Optimization driven actor-critic neural network for sentiment analysis in social media
Pages | 457-476 |
Published date | 11 November 2019 |
Date | 11 November 2019 |
DOI | https://doi.org/10.1108/VJIKMS-12-2018-0116 |
Author | Jayashree Jagdale,Emmanuel M. |
Subject Matter | Information & knowledge management,Knowledge management,Knowledge management systems |
Optimization driven actor-critic
neural network for sentiment
analysis in social media
Jayashree Jagdale
Pacific University, (PAHER), Udaipur, India, and
Emmanuel M.
Department of IT, Pune Institute of Computer Technology, Pune, India
Abstract
Purpose –Sentiment analysis is the subfield of data mining, which is profusely used for studying the
opinions of the users by analyzing their suggestions on the Web platform. It plays an important role in
the daily decision-making process, and every decision has a great impact on daily life. Various
techniques including machine learning algorithms have been proposed for sentiment analysis, but still,
they are inefficient for extracting the sentiment features from the given text. Although the improvement
in sentiment analysis approaches, there are several problems, which make the analysis inefficient and
inaccurate. This paper aims to develop the sentiment analysis scheme on movie reviews by proposing a
novel classifier.
Design/methodology/approach –For the analysis, the movie reviews are collected and subjected
to pre-processing. From the pre-processed review, a total of nine sentiment related features are
extracted and provided to the proposed exponential-salp swarm algorithm based actor-critic neural
network (ESSA-ACNN) classifier for the sentiment classification. The ESSA algorithm is developed
by integrating the exponentially weighted moving average (EWMA) and SSA for selecting the
optimal weight of ACNN. Finally, the proposed classifier classifies the reviews into positive or
negative class.
Findings –The performance of the ESSA-ACNN classifieris analyzed by considering the reviews present
in the movie review database. From, the simulation results, it is evident that the proposed ESSA-ACNN
classifier hasimproved performance than the existing works by having the performanceof 0.7417, 0.8807 and
0.8119, forsensitivity, specificity and accuracy, respectively.
Originality/value –The proposed classifier can be applicable for real-worldproblems, such as business,
politicalactivities and so on.
Keywords Sentiment analysis, Classification, ACNN classifier, Salp swarm algorithm,
Movie review database
Paper type Research paper
1. Introduction
In the recent era, a number of specific tools have been developed for the sentiment analysis
of the reviews arriving in the online platform. The analysis can be done by evaluating the
opinion of the user-providedon the specified topic. The sentimentanalysis can be applied in
several fields, such as news forum, business organization, enterprise and online products
(Prabhat and Khullar, 2017). Opinion mining is related to the field of natural language
processing (NLP).The opinion from the users may differ in language, andhence, NLP can be
considered as the NLP problem. Improving the opinion mining toward the NLP helps
the human-computer interaction.Sentiment analysis can be expressed as the task of finding
the user opinion by analyzing the review.The opinions posted by the user can have positive,
Sentiment
analysis in
social media
457
Received3 December 2018
Revised13 April 2019
Accepted25 April 2019
VINEJournal of Information and
KnowledgeManagement Systems
Vol.49 No. 4, 2019
pp. 457-476
© Emerald Publishing Limited
2059-5891
DOI 10.1108/VJIKMS-12-2018-0116
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/2059-5891.htm
negative or neutral polarity. Sentiment analysis is carried out in the following ways:
document level, sentence level and entity-aspect level. The document-level sentiment
analysis analyzes the entire document for arriving at the opinion class, and the sentence
level analyzes the specific sentence only for the analysis. In the entity level, the results
directly concentrate on the opinions (Liu, 2012). During the sentiment analysis through the
sentence level, it is necessary to understandthe opinions from the phrases and idioms of the
sentence. Sentences, such as interrogative and conditional may lack in opinion in several
cases. Besides, the presence of sarcasm in the sentence may provide the opposite opinion.
Hence, to overcome these challenges, innovative sentiment classifiers need to be designed
(Vaghela and Jadav, 2016).
Developing the sentiment classification systems for sentiment analysis is an interesting
topic of research. Most of the researchers have contributed to the sentiment classification. The
sentiment classification scheme aims to categorize the hidden opinion in the review as a
positive and negative class. For performing the sentiment classification, the neural network
(NN) based techniques are considered to be more successful than other classifiers. Sentiment
classification approach suffers from the domain dependency problem as the words
representing the sentiment may have a different meaning for a different domain (Yuan et al.,
2018). The initial stage in the sentimental analysis is extracting the sentimental related features
from the review using the feature extractor algorithms. The traditional machine learning
algorithms defined in the literature use the extracted feature vectors and perform the sentiment
classification. Sentiment classification estimates the opinions by analyzing the texts. Similarly,
for the sentiment classification using the documents, the features are extracted from the
documents and the opinions are classified. As the features extracted from the sentences are
large, it is necessary to select the required features through the feature selection (Shang et al.,
2016). Sentiment classification approach relays on the bag-of-words or bag-of-n-grams for the
feature extraction (Maas et al., 2011). These techniques omit the general grammar in the
sentence and view the text as the combination of the word (Nio and Murakami, 2018).
The classification algorithms defined for the sentiment analysis fall on three categories,
and they are enlistedas follows:
(1) machine learning-based algorithms;
(2) hybridization schemes; and
(3) lexicon-based techniques.
The machine learning-based schemes perform the sentiment classification by using the
trained and test databases. The lexicon-based schemes perform the classification using the
dictionaries; hence, training is not necessary. The hybrid approach uses both the lexicon-
based techniques and machine learning algorithms for the sentiment classification. For the
sentiment analysis, several tools, such as emoticons, SentiStrength, linguistic inquiry and
word count, SentiWordNet,happiness index, AFINN, SenticNet, positive andnegative affect
schedule-Twitter, Sentiment140, entropy weighted genetic algorithm, NRC and feature
relation network (FRN) (D’Andrea et al.,2015) have been developed. Some of the commonly
used sentiment classifiers are support vector machine (SVM) (Schölkopf and Smola, 2002)
and gradient boosting decision tree (GBDT)(Friedman, 1999). These classifiers classify the
reviews into positive or negative class. The classifiers designed for the sentiment
classification have their strengthsor weaknesses. The SVM classifier has poor performance
while the data are sparse, meanwhile, the GBDT suffers from overfitting issue. As the
existing techniquessuffer from several issues, it is necessary to define the new procedure for
the sentiment classification(Kai et al.,2017).
VJIKMS
49,4
458
To continue reading
Request your trial