International Conference on Global Trends in Engineering, Technology and Management (ICGTETM-2016)
Study and Review of Sentiment Analysis for Social Networks
using Classification Algorithm
Sandeep. J. Patil1, Prashant. C. Harne2, Pravin.K.Patil3
1
Assistant Professor, Department of Information Technology, SSBT’s College of Engineering & Technology, Bambhori,
Jalgaon, North Maharashtra University, Jalgaon, Maharashtra, India
2
Assistant Professor, Department of Information Technology, SSBT’s College of Engineering & Technology, Bambhori,
Jalgaon, North Maharashtra University, Jalgaon, Maharashtra, India
3
Assistant Professor, Department of Information Technology, SSBT’s College of Engineering & Technology, Bambhori,
Jalgaon, North Maharashtra University, Jalgaon, Maharashtra, India
Abstract: In data mining classification is done with
supervised learning and unsupervised learning. Selection of
algorithm depends upon the type and behaviour of data. The
data can be as structured and unstructured. In data mining
text mining has become an important research area. There
are various applications of text mining which includes
information retrieval, machine learning, data mining, and
statics and computation semantics and many more. In form
of text data most of the information is stored. Research is
going on in a direction of multiple language support. In the
presented work the identified social networking web
applications’ data set is reviewed to perform text analysis.
Therefore the entire input data samples are required to
classify in two classes namely positive and negative. And
hence a binary classifier decision tree and their improved
variant algorithms can be utilized for analysis and
performing the classification task. Prior to classification of
text data there is need to improve the quality of data to get
the good results. It is therefore the raw text data collected
from sources is first pre-processed and then tagged
according to the lexical means. Once tagging is done on the
original text data classification algorithms are then trained
to classify the text according to their moods. Additionally for
finding their performance in terms of their efficiency the
time and space complexity needs to be measured that will
show the effective classification with better performance.
Product Reviews
Sentiment Identification
Feature Selection
Sentiment Classification
Sentiment Polarity
Keywords: sentiment analysis, classification, supervised
Fig 1: Sentimental Analysis Process
learning, unsupervised learning.
I. INTRODUCTION
Data mining is the process of extracting information from
the raw data. Here the information is meant to the data which
is required by the application or a program. Text mining is
the computational analysis of text for knowledge mining and
data pattern analysis purpose. These techniques provide ease
in information extraction such as NLP, and IR. Additionally,
these techniques include domains with many different
algorithms including KDD methodologies and others. There
are some applications of text mining such as Enhancing Web
Search, Mining Bibliographic Data, and Sentiment
Classification or detecting mood. In this proposed work the
text data can be mined for semantic analysis of the text from
raw data. Figure 1 shows the process for sentiment analysis.
Initially text data is unstructured e.g. lack of labelling of data
and it becomes a complicated task.
ISSN: 2231-5381
Hence most of the applications are using the cluster
analysis techniques for categorizing data. However if
the data is well labelled then that can be used with the
classification algorithms easily. Therefore the use of
microblog data can be used with the classification
algorithm.
In today’s age of technology most of the computational
parts such as algorithms as well as applications are
hosted on remotely running servers. Users access the
remote data using information superhighway known as
Internet. This network provides services and
information all the time and therefore it becomes a part
of new generation life. Use of Internet connects us with
the imaginary social world such as twitter, Facebook
and others. In use of social networking web
applications by people, a huge amount of text, image
and video data is generated and put on the Internet to be
shared among its users. Due to this manual analysis of
huge amount of data becomes a challenging task.
Therefore computational algorithms or statistical
techniques need to be applied on these data to find the
http://www.ijettjournal.org
Page 163
International Conference on Global Trends in Engineering, Technology and Management (ICGTETM-2016)
targeted patterns.
In this proposed work the text data for microblog analysis
can be used to do the classification. Today microblogs are
frequently used by users. But frequent use of this data which
is shared by people on the Internet increases the amount of
complexity of data for manual analysis. In this model of text
analysis we propose to provide the output in two steps that
will work on labelled data. First of all the data is processed
in order to obtain the text features and then the learning on
evaluation is performed to measure the features. To identify
specific pattern of data on social networking sites, specific
semantic analysis techniques can be applied to obtain better
results. On the other hand for accurate classification of these
data some traditional data mining techniques are developed
to provide ease in classifying text data help mining.
