International Journal of Science and Research (IJSR)

ISSN (Online): 2319-7064

Index Copernicus Value (2013): 6.14 | Impact Factor (2014): 5.611

A Study on Sentiment Analysis: Methods and Tools

Abhishek Kaushik1, Anchal Kaushik2, Sudhanshu Naithani3
1
Kiel University of Applied Sciences, Computer and Electrical Department, Sokratesplatz 1, 24149 Kiel, Germany
2
Amity University, Department of Management Studies, Sector 125, Noida, India
3
Kurukshetra University, Department of Computer Science, Thanesar Taluk, Kurukshetra, India

Abstract: The purpose of social media has created many chances for people to publicly voice their beliefs, simply when they are
employed to deliver an opinion hit a vital problem. Sentiment Analysis is a case of natural language processing which could mark the
mood of the people about any specific product by analysis. Sentiment Analysis is a process of automatic extraction of features by mode
of notions of others about specific product, services or experience. The Sentiment Analysis tool is to function on a series of expressions
for a given item based on the quality and features. Sentiment analysis is also called Opinion mining due to the significant volume of
opinion. Analyzing customer opinion is very important to rate the product. To automate rate the opinions in the form of unstructured
data is been a challenging problem today. Thus, this paper discusses about Sentiment analysis methods and tools used.

Keywords: Data Mining, Opinion Mining, Opinion Summarization, Sentiment Analysis, Text Mining, Web Mining.

Figure 1: Hierarchy of Data Mining

1. Introduction as an alternative technique capable of triangulating

qualitative and quantitative methods through innovative real
The era of electronic information in every phase of life is time data collection and analysis. G.Vinodhini et al [10]
evolving rapidly, which tends to produce a large number of proposed an Overview of different opinion mining
data. As an outcome huge volumes of data are generated in techniques. Blessy Selvam et al [3] proposed different
field of technology, business, healthcare, tourism, e- approaches of sentiment classification and the existing
marketing, etc. Automated analysis systems are meant for methods with the framework. Rudy Prabowo [16] formed a
analysis, summarization and classification of data and new approach by combining rule-based classification,
number of efficient methods to store huge amount of data. supervised learning and machine learning and tested it on
Text mining is an approach used different fields like machine movie reviews, product reviews and MySpace comments.
learning, information retrieval, statistics, and computational And also proposed a semi automatic approach to get better
linguistics for opinion mining. Web mining is a subset of text effectiveness. Archana Shukla [19] introduced a tool to tell
mining used to mine the unstructured web data in the form the quality of the document or its usefulness based on the
Content mining, Web Structure mining and Web Usage annotations. Ayesha Rashid et al [1] presented the limitations
mining. The aim of sentiment analysis is to make an on different sentiment level and the methods used in
automated machine able to recognize and categorize sentiment analysis. Dongjoo Lee et al [4] proposed to use the
emotions [2]. A thought, view, or attitude based on emotion PMI method to use for large corpus to achieve higher
instead of reason is called sentiment. Figure 1 shows the accuracy. Dr.Ritu Sindhu et al [14] presented different levels
different sub level of Data Mining and the branches of of analysis and issues in sentiment analysis. S.Chandrakala et
sentiment analysis. al [5] proposed a work on recent papers on sentiment
analysis and its related tasks with future challenges. Bo Pang
2. Literature Overview [17] gave a new machine learning method that determines
sentiment polarity. Arti Buche et al [11] proposed the Naive
Bakhtawar Seerat et al [15] proposed the method of opinions Bayes algorithm and also Hidden Markov Model to calculate
extraction from an online web page and the limitation of the Entropy and Purity measure in string mining. S.Padmaja
Sentiment analysis. Meena Rambocas [20] concluded all the et al [6] proposed a work on Machine Learning Models for
challenges marketers can face when using sentiment analysis text classification. . Nile M. Shrike et al [13] compared the
Volume 4 Issue 12, December 2015
www.ijsr.net
Paper ID: NOV151832 287
Licensed Under Creative Commons Attribution CC BY
 International Journal of Science and Research (IJSR)
ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2014): 5.611
accuracy using Bayes, Maximum Entropy and Support The views are being extracted from writers review over their
Vector Machine. Raisa Varghese et al [7] proposed the comment. Opinion feature extraction is a sub-process of
structure of sentiment analysis. Vijay B . Roth et al [9] have opinion mining [15]. Pre-processing In this process, raw data
compared the synopsis of different approaches used for taken and is pre-processed for feature extraction.
sentiment analysis. Nidhi Mishra et al [12] proposed the
inner view of sentiment analysis at different levels David
Osimo et al [2] proposed an outline for a new Research
Challenge on Sentiment Analysis. Sindhu, Chandrakala et al
[8] proposed a systematic flow and Machine learning
approaches to optimize the performance. Alec Go [18]
proposed a novel approach to classify sentiment of the twitter
message automatically and showed that machine learning
algorithms (Naive Bayes, Maximum Entropy, and SVM)
have accuracies above 80% when trained with emoticon data.

