends and agendas in
predicting sentiments could enable us to understand customers'
preferences, their views on the product and services offered by topics that range from nature and governmental issues to
the companies. Companies and organization can use this innovation and the diversion business. Since social media
information to formulate the future planning. These can can also be inferred as a form of collective wisdom, we
possibly make positive or negative wave in the business sectors decided to inquire its power at predicting real-world
as well as social sectors. The primary center of my proposed outcomes. Amazingly we discovered that the communication
work is to analyze the emotions communicated on social of people can indeed be used to make quantitative forecast
networking Twitter so that peoples feelings, habits and choices that beat those of fake markets. These data advertises for the
are extracted, investigated and used to understand the behavior
most part include the exchanging of state-unexpected
of the people.
securities, and if sufficiently huge and legitimately planned,
Index Terms Twitter, Sentiment Analysis, Machine they are typically more exact than different systems for
Learning Algorithms, Movies and Songs. removing diffuse data, for example, overviews and
suppositions surveys. In particular, the costs in these
business sectors have been appeared to have solid
I. INTRODUCTION connections with watched result frequencies, and in this
manner are great pointers of future results. [1].
With the rapid growth of mobile information systems and
In the case of social media, the volume and high
the increased availability of smart phones, social media has
fluctuation of the data that produce through huge client
become a integral part of daily life in most societies. This
groups introduces an open door for bridling that information
improvement has involved the creation of huge amounts of
into a structure that consider particular expectations about
information: information which when analyzed can be used
specific results, without instituting market components. One
to extract valuable information about a variety of subjects.
can also build models to collect the opinions of the
aggregate population and gain useful insights into their
People are able to gather the relevant information and are
behavior, while predicting future trends. Moreover,
able to share the same on social web. The Web is a virtual
gathering information on how individuals talk with respect
environment where people are able to put their experience
to specific products can be useful when designing marketing
about the products before buying, the perception of being
and promoting efforts [2].
present rather than actually being there in a real
This paper is the classification of sentimental movie review
environment. Marketing communication strategies have
using to experiment using different machine learning
always been around. It combines the prospect of overcoming
algorithms to predict the sentiment of movie and songs
public resistance with significantly lower costs and faster
reviews. The purpose of this work is to analyze the tweets of
delivery. A large no people are now availing access to social
movies and songs. This paper presents the results of the
research work performed on Twitter to classify Twitter
messages of movies and songs as positive sentiment,

Vijay Kumar Mishra,Assistant Professor, Department of Computer negative sentiment and neutral statements.
Application, Feroze Gandhi Institute of Engineering And
Technology,,RaeBareli , India.
Dr. Neelendra Badal, Associate Professor, Department of Computer
Science and Engineering, Kamla Nehru Institute of Technology,Sultanpur ,
India.

