An Efficient Approach for Sentiment Analysis using Convolutional
Neural Network
Alpna Patel
1
, Arvind Kumar Tiwari
1
, S. S. Ahmad
2
1
Department of Computer Science & Engineering, Kamla Nehru Institute of Technology, Sultanpur, India
2
KL Ghurair University, Dubai, UAE
Keywords Natural Language Processing; Sentiment Analysis; Deep Learning Classifiers; LSTM; CNN
Abstract: Sentiment analysis is a substantial area of research nowadays. Many researchers have proved the subject in
recent years. The reason behind that is the rapidly growing opinionated data on social media. With the aim
of surmounting this obstacle, we introduce an efficient approach for sentiment analysis that ensemble the
advantages of two deep learning models. Sentiment mining is the process of extracting opinion, feelings,
emotions and attitude towards a specific task. Here, we have collected the IMDB movie review dataset as
well as used two kind of deep learning classifiers to analyze the experimental result. Hence, the
contemplated models are Long Short Term Memory and Convolutional Neural Network. The efficiency of
the proposed model is compared with other traditional approaches in experimental work and outcome of the
result shows that the ensemble approach can effectively improve the accuracy to predict the sentiments.
1 INTRODUCTION
Natural Language Processing(NLP) is a discipline in
computer science that deals with the communication
between machines and humans in natural language.
NLP is dealt with enabling machines to understand
and develop the human language. There are
numerous application of NLP such as text
categorization, sentence classification, Named Entity
Recognition, speech recognition, Language
detection and summarization, character recognition,
structure prediction, decision making, computer
vision, and others. It is one of the substantial
applications of Natural Language Processing.
Sentiment classification has become the most
effective research area in NLP due to an increase in
public interest in movies, outlooks, and elections. It
aims to identify opinions, emotions, and attitudes
towards specific tasks like movies, events, elections
and many more. The rich data sources are used to
perform analysis, likewise social media sites,
blogging sites, RSS news feed etc. To perform text
classification, it includes different processes
likewise feature extraction, emotion detection, and
attitude extraction. In the real world, there exist
various application areas likewise in the medical
field, box office, commercial sites, politics, and
business intelligence.
If we look further in the analysis process, there
are used some notations such as subjective,
objective, polarity and sentiment level. When the
sentence holds subjective orientation in a given
piece of text then it is known as subjective. For
example “Newton is an awesome movie.” An
objective can be defined as the sentence holds
objective orientation. For example “Amit Masurkar
is the director of Newton movie”. The other notation
is polarity detection and it can be split up into three
parts likewise positive, negative and neutral.
Sentences show different levels of polarity such as I
love my friends, I hate liars,I usually go outside
every weekend. The first sentence contains positive
polarity, the second sentence contains negative
polarity and the third one contains neutral polarity
respectively.
2 RELATED WORK
In order to do the analysis of sentiments, numerous
researchers have made their efforts to ensemble deep
learning andmachine learning classifiers to achieve
outstanding result in ongoing years. The related
12
Patel, A., Tiwari, A. and Ahmad, S.
An Efficient Approach for Sentiment Analysis using Convolutional Neural Network.
DOI: 10.5220/0010561700003161
In Proceedings of the 3rd International Conference on Advanced Computing and Software Engineering (ICACSE 2021), pages 12-21
ISBN: 978-989-758-544-9
Copyright
c
 2022 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

work briefly elaborates on the numerous researches,
associated to text classification of social media
contents about people’s sentiments, feelings, reviews
towards various subjects like movies and products
using NLP techniques.
The authors have proposed an efficient deep
learning classifier for sentiment classification which
calculates the accuracy of 82.53% on Bengali text.
