helpfn

bcs

Support_Vector_Machine_based_Word_Embedding_and_Feature_Reduction_for_Sentiment_Analysis-A_Study.pdf

Home >Computer Science homework help >helpfn

Support Vector Machine based Word Embedding and Feature Reduction for

Sentiment Analysis-A Study

Prof. Prajakta P. Shelke Department of Computer Science and Engineering,

Government College of Engineering, Amravati, India

[email protected]

Ankita N. Korde Department of Computer Science and Engineering,

Government College of Engineering, Amravati, India

[email protected] m

ABS TRACT: S entiment analysis (S A), also called as opinion mining is the technique used to bring together the opinions of a specific entity or feature from reviews dataset. The opinions of other users help in performing the decision making process. This paper studies different methods that are aimed at performing sentiment analysis. These approaches vary from semantic based methods, machine learning, neural networks, and syntactical methods with each having its own strength. Although hybrid approach also exists, the main idea is to combine the strengths of two or more methods to increase the

accuracy. A framework in which sentiment analysis is done by using the proposed word embedding and feature reduction techniques. Word embedding is a technique in which low-dimensional vector representation of words i s provided. Feature reduction method employs a support vector machine (S VM) classifier. The framework will perform sentiment analysis of user opinions by using a machine learning approach and provides a recommendation system for the ease of decision making to users.

INDEX TERMS : S entiment analysis, opinion mining, genetic algorithm, word embedding, feature reduction.

I. INTRODUCTION Sentiment analysis is an important research area as people are being more express ive in social networks like Facebook as well as Twitter. Several approaches aimed at sentiment analysis has been used among which the most common approach is machine learning (ML) which learns the different sentiments by using a significant dataset. ML methods are used to train the classifiers and determine the sentiments by using different learning algorithms and datasets. Sentiments are generally found in comments, reviews or feedbacks. These sentiments are either positive or negative. In respect to this, sentiment analysis works as a task of classification in which every classified set signifies the sentiment. SA shows the customer satisfaction for a product or an entity.

This paper proposes a sentiment analysis which employs SVM algorithm and UCI ML Repository dataset to analyse and classify the reviews. Feature reduction is also done by using by using genetic algorithm (GA) technique which works by developing a fitness function.

Word embedding is used in sentiment analysis task which provides low dimensional vector representation of words. Sentiment embedding is used for acquiring both semantic and syntactic similarity between the words which avoids generating similar vector representation for semantically similar words.

There are various opinions of users on different products. For example, ‘The phone is good but the voice quality is poor’. In such reviews where both negative and positive sentiments are included in a single sentence, the word embedding technique makes it easy to analyse the sentiments. Feature reduction is an important technique in sentiment analysis which makes the classifiers more accurate and efficient. A there are plenty of features in online reviews, it makes classifier infeasible. So to eliminate the unnecessary features, feature reduction technique is used.

II. LITERATURE SURVEY This section describes the research work related to the field of sentiment analysis. The researchers have developed and tested various approaches to sentiment analysis. Studying these approaches helps us to see how user’s opinions are helpful in analysing effectiveness of reviews.

In [5] the author describes the different ways to enable sentiment analysis systems. They focus on methods to fulfil challenges produced by sentiment analysis applications in comparison with the available applications in more traditional systems. They included the problems related to confidentiality, management, and financial influence of opinion leaning facilities .

Proceedings of the Fourth International Conference on Computing Methodologies and Communication (ICCMC 2020) IEEE Xplore Part Number:CFP20K25-ART; ISBN:978-1-7281-4889-2

20 20

F ou

rt h

In te

rn at

io na

l C

on fe

re nc

e on

C om

pu ti

ng M

et ho

do lo

gi es

a nd

C om

m un

ic at

io n

(I C

C M

C )

97 8-

1- 72

81 -4

88 9-

2/ 20

/$ 31

.0 0

20 I

E E

E 1

0. 11

09 /I

C C

M C

48 09

2. 20

20 .I

C C

M C

-0 00

Authorized licensed use limited to: University of the Cumberlands. Downloaded on July 24,2021 at 03:45:57 UTC from IEEE Xplore. Restrictions apply.

In [6], they provided a system used for SA of small, social network grades. The system is demonstrated with twitter reviews to show glad and down sentiments as well as demonstrate that the system gives better performance compared to Naïve Bayes . They analysed the twitter reviews and provided a scheme containing data acquisition as well as calculation to analyse the sentiments, which is a scalable system.