3. Information Source
User view is an important factor for the improvement of the
quality of services. Blogs, review sites, data and micro blogs
provide a good information of the products and services
provided to clients.

Blogs: The name relates all the blog sites is called

blogosphere [1]. People express about their thoughts they
want to share with others on a blog. Blog pages [1] have
become the popular platform to share ones personal views
about specific products .

Review sites: The opinions of others is being an important

factor while purchasing anything. A large number of users
express their views on a particular product. These reviews
are easily available on the Internet. The re-viewer’s data used Figure 2: Work flow of Opinion Mining
in most of the opinion classification gather from the e-
commerce websites [10] like www.flipkart.com . The preprocessing phase [1] has been further divided into a
number of sub phases as follows:
Data Set: The dataset contains different types of product
reviews (including Books, DVDs, Electronics and Kitchen Tokenization is the process to split up into tokens by
appliances) and movie reviews extracted from Flip-kart and removing white spaces, commas and other symbols, etc. Stop
IMDB webpage. word Removal removes words (like „a, an, the, of, for,).
Stemming reduce the relevant tokens into a single type.
4. Sentiment Analysis Normalization is a process that has English texts to be
published in both higher and lower case characters and turns
Sentiment analysis is a technique which is used to extract the the entire document or sentences into lowercase/uppercase.
meaningful information in the documents [6]. In general,
opinion mining tries to figure out the sentiment of a writer Feature extraction phase deals with feature types [3] (which
about some specific aspect and also the overall contextual identifies the type of features used for opinion mining),
polarity of a document. The sentiment may be a judgment, feature selection (used to select good features for opinion
mood or evaluation of the writer [2]. A core issue in this field classification), feature weighting mechanism (weights each
is an opinion classification, where a review is classified as a feature for good recommendation) reduction mechanisms
positive or negative evaluation of a subjected object (film, (features for optimizing the classification process).
book, etc.). The assessment of sentiment can be done in two
ways: Types of features used for opinion mining could be:
1) Term frequency (The presence of the term in a document
4.1 Direct opinions: It gives positive or negative sentiment carries a weight age).
about the product directly [12]. For example, “The food 2) Term co-occurrence (features which occurs together like
quality of this hotel is poor” expresses a direct opinion. uni-gram, bi-gram or n-gram),
3) Part of speech information (POS tagger is used to separate
4.2 Comparison: It means to compare the subject with any POS tokens).
other similar objects [12]. For example, “The food quality of 4) Opinion words (Opinion words are words which express
the hotel-a is better than that of hotel-b.” expresses a positive (good) or negative (bad) emotions) [3].
comparison. Figure 2 had a workflow of Opinion Mining. 5) Negations (Negation words (not, not only) shift sentiment

Volume 4 Issue 12, December 2015

www.ijsr.net
Paper ID: NOV151832 288
Licensed Under Creative Commons Attribution CC BY
 International Journal of Science and Research (IJSR)
ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2014): 5.611
orientation in a sentence) Summarization. Review Text is retrieved from review
6) Syntactic dependency (It is represented as a parse tree and websites. Opinion text in blog, reviews, comments, etc.
it contains word dependency based features) contains subjective information about the topic.

