Abstract is helpful for purchasers and also producer. While


The main reason of rapid development of social web is to user generated contents like user surveys, reviews, comments, remarks and opinions which is added frequently. These user responses related to any items, peoples, events and so on. This data is exceptionally valuable for organizations, governments and people. While this content to be useful in analyzing the entire content is tedious and too much hectic. So there is a need to build up such framework which consequently mine such huge contents and characterize them into positive, negative and neutral category. Sentiment analysis will be the robotized mining about attitudes, opinions and feelings extracted from text, speech and database resources through Natural Language Processing (NLP). The purpose for this paper is to cover the Sentiment Analysis in the field of Natural Language Processing (NLP), and compare related techniques in this field.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now

Keywords: Sentiment Analysis, Natural Language Processing, Naïve Bayes, Maximum Entropy Model, Support Vector Machines (SVM)



Human basic leadership is dependably impact by others considering, thoughts and suppositions. The development of social web contributes huge measure of client created, for example, remarks, surveys and assessments about items, administrations and occasions. . This is helpful for purchasers and also producer. While making any online purchasing, consumer read out others reviews about the items. Maker can pick up understanding into its items quality and shortcomings in view of the assessment of the clients. In spite of the fact that these feelings are useful for both business associations and people, the immense measure of such obstinate  information winds up plainly overpowering to clients. The most effective method to break down and outline the feelings communicated in this gigantic obstinate information is an extremely fascinating space for specialists. This new research area is generally called sentiment analysis or opinion mining.

Sentiment analysis is the mechanized mining of states of mind, opinions, feelings, and attitudes from speech, text and database sources through Natural Language Processing (NLP). Sentiment analysis includes classifications opinions into grouping like “positive” or “negative” or ” neutral”. It’s frequently called to as subjectivity investigation, opinion mining, and evaluation extraction. Clients need to perceive the idea of others about an item before getting it. Business associations need to recognize what clients are saying in regards to their item or administration that business organization is giving, to settle on future steps It might give capable purpose to voice about client and make trade mark.

Fundamental fields of research in Sentiment examination are Subjectivity Detection, Sentiment Prediction, Aspect Based Sentiment Summarization, Text summarization for Opinions, Contrastive Viewpoint Summarization, Product Feature Extraction, detecting opinion spam. Subjectivity Detection will be an undertaking from claiming deciding if it will be reviewed or not. Sentiment Prediction is tied in with foreseeing the extremity of whether it is certain or negative. Aspect Based Sentiment Summarization gives conclusion synopsis as star ratings or scores of highlights of the item. Text summarization creates a couple of sentences that compress the surveys of an item. Contrastive Viewpoint Summarization puts a stress with respect to contradicting slants. Product Feature Extraction is an undertaking that concentrates the item includes from its survey. Identifying assumption spam will be worry for distinguishing fake alternately fake opinions from reviews.

Sentiment classification drilled down at Document level, Sentence level and Aspect level. Whole document is grouped into positive or negative and named as document level. Same as document level, sentence level grouped into positive or negative. Aspect level related to items features collecting from data.

Review of literature

Sentiment Analysis

Sentiment analysis is a name of process in which sentiments, views, feelings, opinions, attitudes and emotions mined automatically from textures, speech, posts, blogs, tweets and other external sources like database using Natural Language Processing (NLP). With the help of sentiment analysis, opinions and user sentiments in texts can be classified into categories e.g. positive or negative etc. Sentiment analysis can also be referred with other names like subjectivity analysis, appraisal extraction and opinion mining. Basically, it is analysis of user perceptions regarding products, topics and personals.

Supervised Learning Approach

Supervised learning approach totally worked on labeled data set and applied different models to get refined outputs which are beneficial in decision making and forecasting. Specific features is selected and then extracted to detect sentiments.

Machine learning based techniques

The machine learning approach is applied for sentiment analysis which needs to things; one is training set and second is test set. Training set algorithm depends upon characteristics of document to classify it automatically while test set basically works as quality assurance to give surety of results and its percentage. Naive Bayes (NB), maximum entropy (ME), and support vector machines (SVM) are the successful techniques in sentiment analysis. Machine learning totally based on training dataset and gets trained by using training dataset for supervised classification. After applying feature selection, document is able to give meaningful results.

Pang et al. 1 perception, uni-grams have more accuracy than bi-grams in movie review while Dave et al. 6 bigrams and tri-grams more accurate in sense of product-review polarity. POS is used to disambiguate sense which thus will be used to aide characteristic determination 11. For POS tagging every haul in penalties will make allocated. A label which speaks to its position/role in the grammatical setting. To example, POS tags, we might recognize adjectives and adverbs which need normally utilized as sentiment indicators. Pak and Paroubek(2010) 2 gives a model which is based on twitter’s tweets, it classify into three as positive, negative and objective. They are successful fulfill basic need i.e. corpus by gathering twitter’s tweets with the help of twitter’s API and use emoticons for annotating those tweets. With the help of corpus, they are able to develop a sentiment classifier which is based on multinomial Naïve Bayes method, uses N-gram and POS-tags features. As they only used those tweets which contains emoticons so their training set is not too much efficient.Negation is also an important feature to take into account since it has the potential of reversing a sentiment 11. In sentiment classification, these three classifiers named Naïve Bayes, Maximum Entropy and Support Vector Machines (SVM) have been compared in respect to their performance at document level with different features like considered only unigrams, bigrams, or both, combining unigrams and parts of speech, taking only adjectives and combining unigrams and position information by the Pang et al. 1. The outcomes represent that presence of feature is more significance than its frequency while feature set is small. Naïve Bayes gives better results than SVM. If feature space is increased then SVM is better option. Maximum Entropy also have show better performance than Naïve Bayes due to it’s over fitting. Abbasi et al. 12 also worked on sentiment analysis techniques related to classification of hate / extremist web forum posting which is based on different languages like English, Arabic etc by using stylistic and syntactic features. Parikh and Movassate (2009) 2 proposed models that are Naïve Bayes bigram model and Maximum Entropy Model to group twitter’s tweets. They believed that Naïve Bayes gives much better results and performance than Maximum Entropy Model. Go and L.Huang (2009) 4 suggested a solution based on twitter data related to sentiment analysis by using variant supervision, as their training data based on twitter’s tweets which have emoticons and contains noisy labels as well. They use Naïve Bayes, Maximum Entropy Model and Support Vector Machines (SVM). Their feature space contains unigrams, bigrams and POS. Now they believed that SVM performed well as compared to others and unigram is more effective feature. Po-Wei Liang et.al. (2014) 8 have also worked on twitter data and training data based on three different categories i.e. mobile, camera and movie. Positive, negative and non-opinions is labeled using data. Opinion based tweets were filtered. The Naïve Bayes is used to simplifying assumptions independently and eliminated useless features with the help of extraction method like Mutual Information and Chi Square.