Therefore the proposed work involves the improved
classification algorithm for classifying the text. This
classification would help us to calculate the sentiments of
users during communication over the social networking web
applications such as twitter.
II. LITERATURE SURVEY
Ziato Liu [1] suggested a new feature selection method
predicate on How Net and Parts of Speech in his paper
“Short Text Feature Selection for Micro-blog Mining”.
According to the composition of text property they utilize
test data set accumulated from sina microblog. The result
shows that the short text feature selection method has a
substantial amount of information, and good classification
result.
Stefan Stieglitz [2], seek to examine whether sentiment
occurring in politically germane tweets has an effect on
their retweet ability. Predicate on dataset of 64,431 political
denoting affective dimension, including positive and
negative emotions associated with certain political parties
or politicians, in a tweet and its retweet rate. Furthermore,
they investigate how political discussion takes place in the
Twitter network during the periods of political elections.
Determinately, authors conclude by discussing the
implicative insinuation of results.
Pravin Patil [3], suggested a combinational approach for
sentiment analysis of twitter messages. The approach
combined the NB classifies with a lexicon basesd approach
to analyse the tweets into positive, negative and neutral.
The results showed improvement of 10% over the
traditional approach.
Apoorv Agarwal [4] examines sentiment analysis on
Twitter data. Their contributions of this paper are: (1) First
introduce POS prior polarity feature. (2) Explore the
utilization of tree kernel to obviate the desideratum for
tedious feature engineering. The incipient feature and tree
kernel perform approximately at same level, both
outperforming the state of art of baseline.
Raymond Kosala, et.al., in [5] surveyed the research in
the area of web mining. It also explores the connection
between the web mining categories and the related agent
paradigm. For this survey, focus is on the representation
issues, on the process, on the learning algorithm and on the
application of the recent works as the criteria. It describes
ISSN: 2231-5381
the research done for the information retrieval giving an
IR view of the unstructured documents. Also, the
information retrieval view for semi-structured
documents is discussed. The database view for the web
content has been explained in detail which mainly tries
to model the data on the web and to integrate them so
that sophisticated queries other than keywords based
search could be performed.
Sandra Stendahl, et.al., in [6] focused on different
implementations on web mining and the importance of
filtering out calls made from robots to get knowledge
about the actual human usage of a website. This is to
find patterns between different web pages and create
more customized and accessible web pages to users,
which in turn creates more traffic and trade to the
website. Also, some common methods to find and
eliminate the web usage made from robots while
keeping browsing data made from human users intact
are addressed. Web mining is viewed as seen to consist
of three major parts: collecting the data, preprocessing
the data and extracting and analyzing patterns in the
data.
D. Jayalatchumy, et.al., in [7] worked on survey on the
existing techniques of web mining and the issues
related to it. It primarily reports the summary of various
techniques of web mining approached from the
following angles like feature extraction, transformation
and representation and data mining techniques in
various application domains. The survey on data
mining technique is made with respect to Clustering,
classification, sequence pattern mining, association rule
mining and visualization. It also gives the overview of
development in research of web mining and some
important research issues related to it. It describes the
process of web cleaning which is needed to remove
noise and correct inconsistencies in the data.
Abdelhakim Herrouz, et.al., in [8] discussed the
overview of different web content mining tools. Web
content mining is the process of extracting useful
information from the web documents. With the flood
of information and data on the Web, the content
mining tools helps to download the essential
information that one would require.
R.Malarvizhi, et.al., in [9], made a comprehensive
study of the various web content mining techniques
tools & algorithms.
Arvind Arasu, et.al., in [10], proposed an algorithm for
structure data extraction.
III. PROPOSED SYSTEM
To perform the sentiment text analysis and to have its
evaluation following methods shown in the figure can
be used.
http://www.ijettjournal.org
Page 164
International Conference on Global Trends in Engineering, Technology and Management (ICGTETM-2016)
Step 6: Test data is applied to perform testing on the
model and prediction on dataset is done [11].