Feature Selection Reviews classified as positive or negative review. Opinion

1) Information gain (based on the presence and absence of a summary is generated based on features opinion sentences by
term in a document a threshold is set and the terms with considering frequent features about a matter.
less information gain is removed).
2) Odd Ratio (It is suitable for binary class domain where it 5.1 Opinion Retrieval
has one positive and one negative class for classification.
3) Document Frequency measures the number of appearances It is the procedure of collecting review text from review sites.
of a term in the available number of documents in the Different review websites contain reviews for products,
corpus and based on the threshold computed the terms are movies, hotels and news.
removed. Features weighting mechanism The mechanisms
are of two types. They are 1: Term Presence and Term 5.2 Information retrieval
Frequency- word which occurs occasionally contains more
information than frequently occurring words. 2: Term Techniques such as web crawler can be employed to collect
frequency and inverse document frequency (TFIDF) - the review text data from many sources and store them in a
Documents are rated where highest rating is given to words database. This step involves retrieval of reviews, micro-blogs
that appear regularly in a few documents and lowest rating and comments by user.
for words that appear regularly in every document. Feature
Reduction Feature reduction reduces the feature vector size 5.3 Opinion Classification
to optimize the performance of a classifier.
Primary steps in sentiment analysis are a classification of
Reduction of the number of features in the feature vector can review text. Given a review document M = {M1….. M1} and
be done in two different ways in which top n-features can be a predefined category set K = {positive, negative}, sentiment
left in the vector and either low level or unwanted linguistic classification is to classify each day in M, with a label
features could be removed. Adjectives only Adjectives have expressed in K. The approach involves classifying review
been used most frequently as features amongst all parts of text into two forms namely positive and negative [9].
speech. A strong correlation between adjectives and Machine learning and dictionary based approach is more
subjectivity has been found. Although all the parts of speech popular [3].
are important people most commonly used adjectives to
depict most of the sentiments and a high accuracy have been 5.4 Opinion Summarization
reported by all the works concentrating on only adjectives for
features generation. Adjective-Adverb Combination Most of Summarization of opinion is a major character in the opinion
the adverbs have no prior polarity. mining process. Summary of reviews provided should be
based on features or subtopics that are mentioned in the
But when they occur with sentiment bearing adjectives, they reviews. Many works have been done on summarization of
can play a major role in determining the sentiment of a product reviews [9].
sentence. Adverbs alter the sentimental value of the adjective
that they are used with. Adverbs of degree, on the basis of the The opinion summarization process mainly involves the
extent to which they modify this sentimental value, are following two approaches. Feature based summarization a
classified as: type summarization involves the finding of frequent terms
 Adverbs of affirmation: certainly, totally (features) that are appearing in many reviews. The summary
 Adverbs of doubt: maybe, probably is submitted by selecting sentences that contain particular
 Strongly intensifying adverbs: exceedingly, immensely feature information. Characteristics present in review text can
 Weakly intensifying adverbs: barely, slightly be identified using Latent Semantic Analysis (LSA) method.
 Negation and minimizers: never Some of the positive
Term frequency is a count of term occurrences in a
Adjectives are as follows dazzling, brilliant, phenomenal, document. If a term has higher frequency it means that the
excellent and fantastic. Negative Adjectives: suck, terrible, condition is more import for summary presentation. In many
awful, unwatchable, hideous. product reviews certain product features come out frequently
and associated with user opinions about it. Fig. 3 has the
architecture of Opinion Mining which says how the input is
5. Standard Structure of Sentimental Analysis being classified on the various steps to summarize the
reviews.
Opinion Mining also called sentiment analysis is a process of
finding user’s opinion towards a topic or a product. Opinion
mining concludes whether the user’s view is positive, minus,
or neutral about a product, issue, event, etc. Opinion mining
and summarization process involve three primary steps, first
is Opinion Retrieval, Opinion Classification and Opinion