100 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-6, Issue-1, September 2016
II.LITERATURE REVIEW models taking the activity level of editors on Wikipedia as a
Although Twitter has been very popular as a web service, popularity parameter [14].
there has not been considerable published research on it.
Huberman and others [3] studied the social interactions on III. PROPOSED METHODOLOGY
Twitter to reveal that the driving process for usage is a In the case of social media, the outrageousness and high
sparse hidden network underlying the friends and followers, variance of the information that propagates through large
while most of the links represent meaningless interactions. user communities presents an interesting opportunity for
Java et al [4] investigated community structure and isolated using that data into a form that allows for specific
different types of user intentions on Twitter. Jansen and predictions about particular results, without having to
others [5] have examined Twitter as a mechanism for word- institute market mechanisms. Specifically we consider the
of-mouth advertising, and considered particular brands and task of predicting box-office sentiments for movies and song
products while examining the structure of the postings and using the information post on Twitter, one of the fastest
the change in sentiments. However the authors do not growing social media in the Internet. Twitter, a micro-
perform any analysis on the predictive aspect of Twitter. blogging social media site, has experienced a full of
There has been some prior work on analyzing the popularity in recent years leading to a huge user-base,
correlation between blog and review mentions and consisting of several tens of millions of users who actively
performance. Gruhl and others [6] showed how to generate participate in the creation and propagation of infofmation.
automated queries for mining blogs in order to predict spikes Our Methodology divided in several blocks.
in book sales. And while there has been research on
predicting movie sales, almost all of them have used meta-
data information on the movies themselves to perform the
forecasting, such as the movies genre, MPAA rating, running
time, release date, the number of screens on which the movie
debuted, and the presence of particular actors or actresses in
the cast. Joshi and others [7] use linear regression from text
and metadata features to predict earnings for movies. Mishne
and Glance [8] correlate sentiments in blog posts with movie
box-office scores. The correlations they observed for Figure 1 Proposed system block diagram
positive sentiments are fairly low and not sufficient to use
for predictive purposes. Sharda and Delen [9] have treated A. Data Set
the prediction problem as a classification problem and used The dataset that we used was obtained by crawling hourly
neural networks to classify movies into categories ranging feed data from Twitter.com. To ensure that we obtained all
from flop to blockbuster. Apart from the fact that they are tweets referring to a movie and songs, we used keywords
predicting ranges over actual numbers, the best accuracy that present in the movie title and songs title as search
their model can achieve is fairly low. Zhang and Skiena [10] arguments. We extracted tweets over frequent intervals using
have used a news aggregation model along with IMDB data the Twitter Search Api., thereby ensuring we had the
to predict movie box-office numbers. We have shown how timestamp, author and tweet text for our analysis. With an
our model can generate better results when compared to intention to discover the social media signals that potentially
their method. possess stronger correlation with the profitability of a film;
Akcora et al., [11] in their experiment, try to identify the we identify signals which reflect audience approval from
emotional pattern and the word pattern that claims to change different social media domains. Note that the type of data in
the public opinion, using Twitter data. To identify the each domain may be different, e.g., Twitter is a social stream
breakpoint, researchers use Jaccards similarity of two whereas YouTube is a social video publishing website. Most
successive intervals of words. Sun et al., [12] study fan of social media buzz around a movie that have been captured
pages on Facebook to understand diffusion trees. Kwak et for this study are before its release or during the first 1-2
al., [13] compare the number of followers, page-ranks and weeks.
the number of re-tweets as three different measures of
B. Classification of Data
influence. Their finding is that the ranking of the most
influential users differed depending on the measure. Data can be classified into different categories depending on
Movies and songs are released across different parts of the the domain i.e.
world and Twitter users are also from different parts of the Political, Business, Sports etc.
world. The study aims to analyze whether an attitude in the Recommendations, Complaints
tweets is positive or negative sentiment or a cognitive Positive, Negative
statement, understand the flow of interpersonal messages Lexicon based classification
across different countries and understand people behavior. Requires a dictionary of words and their polarity
Many previous studies have shaped the goal of our work. scores.
Ishii et al. developed a mathematical framework for the Supervised Learning classification
spread of popularity of the movie in society. Their model Requires training data to create a classification
considers the activity level of the bloggers estimated through model.
number of weblog posts on particular movies in the Japanese Various visualization techniques
Blogosphere as a representative parameter for social WEKA can be applied to classification of data.
popularity. Similarly, other researchers have developed

101 www.erpublication.org
 Sentiment Analysis on Unstructured Social Media Data Compare with Different Classification Algorithms