Training Samples
B) Training Samples
The key aim of the system is to classify the tweets
on the social media. So we can use microblog on
any site as data set to perform the experiment.
Pre-processing
C) Test Samples
The available data set can be divided in to training
and testing as per requirement.
Tagging
Feature Estimation
D) Pre-Processing
In both training and test, data is pre-processed. The
pre-processing phase of data involves removal of
punctuations and removal of frequently occurred
words.
Decision Tree Algorithm
Trained Data Model
Training Samples
Classification and Performance
Figure 2 a. Process Model of Proposed System
(Training Samples)
Test Samples
Pre-processing
E) Tagging
It is required to involve feature on data after preprocessing. Therefore the user input tags are applied
with the text such as.
Mango
is a good fruit.
F)
Can be converted into: Noun adjective noun
G) Features Estimation
After tagging the original data is converted into a new
encoded format. Therefore the tagged data and the
associated tag are stored on a relational data base which
contains the encoded attributes and their class labels.
Following table shows sample of featured data.
Table 1: Feature Data
Noun Pro-no Verb Adv Adj Pre
2
0
1
0
1
0
G)
Tagging
G) ID3 Training
The given table data is used to learn the traditional ID3
algorithm. Initially a provision can be made to select
algorithm for training. The system gets training from
traditional ID3 algorithm if we select it. Entropy and
Information Gain are the important factors which are
used to select the most useful attribute for
classification.
Feature Estimation
Test Data
Trained Data Model
Classification and Performance
Figure 2 b. Process Model of Proposed System
(Test Samples)
A) Method
Step 1: Both Test and Training Data set needs to be preprocessed separately.
Step 2: Speech tagging is done on pre-processed dataset.
This is a process of marking up a word and tag with
corresponding function in parts of speech.
Step 3: Every word is replaced with its tag
Step 4: Tagged data and associated tag is stored in relational
database. Traditional ID3 algorithm and Modified ID3
algorithm is applied on tagged datasets which develops a
decision tree model.
Step 5: The data model is generated from the training set for
classification.
ISSN: 2231-5381
Conj
0
H) Improved ID3 Training
If we select the Improved ID3, a decision tree data
model from the training sample is developed.
Association function correlation is the important factor
used to carry out the importance of attribute. AF not
only can well overcome the ID3’s deficiency of tending
to take value with more attributes, but it can also
represent the relations between all elements and their
attributes [1].
AF =∑ |X i1 - X i2 | / n
Then, the normalization of relation degree function
value is followed.
V (K) = AF (k) / AF (1) + AF (2)……..AF (m)
Gain (A) = I(S (1), S (2), S (3)………..S (m)) – E (A)*V (A)
I) Trained Data Model
Trained model is a finalized decision tree that makes
use of input data and converts into a tree structure. This
http://www.ijettjournal.org
Page 165
International Conference on Global Trends in Engineering, Technology and Management (ICGTETM-2016)
trained data model is prepared by ID3 and improved ID3.
J) Test Data
This is part of training sample which is used to perform
testing of trained data model using the cross validation
technique. The cross validation results the accurate amount
of data that is correctly recognized using the decision tree.
K) Classification and Performance
Finally the performance of the entire system is computed in
terms of accuracy, error rate, time consumption, and the
memory consumption during training and testing of data.
IV. FACTORS FOR PERFORMANCE ANALYSIS
Following are some of the factors based on which the
performance of the proposed ID3 algorithm and the
traditional algorithm are compared and reviewed.
CONCLUSION
Classification algorithm can be proved to be better in
performance in the process of analysing the text data
according to the sentiments. Therefore the study is
focused on analysing the text sentiments. The
classification model for text data is prepared using
classification algorithms such as Decision Tree
Algorithm and its variants. In this work twitter dataset
can be used for sentiments based text classification.
The raw data can be improved by pre-processing, and
tagging. The model can be trained and tested to
recognize the sentiments of users on microblogging.
REFERENCES
[1]
[2]
a) Accuracy
In a data mining based classification system the amount of
correctly recognized patterns are known as the classification
accuracy. The accuracy of the system in terms of percentage
can be computed using the following formula.