Naïve Bayes

Naïve Bayes is an algorithm related to binary and multiclass classification problems to describe binary or categorical input values. This technique is also known as Bayesian theorem because it is based on different theorems.

Maximum Entropy Model

It is widely used in Natural Language Processing and able to classification based on speech and data problems by using different numerical analysis

Support Vector Machines (SVM)

Support Vector Machine is an algorithm usually used to resolve in machine learning problems i.e. classifications and regression issues.

Bayesian Network

A graphical model using probability from data is Bayesian Network. Nodes act as variables and links connect one node to other. Using this connectivity, it is easily for visualization. Statistical distribution like probability is needed for each node in Bayesian Network which is conditional on its parent and advanced algorithm is used to combine these distributions for results as predictions using Bayes Server. Using time series or sequence approach in Bayesian Network makes Dynamic Bayesian Network with the concept of time duration.

Neural Network

Neural network are like neural system which grow and improve their performance with new conditions and tasks. It has capability to solve complex problems by dividing into smaller parts of any big problems. It is mature with millions of different examples and conditions.

Methods and Material

An artificial dataset is developed for modeling different aspects of real data especially focused to evaluate its performance and quality of different supervised learning approach regarding sentiment analysis.

In this, we used different past adapted methods for developing random datasets with covariance matrices. We categorized the classification of data with its features of each nodes and objects including other properties as well. Generally, most of past methods which developed classes show same data variances, relationships and dependencies.

To work on data with parameters, we need covariance data matrices separately for each class. The common practice to draw mentioned matrix elements is probability distribution to get desired matrix. Another practice is to use mapping functions by using algorithm. Suppose x is the input variables or data and y is the output variable or result set, so

Y = F(X)

This is most general form which needs different algorithm to get specific resultant.



A. Abbasi, H. Chen, and A. Salem, “Sentiment analysis in multiple languages: Feature selection for opinion classification in web forums,” In ACM Transactions on Information Systems, vol. 26 Issue 3, pp. 1-34, 2008.

A.Pak and P. Paroubek. „Twitter as a Corpus for Sentiment Analysis and Opinion Mining”. In Proceedings of the Seventh Conference on International Language Resources and Evaluation, 2010, pp.1320-1326

B. Pang and L. Lee, “Opinion mining and sentiment analysis,” Foundations and Trends in Information Retrieval 2(1-2), 2008, pp. 1–135.

B. Pang, L. Lee, and S. Vaithyanathan, “Thumbs up?: sentiment classification using machine learning techniques,” Proceedings of the ACL-02 conference on Empirical methods in natural language processing, vol.10, 2002, pp. 79-86.

Dubey, G., Rana, A., & Ranjan, J. (2016). A research study of sentiment analysis and various techniques of sentiment classification. International Journal of Data Analysis Techniques and Strategies, 8(2), 122. doi:10.1504/ijdats.2016.077485

Go, R. Bhayani, L.Huang. “Twitter Sentiment ClassificationUsing Distant Supervision”. Stanford University, Technical Paper,2009

Hirschberger M, Qi Y, Steuer RE (2007) Randomly generating portfolio-selection covariance matrices with specified distributional characteristics. European Journal of Operational Research 177: 1610–1625.

K. Dave, S. Lawrence, and D. M. Pennock, “Mining the peanut gallery: Opinion extraction and semantic classification of product reviews,” Proceedings of WWW, 2003, pp. 519–528.

Liu, B. (n.d.). Document Sentiment Classification. Sentiment Analysis, 47-69. doi:10.1017/cbo9781139084789.004

Madhoushi, Z., Hamdan, A. R., & Zainudin, S. (2015). Sentiment analysis techniques in recent works. 2015 Science and Information Conference (SAI). doi:10.1109/sai.2015.7237157

Po-Wei Liang, Bi-Ru Dai, “Opinion Mining on Social MediaData”, IEEE 14th International Conference on Mobile Data Management,Milan, Italy, June 3 – 6, 2013, pp 91-96, ISBN: 978-1-494673-6068-5, http://doi.ieeecomputersociety.org/10.1109/MDM.2013.

R. Parikh and M. Movassate, “Sentiment Analysis of User- GeneratedTwitter Updates using Various Classi_cation Techniques”,CS224N Final Report, 2009

Read, J., & Carroll, J. (2009). Weakly supervised techniques for domain-independent sentiment classification. Proceeding of the 1st international CIKM workshop on Topic-sentiment analysis for mass opinion – TSA ’09. doi:10.1145/1651461.1651470