C.System Process understanding of consumers and selecting strategies to

Opinion mining or sentiment analysis is the process of influence them. In order to understand how consumers talk
determining the feelings expressed by an individual in his about the movies, sentiments are measured based on the
writing. There are two basic methods that exist; the first is following classification: Positive Sentiment, Negative
the document level and the second is the sentence level. In Sentiment and Cognitive Statement.
the document level, the analysis is based on the complete Here for preprocessing, filtering, visualization Weka API is
document, where as in the sentence level, the analysis is used, it is a popular suite of machine learning software
performed at the sentence level. Since tweets comprise only written in Java. All algorithms that software provides can be
140 characters, we have used the methods that are related to used directly from code by importing weka.jar file. Weka
the sentence level. The following tools and methods will be contains tools for data preprocessing, classification,
developed to address the objective: regression, clustering and visualization.
An in-house tool using Twitter API will be Precision represent proportion of samples that are truly of a
developed to download related tweets from Twitter class divided by the total sample classified as that class.
database in an automated manner. Recall represents proportion of samples classified as a
A Sentiment Analyzer tool will be developed using given class divided by the actual total in that class.
python/java and natural language tool kit libraries F-Measure which is a combined measure for precision and
by trying different supervised machine learning recall calculated as
algorithms. The best classifier model will be chosen F-measure = 2 * Precision * Recall / (Precision + Recall)
for the final classification.
The collected tweets are classified, using the above III. EXPERIMENTAL ANALYSIS AND RESULTS
classifier, into three different classes positive The results are used to compare with machine learning
sentiment, negative sentiment and natural algorithms of sentimental analysis of social data of movies
statement. and songs. For the better results, we have to measure its
Unwanted tweets that fell into none of the above performance by applying suitable measures.
mentioned classes are classified using any filter. This work, we have used to different machine algorithms.
This project is the classification of sentimental movie review These algorithms are used to sentiment analysis of tweeted
using to experiment using machine learning algorithms to data of movies and songs. For sentiment analysis, we search
predict the sentiment of movie reviews. This prediction was the movie name as like X-man, and then we recolonized the
performed on the movie review dataset [15]. tweet data about movie and song.

IV. MACHINE LEARNING ALGORITHMS

D.Support Vector Machine

Support vector machine is a supervised learning model, very

similar to linear regression, which analyzes data, recognizes
patters and uses these results for predictions. Here we used
the SVM library to use the complete functionality of the
algorithms. The advantage of Support Vector Machines is
that they can make use of certain kernels in order to
transform the problem, such that we can apply linear
classification techniques to non-linear data.
E. J48 Decision Tree
J48 Decision tree is a predictive machine-learning model
that decides the target value of a new sample based on
various attribute values of the available data.
F. Logistic Regression
Figure 2 Entered Movies Name
Logistic Regression is one of the best probabilistic
classifiers, measured in both log loss and first-best After collection of tweet data, we applied machine
classification accuracy across a number of tasks. The algorithms for analyzing popularity.
dimensions of the input vectors being classified are called
"features" and there is no restriction against them being
correlated.

II. TRAINING AND TESTING THE CLASSIFIERS

Consumer behavior involves the thoughts and feelings
people experience and the actions they perform in the
consumption and usage of a product. Consumer thinking,
feelings, actions are constantly changing. The Wheel of
Consumer Behavior is critical for developing a complete

102 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-6, Issue-1, September 2016

Figure 3 Output of Tweet data

This work applied as different sentiment algorithms for

analysis. For analytical comparisons, we make use of Pie
Chart and Bar Graph.
Figure 6 Pie-Chart after applying Logicstic reression

Figure 4 Basic Pie-Chart without apply any algorithm

Figure 7 Pie-Chart after applying SVM Algorithm

Performance compression of SVN, Logistic Regression

and J48 Decision tree algorithms using WEKA:

Table 1 Performance compression of algorithms with different

parameters

Logistic
Parameters SVM Regression J48
TP Rate 0.501 0.752 0.372
FP Rate 0.11 0.350 0.262
Precision 0.82 0.752 0.362
Recall 0.501 0.752 0.372
F- Measure 0.622 0.752 0.37
ROC Area 0.696 0.807 0.552
Figure 5 Pie-Chart after applying J48 algorithm

103 www.erpublication.org
 Sentiment Analysis on Unstructured Social Media Data Compare with Different Classification Algorithms

[9] Asur S, Huberman BA,Predicting the future with social media in

Proceedings of the IEEE/WIC/ACM International Conference on
Performance of Positive Class Web Intelligence and Intelligent Agent Technology, DC, pp. 492-
499, 2010.
1 [10] Joshi M, Das D, Gimpel K, Smith N,Movie reviews and revenues:
0.8 An experiment in text regression, in Proceedings of NAACL-HLT,
PA, pp. 293-296, 2010.
0.6 [11] Akcora, C. G., Bayir, M. A., Demirbas, M., and Ferhatosmanoglu,
0.4 H., Identifying breakpoints in public opinion, In ACM Proceedings
Value

of the First Workshop on Social Media Analytics, July, pp. 62-66.