Volume 4 Issue 12, December 2015

www.ijsr.net
Paper ID: NOV151832 289
Licensed Under Creative Commons Attribution CC BY
 International Journal of Science and Research (IJSR)
ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2014): 5.611
7. Techniques
Major data mining techniques used to dig the knowledge and
information are: generalization, classification, clustering,
genetic algorithm, association rule mining, data visualization,
neural networks, fuzzy logic, Bayesian networks, and,
decision tree. Number 5 has the techniques of Opinion
Mining. Figure 5. Techniques of Opinion Mining

 Supervised Machine Learning: Classification is most

oftenly used and very popular data mining technique [11].
Classification used to divide the possible results from a
given data set is based on the basis of a defined set of
attributes and a given predictive attributes. The given
dataset is used as the training dataset consist of
independent variables (properties of the dataset) and a
dependent attributes (predicted attribute). A training
dataset created model test on text corpus holds the same
attributes but no predicted attribute. Accuracy of model
checks on how faultless it is making a prediction. Double
Propagation Algorithm is used to extract Product features
and sentenced words.

Figure 3: Architecture of opinion mining

6. Hierarchy of Opinion Mining

 Document level Opinion Mining- A single document of Figure 4: Techniques of Opinion

opinionated text works as a basic data unit in this level [7].
Here the document level classification is a single review  Unsupervised Learning- It differs to supervised learning,
about a topic is viewed. Merely in the forums or blog because unsupervised learning does not have definite
scenario, there are possibilities for comparative sentences targeted output connected with the input. Class label for
to appear and for clients to compare one product with any instance is not known so this technique of learning is
another that has alike characteristics and that’s how about to learn by observation. Clustering is a technique
document level analysis is not suitable for forums and web which is also used in unsupervised learning. Clustering is
logs. Therefore the subjectivity/objectivity arrangement is an approach of bunching objects with similar properties
very vital in this type of classification. into a group. Objects in a cluster are always dissimilar to
 Sentence level Opinion Mining- The calculated polarity of the objects in other clusters.
each sentence is considered in the case of sentence level  Case Based Reasoning- Case based reasoning is one of the
Opinion Mining. The same classification approach as emerging Artificial Intelligence supervised techniques.
applied in document level, can be reactive to the sentence CBR is a fierce tool of computer reasoning and crack the
level classification problem also, but Objective and problems (cases) in the closest way to real time scenario.
subjective sentences [12] necessarily be localized. Opinion This is a problem solving technique in which knowledge is
words are carried by subjective sentences. These sentiment personified as past cases in the library and it is not
words aid to determine the sentiments related to that entity. dependent on classical rules. The solutions of all the cases
After which the polarity classification takes place into are stored in CBR warehouse known as Knowledge base or
positive and negative classes. Case base.
 Phrase level Opinion Mining- This level of classification is
much more pinpointed approach to opinion mining. Here 8. Semantic Orientation
phrases containing opinion words are observed and the
phrase level class is completed. Only in some special Problem of Opinion mining can be divided into two parts
cases, where contextual polarity also matters, the effect which are sentiment classification [13] and feature based
may not be fully precise. opinion mining. The trouble of taking out the semantic
orientation (SO) of a text (i.e., whether the text is positive or
negative towards a peculiar subject matter) often takes as a
Volume 4 Issue 12, December 2015
www.ijsr.net
Paper ID: NOV151832 290
Licensed Under Creative Commons Attribution CC BY
 International Journal of Science and Research (IJSR)
ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2014): 5.611
starting point the problem of determining semantic extracted from the customer reviews. The results are
orientation for individual speech. The hypothesis is that, if displayed by a web based interface [1].
the SO of relevant words in a text is given, SO for the entire  Opinion observer-This is an opinion mining system which
text can be determined. The SO approach to Sentiment is used to analyze and compare different opinions [5] on
analysis is an unsupervised learning because it does not need the cyber space by using user generated contents. This
advance training in order to mine the data. Figure 6 shows system illustrates the results in a graph format clearly
the details of the classification of approaches of semantic showing opinion of the product feature by feature. It uses a
orientation. WordNet Exploring method to assign prior polarity.