In [7], author has attempted a complete summary of last update in field of SA. Many projected algorithms’ improvements and several SA applications have been examined and presented concisely. The correlated fields to SA that are concerned by researchers recently are deliberated. They offered almost full image of SA techniques and the correlated fields with brief details and included the refined categorizations of an outsized number of recent articles and the design of the recent development of research in the sentiment analysis and its related zones . Sentiment analysis of data on Twitter is examined in [8]. They presented POS-specific preceding divergence features. The practice of a tree kernel on the way to prevent requirement of deadly feature manufacturing is discovered. The novel features and tree kernel accomplished nearly with a similar way, together outstripping an advanced reference point. In [9], usefulness of etymological structures for identifying sentiment in Twitter communications is studied. It estimated worth of current lexical assets and also features which seizure data around creative as well as casual language applied with microblogging. They acquired the supervised method for problem, nonetheless for construction of training data which influenced current hashtags in the Twitter data. In [10], word embedding aims to the vector illustration of arguments through leveraging the appropriate data with huge reviews datasets is stated. A pioneering work is proposed and castoff the neural network language system towards study of word embedding built with a prior context of every word [11]. For eluding the production of analogous vector representations of sentimentally differing words, current research ought to propose embedding [12] to capture both semantic and syntactic data so that sentimentally related words possess comparable symbols [13].

III. PROPOSED METHODOLOGY In this system, we present a sentiment analysis framework which includes data cleansing, data pre- processing and sentiment analysis. For better recommendation system, a summarized review is necessary to the customer for decision making ability. The proposed framework illustrates the review summarization and allows users to make a quick decision.

Fig.: Proposed System

The proposed system works in different stages such as data cleansing, data preprocessing, sentiment analysis as explained below. A. DATA CLEANSING The first module of this methodology is data cleansing. In this process, unwanted data is removed. The slang and abbreviation is translated in their unique and reduced arrangement. Stopword removal is also done. B. PREPROCESSING This module includes tokenization, word stemming and POS tagging. Tokenization is breaking text into meaningful elements. The inflected word is reduced to its root word is done in stemming. POS tagging includes tagging a word as corresponding part of speech. C. SENTIMENT ANALYSIS USING SVM SVM is supervised ML algorithm offered in classification. After the pre-processing of data, the extracted features are obtained called as bag -of-words. By using SVM the sentiment values of given data is classified. Here word embedding is used and the feature vector is generated for each feature. This makes the data ready for classification. SVM performs classification by finding the hyperplane that differentiate the classes plotted in n-dimensional space. Thus, SVM gives the classified data and we can predict whether the analysis is positive or negative. D. RECOMMENDATION SYSTEM Once the sentiment analysis is done over the reviews, we provide a better recommendation system according to the analysis of reviews for specific product. This provides user quick decision making ability.

Proceedings of the Fourth International Conference on Computing Methodologies and Communication (ICCMC 2020) IEEE Xplore Part Number:CFP20K25-ART; ISBN:978-1-7281-4889-2

Authorized licensed use limited to: University of the Cumberlands. Downloaded on July 24,2021 at 03:45:57 UTC from IEEE Xplore. Restrictions apply.

IV. COMPARISON WITH EXISTING SENTIMENT ANALYSIS TECHNIQUES

Paper Name Authors Technique used Implementation SA- A Combined Approach RUDIY PRAABOWO Combined approach to rule

based ordering, supervised as well as ML

The system has been applied for generating movie review, also showed effectiveness of hybrid classification.

ML Approach to SA in Multi lingual Web Texts

ERIK BOEY Machine learning approach Implemented on blog, appraisal as well as medium copies originate in WWW, inscribed with English, Dutch or French for recognizing positive, negative as well as neutral emotional state for unit providing better accurateness

Lexicon-Based Methods for SA

MAEITE TABODA Lexicon based approach A (SO-CAL) applies lexicons of features marked by its semantic coordination, also includes escalation, cancellation.

Holistic Lexicon Bas ed Method for Opinion Mining

XIAO WENDDING Holistic lexicon-based approach

The method lets a classification to grip estimation features those are reliant on situation, grounding main complications aimed at current algorithms.