Accuracy = (Accurately Classified Patterns /
Total Input Patterns) x100
b) Error rate
The amount of data misclassified during classification of
algorithms is known as error rate of the system. That can
also be computed using the following formula:
Error rate % = (Total Misclassified Patterns / Total
Input Patterns) x 100
[3]
[4]
[5]
[6]
[7]
[8]
c) Memory Consumption
Memory consumption of the system also termed as the
space complexity in terms of algorithm performance. That
can be calculated using the following formula:
Memory Consumption = Total Memory – Free
Memory
V. MERITS AND DEMERITS OF TEXT
CLASSIFICATION BASED ON SENTIMENT
ANALYSIS
The text classification based on the sentiment analysis is
very useful in understanding the thinking process of the
youth on the basis of microbloging.
However tagging could be done to involve the feature on
data but it can be a time consuming task if it is done
manually token by token not line by line.
ISSN: 2231-5381
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Xia Hu, Lei Tang, Jiliang Tang, Huan Liu, “Exploiting Social
Relations for Sentiment Analysis in Microblogging”, WSDM,
Rome, Italy, Copyright 2013 ACM 978-1-4503-18693/13/ 02,
February 4–8 2013.
Stefan Stieglitz, Linh Dang-Xuan, “Political Communication
and Influence through Microblogging 6 An Empirical Analysis
of Sentiment in Twitter Messages and Retweet Behavior”, 45th
Hawaii International Conference on System Sciences 2012.
Pravin Keshav Patil, K. P. Adhiya, “Automatic Sentiment
Analysis of Twitter Messages using Lexicon based Approach
and Naive Bayes Classifier with Interpretation of Sentiment
Variation”, International Journal of Innovative Research in
Science Engineering and Technology, Vol. 4, Issue 9,
September 2015.
Apoorv Agarwal, Boyi Xie, Ilia Vovsha, Owen Rambow,
Rebecca Passonneau, “Sentiment Analysis of Twitter Data”,
Proceedings of the Workshop on Language in Social Media
(LSM 2011), p.p 30–38, 23 June 2011
Raymond Kosala, Hendrik Blockee, “Web Mining Research: A
Survey”, ACM, Vol.1, pp.1-15, 2000.
Sandra Stendahl, Andreas Andersson, Gustav Strömberg, “Web
Mining”, pp. 1-7.
D. Jayalatchumy, Dr. P. Thambidurai, “Web Mining Research
Issues and Future Directions – A Survey”, IOSR Journal of
Computer Engineering (IOSR-JCE), Vol.14, pp. 20-27, 2013.
Abdelhakim Herrouz, Chabane Khentout, Mahieddine Djoudi,
“Overview of Web Content Mining Tools”, International
Journal of Engineering And Science, Vol 2, pp 1-6, 2013.
R.Malarvizhi, K.Saraswathi, “Web Content Mining Techniques
Tools and Algorithms – A Comprehensive Study”, International
Journal of Computer Trends and Technology, Vol4, pp. 29402945, 2013.
Arvind Arasu, Hector Garcia-Molina, “Extracting Structured
Data from Web Pages”, ACM, pp. 337-348, 2003.
Sameesksha Shrivastava, Dr. Pramod S. Nair, “Mood Prediction
on Tweets Using Classification Algorithm” International
Journal of Science and Research (IJSR), pp. 295-299, 2015.
Chen Jin, Luo De-lin, Mu Fen-xiang, “An Improved ID3
Decision Tree Algorithm”, Proceedings of 2009 4th
International Conference on Computer Science & Education,
IEEE©2009
B. V. Rama Krishna, B. Sushma, “Novel Approach to Museums
Development & Emergence of Text Mining”, International
Journal of Computer Technology and Electronics Engineering
(IJCTEE), Volume 2, No.2.
Andreas Hotho, Andreas Nurnberger, Gerhard Paaß, Fraunhofer
AiS, “A Brief Survey of Text Mining”, Knowledge Discovery
Group Sankt Augustin, May 13, 2005.
Umajancy. S, Dr. Antony Selvadoss Thanamani, “An Analysis
on Text Mining –Text Retrieval and Text Extraction”,
International Journal of Advanced Research in Computer and
Communication Engineering, Volume 2, No.8, August 2013.
http://www.ijettjournal.org
Page 166