10. Conclusion
Opinion mining is an emerging sphere of data mining used to
receive the knowledge of the huge mass of data (data may be
customer comments, feedback and reviews on whatever
product or topic etc). Much research has been carried on to
mine the opinions in the contour of a document, sentence and
feature level sentiment analysis. It has been examined that
now the opinion mining trend is proceeding to the
sentimental reviews of twitter data, comments used in
Figure5: Classification of Approaches of Semantic Facebook on pictures, videos or Facebook status. Therefore,
Orientation this paper discusses about an overview of the sentimental
analysis approach of Opinion Mining in detail with the
 Corpus Based Approach- Emotional affinity of words is techniques and tools.
determined by Popular corpus-driven method. Emotional
affinity is meant to learn their probabilistic affective scores References
from large corpora. The method to assign a happiness
factor to words depending on the frequency of their [1] Ayesha Rashid et al, “A Survey Paper: Areas,
occurrences in happy-labelled blog posts compared to their Techniques and Challenges of Opinion Mining”,
total frequency in a corpus containing blog posts labelled International Journal of Computer Science (IJCSI), Vol
with “happy” and “sad” mood annotations. They also 10 Issue 6 No 2, Nov 2013.
compare the happiness factor scores of words with the [2] David Osimo and Francesco Mureddu, “Research
scores in the list. Challenge on Opinion Mining and Sentiment Analysis”.
 Dictionary Based Approach- Dictionary based approach [3] Bluesy Selvam, A. Abirami, “A Survey on Opinion
contains used lexical resources (e.g-Word Net) which work Mining Framework”, International Journal of Advanced
as an asset to automatically acquire emotion-related words Research in Computer and Communication
for emotion classification experiments. They start from a Engineering,Vol 2, Issue 9, Sep 2013Pg No 3544-3549.
set of primary emotion adjectives, and then retrieve alike [4] Dongjoo Lee et al, “Opinion Mining of Customer
words from Word Net by utilizing all senses of all words in Feedback Data on the Web”. Seoul National University.
the synsets that contain the emotion adjectives. The [5] S. Chandrakala, C. Sindhu, “Opinion Mining and
process takes advantage of the synonym and hyponym Sentiment Classification: A Survey”,ICTACT Journal on
relations in Word Net to manually find alike words to Soft Computing, Oct 2012 Vol 3 Issue 1,Pg No 420-425.
nominal emotion words. The affective weights are [6] S.Padmaja et al, “Opinion Mining and Sentiment
automatically acquired from a very large text corpus in an Analysis – An Assessment of Peoples’ Belief: A
unsupervised fashion. Survey”, International Journal of Ad hoc, Sensor &
Ubiquitous Computing IJASUC, Vol 4 No 1, Feb 2013.
9. Tools Used In Opinion Mining [7] Raisa Varghese, Jayasree, “A Survey on Sentiment
Analysis and Opinion Mining”, International Journal of
The tools used in the process of tracking the opinion or Research in Engineering and Technology (IJRET), Vol 2
polarity from the user’s generated contents are: Issue 11 Nov 2013.
 Review Seer tool – Work done by aggregation sites is [8] Sindhu, Chandrakala, “A Survey on Opinion Mining and
automated by this tool. To collect positive and negative Sentiment Polarity Classification”, International Journal
opinions for assigning a score to the extracted feature of Emerging Technology and Advanced Engineering.Vol
terms, the Naive Bayes classifier approach is used. The 3 Issue 1, Jan 2013.
results are displayed as a simple opinion sentence [10]. [9] Vijay. B. Roth et al, “Survey on Opinion Mining and
 Web Fountain - Beginning definite Base Noun Phrase Summarization of User Reviews on Web”, International
(BNP) heuristic approach is used here for extracting the Journal of Computer Science and Information
product features. Development of a simple web interface is Technologies (IJCSIT),Vol 5(2), 2014. 1026-1030.
also possible. [10] G. Vinodhini et al, “Sentiment Analysis and Opinion
 Red Opal –This tool allows the users to determine the Mining: A Survey”, International Journal of Advanced
features based opinion orientations of products. It assigns Research in Computer Science and Software
the scores to each and every product based on features Engineering (IJARCSSE), Vol 2, Issue 6, June 2012.