V. CONCLUS ION

In this paper, we have presented the sentiment analysis framework by using word embedding technique. The feature reduction technique makes the process of sentiment analysis easier. The proposed system provides more accuracy and scalability. This system is designed by considering the e-commerce websites. In future, this framework can be extended as a generalized sentiment analysis system to analyse the travel and hotel reviews. REFERENCES [1] Farkhund Iqbal,Jahanzeb Maqbool Hashmi and Benjamin C.

Fung,”A Hybrid Framework for Sentim ent Analysis Using Genetic Algorithm based Feature Reduction”, 2017.

[2] M. P ont iki et al., ``Sem Eval-2016 task 5: Aspect based

sentim ent analysis,'' in P roc. 8t h Int . Workshop Semant ic Eval. (SemEval), 2014.

[3] P . C. S. Njølst ad, L. S. Høysæter, W. Wei, and J. A. Gulla,

``Evaluating feature sets and classifiers for sentim ent analysis of financial news,'' in P roc. IEEE/WIC/ACM Int . Joint Conf. Web Intell. (WI) Intell. Agent T echnol. (IAT ), vol. 2, Aug. 2014.

[4] M. Govindarajan, ``Sentim ent analysis of m ovie reviews

using hybrid method of naive Bayes and genetic algorithm,'' Int . J. Adv. Comput . Res., vol. 3, no. 4, 2013.

[5] B. P ang and L. Lee,“Opinion m ining and sentim ent analysis',' Found. T rends Inf. Ret r., vol. 2, 2008.

[6] A. Davies and Z. Ghahramani, ``Language-independent

Bayesian sentiment mining of Twitter,'' in P roc. Workshop Social Net w. Mining Anal., 2011.

[7] W. Medhat , A. Hassan, and H. Korashy, “ Sentim ent

analysis algorithms and applications: A survey,'' Ain Shams Eng. J., vol. 5, no. 4, 2014.

[8] A. Agarwal, B. Xie, I.Vovsha, O. Rambow, and R.

P assonneau, “Sentiment analysis of Twitter data,'' in P roc. Workshop Lang. Social Media, 2011.

[9] Duraipandian, M. "P erformance Evaluat ion of Rout ing

Algorit hm for MANET based on t he Machine Learning T echniques." Journal of trends in Com puter Science and Sm art technology (TCSST) 1, no. 01 (2019): 25 -38.

[10] Ling-Chih Yu, J. Wang, K. Robert and X. Zhang,”Refining

Word Em beddings using intensity scores for Sentim ent Analysis”, Vol. 26, March 2018

[11] Y. Beingo, R. Ducharme,”A Neural Probabilistic Language

Model”, J. Mach Learn Res., Vol. 3. [12] A. L. Maas, R.E.Daly ,”Learning Word Vectors for

Sentim ent Analysis”. In P roc. ACL, 2011. [13] Y. Ren, Y. Zhang and D. Ji,”Im proving Twitter Sentim ent

Classification using Topic-enriched Multi prototype Word Em bedding”, in P roc. AAAI, 2016.

Proceedings of the Fourth International Conference on Computing Methodologies and Communication (ICCMC 2020) IEEE Xplore Part Number:CFP20K25-ART; ISBN:978-1-7281-4889-2

Authorized licensed use limited to: University of the Cumberlands. Downloaded on July 24,2021 at 03:45:57 UTC from IEEE Xplore. Restrictions apply.

[14] A. Collomb, C. Cost ea, D. Joyeux, O. Hasan, and L. Brunie, ``A study and comparison of sentiment analysis m ethods for reputation evaluation,'' T ech. Rep. RR-LIRIS-2014-002, 2014.

[15] L. M. Schmit t , ``Theory of genetic algorithm s,'' T heor.

Comput . Sci., vol. 259, May 2001. [16] A. P ak and P . P aroubek, ̀ `Twitter as a corpus for sentim ent

analysis and opinion m ining,'' in P roc. LREC, 2010.

Proceedings of the Fourth International Conference on Computing Methodologies and Communication (ICCMC 2020) IEEE Xplore Part Number:CFP20K25-ART; ISBN:978-1-7281-4889-2

Authorized licensed use limited to: University of the Cumberlands. Downloaded on July 24,2021 at 03:45:57 UTC from IEEE Xplore. Restrictions apply.