0.2 2010.
[12] Sun, E., Rosenn, I., Marlow, C. and Lento, T. (2009), Gesundheit!
0 modeling contagion through Facebook news feed, In Proc. Of
FP Rate Recall ROC Area International AAAI Conference on Weblogs and Social Media, May,
TP Rate Precision F- Measure pp. 22.
[13] Kwak, H., Lee, C., Park, H. and Moon, S. (2010), What is twitter, a
Parameter social network or a news media?, In ACM Proceedings of the 19 th
International Conference on World Wide Web, April, pp. 591-600.
[14] Mestyn, M., Yasseri, T., and Kertsz, J.,Early Prediction of Movie
SVM LR J48 Box Office Success based on Wikipedia Activity Big Data, arXiv
preprint arXiv:1211.0970, 2012. Science and Technology, Vol. 3,
Figure 8 Performance comparisons of SVN, Logistic Regression and issue 3, pp. 1878-1884, March 2013.
J48 Decision tree algorithms with different parameters [15] http://www.cs.cornell.edu/people/pabo/movie-review-data/review
polarity.tar.gz

IV. CONCLUSION Vijay Kumar Mishra Assistant Professor of Department of Computer

Application, Feroze Gandhi Institute of Engineering and Technology,
The study attempts to examine the use of micro-blogging as Raebareli, India.
a communication channel. The messages expressed in Dr. Neelendra Badal is a working as Associate Professor of
Twitter micro-blogging can be related to human behavior. Department of Computer Science and Engineering, Kamla Nehru Institute
People can express either positive or negative sentiments of Technology, Sultanpur, India.
and also information can be neutral in nature. Different
regions express different sentiments depending on the nature
of the movie and how the movies impact cultural sentiments.
After applying different machine algorithms, we analyze the
sentiment behavior expression though the overall sentiment
behavior expressed is positive, people have also expressed
negative sentiments, which cannot be ignored. Classification
of the tweets into positive sentiments, negative sentiments
and neutral statements specify the extent of varied consumer
view on any specific aspect which provide important and
useful input to various enterprises in formalizing their
strategies and carrying out course corrective actions with
respect to their strategies. The extents of the inputs /
feedback through Tweets are far larger than any other
normal means of collecting customer feedback.

REFERENCES
[1] Bruns, A., The Active Audience: Transforming Journalism from
Gatekeeping to Gatewatching. In Making Online News: The
Ethnography of New Media Production. Eds. Chris Paterson and
David Domingo. New York: Peter Lang, 2008.
[2] Boyd, Danahd and Ellison, N.. Social Network Sites: Definition,
History and Scholarship. Journal of Computer-Mediated
Communication, 13(1), 1, 210-230, 2007.
[3] Bernardo A. Huberman, Daniel M. Romero, and Fang Wu. Social
networks that matter: Twitter under the microscope. First Monday,
14(1), Jan 2009.
[4] Akshay Java, Xiaodan Song, Tim Finin and Belle Tseng. Why we
twitter: understanding microblogging usage and communities.
Proceedings of the 9th WebKDD and 1st SNA-KDD 2007 workshop
on Web mining and social network analysis, pages 5665, 2007.
[5] B. Jansen, M. Zhang, K. Sobel, and A. Chowdury. Twitter
power:Tweets as electronic word of mouth. Journal of the American
Society for Information Science and Technology, 2009.
[6] Karniouchina, Ekaterina, Impact of star and movie buzz on motion
picture distribution and box office revenue, Intern. J. of Research
in Marketing, vol. 28, pp. 62-74, 2011.
[7] Spann, Martin & Bernd Skiera, InternetBased Virtual Stock Markets
for Business Forecasting, Management Science , vol. 49, no. 10, pp.
1310-1326, 2003.
[8] http://www.stateofdigital.com/how-to-recognize-twitter-bots-6-
signals-to-look-out-for

104 www.erpublication.org