Volume 4 Issue 12, December 2015

www.ijsr.net
Paper ID: NOV151832 291
Licensed Under Creative Commons Attribution CC BY
 International Journal of Science and Research (IJSR)
ISSN (Online): 2319-7064
Index Copernicus Value (2013): 6.14 | Impact Factor (2014): 5.611
[11] Arti Buche, Dr.M.B.Chandak, Akshay Zadgoanakar
“Opinion Mining and Analysis: A Survey”, International
Journal on Natural Language Computing (IJNLC) Vol 2
No 3 June 2013Pg No 39-48.
[12] Nidhi Mishra et al, “Classification of Opinion Mining
Techniques”, International Journal of Computer
Applications, Vol 56, No 13, Oct 2012Pg No 1-6.
[13] Nilesh M. Shrike et al, “Survey of Techniques for
Opinion Mining”, International Journal of Computer
Applications, Vol 57 No 13. Nov 2012Pg No 30-35.
[14] Dr. Ritu Sindhu, Ravendra Ratan Singh Jandail, Rakesh
Ranjan Kumar, “A Novel Approach for Sentiment
Analysis and Opinion Mining”, International Journal of
Emerging Technology and Advanced Engineering
(IJETAE), Vol 4, Issue 4, April 2014.
[15] Bakhtawar Seerat, Farouque Azam, “Opinion Mining:
Issues and Challenges (A Survey)”, International Journal
of Computer Applications, Vol49 No 9 July 2012Pg No
42-51.
[16] Rudy Prabowo, Mike Thelwell, "Sentiment Analysis: A
Combined Approach".
[17] Bo Pang, Lillian Lee, "A Sentimental Education:
Sentiment Analysis Using Subjectivity Summarization
Based on Minimum Cuts".
[18] Alec Go, Richa Bhayani, Lei Huang, "Twitter Sentiment
Classification Using Distant Supervision".
[19] Archana Shukla, "Sentiment Analtsis of Document
Based on Annotation".
[20] Meena Rambocas, Joao Gama, "Marketing Research:
The Role of Sentiment Analysis".

Author Profiles
Abhishek Kaushik is currently working in Siemens,
Germany as a Master thesis student. He is in the final
phase of completing his Masters degree in Information
Technology from Kiel University of Applied Sciences.
Before starting his Masters he received his Bachelor’s of
Technology in Computer Science Engineering from Kurukshetra
University in 2012.

Anchal Kaushik is pursuing her MBA in Competitive

Intelligence and Strategy Management from Amity
University Noida. Before this she received her
Bachelor's of Commerce Degree in 2014 from CCS
Meerut

Sudhanshu Naithani has received his Bachelor's of

Technology in Computer Science Engineering from
Kurukshetra University in 2015. He is currently
working as a research assistant under Assistant
Professor Ravinder Madan at Manav Bharti University,
Solan.

Volume 4 Issue 12, December 2015

www.ijsr.net
Paper ID: NOV151832 292
Licensed Under Creative Commons Attribution CC BY