In order to evaluate the performance, they used two
deep neural network models such as deep RNN with
BiLSTM (Sharfuddin AA, 2018). Deep
learningtechniques achieved significant results in
textanalysis. (Chen S, 2018) has been proposed an
innovative method for target-based sentiment
analysis which reduces the training time of the
proposed model through regional LSTM. Deep
learning models are frequently used in NLP
applications. An efficient approach has been
proposed for the multi-domain system that is based
on word embedding. The tool named NeuroSent
gives an accuracy of 85.15% by using the Amazon
web site dataset for multi-domain (Dragoni M,
2017). Some of the deep learning models are based
on sentence classification in Natural Language
Processing and some of them are based on
traditional models like SVM, RNN, LSTM and
much more. In this literature survey, we basically
study the ensemble approaches to improve the
performance. Some authors proposed an ensemble
method for text classification by using Vietnamese
text. In this technique, they have merged the
traditional modelswith deep learning models and
achieved the remarkable result that is 89.19%
(Nguyen HQ, 2018). Some authors have done their
research study in artificial intelligence on deep
learning models. The review basically focuses on
text classification by using different datasets
(AlwehaibiA). A novel approach has been proposed
by authors that are based on an ensemble of two
models and achieves the accuracy 89%. They have
used the IMDB movie review datasetfor the
analysis process [6]. The ensemble approach gives
outstanding results over traditional models in text
analysis. We noticed that the ensemble approaches
performed much better than traditional models.
Some authors have proposed a machinelearning-
based approach for improving the performance of
sentiment analysis. They have used LSTM, Naïve
Bayes and SVM for analysis process (Day MY,
2017). Some authors have gained remarkable results
in the field of Natural Language Processing by using
deep learning techniques for text classification
(Hassan A, 2017). In order to do classification, the
authors have used Tibetan microblogs and achieved
the result up to the mark (Sun B, 2018).
The deep learning-based models improve the
result in the field of NLP over the years. The authors
have proposed a model named SentiWordNet and
achieved better results. The model used word2Vec
to perform analysis (Alshari EM, 2018). A novel
approach ECNN has proposed that is used to
identify opinion, polarity, and emotions in
microblogs (Yang G, 2019). Numerous researchers
have proposed a model related to sentiment
classification. They have used word embedding
methods of learning at the word level and sentence
level (Zhang Z, 2015). In this field of research, we
can achieve better results by using deep learning-
based approaches. The authors have proposed an
ensemble approach that is the result of two machine
learning models CNN, SVM for text sentiment
analysis (Cai J, 2018). Many researchers have
proposed an efficient method to perform
classification processes on the IMDB review dataset
and they found that RNN performs effectively in
terms of words semantic and they achieved an
accuracy of 89.8% (Zharmagambetov AS, 2015). In
order to perform analysis, there are different
parameters used such as feature extraction, opinion
mining, applying different kinds of machine learning
algorithms. An approach has been proposed that is
based on a machine learning and Lexicon based
features to perform sentiment analysis on the movie
review dataset (Bandana R. 2018). Word
embedding is a technique that is used to convert the
words into vectors. The researchers have been using
the word embedding method for sentiment analysis.
An efficient approach has been proposed for
sentiment analysis by using word embedding (Deho
BO, 2018).
The comparative study is done by researchers on
different tools and techniques of machine learning
approaches of Natural Language Processing. The
paper presents the various feature selection methods
and machine learning techniques (Mejova Y, 2009).
A joint framework has been proposed for sentence
classification based on CNN and RNN. It gives the
accuracy of 93.3%, 48.8%,89.2% on the movie
review dataset, fine-grained and binary accuracy
respectively (Hassan A, 2018). The authors have
proposed a deep learning approach for the
classification process. (Patel Alpna, 2019) have
achieved an accuracy of 87.42% by using RNN. The
researchers have presented a novel approach to
extract features and textual modalities and to
improve the performance, they have used a deep
CNN approach for the classification process (Poria
An Efficient Approach for Sentiment Analysis using Convolutional Neural Network
13


S., 2017). (Ruangkanokmas, 2017) have been
proposed a model named Deep Belief Network.
They have used a semi-supervised learning method
called Deep Belief Network. The authors have
proposed a method for users’ interests classification
based on CNN and Word2Vec. The proposed
framework is based on deep learning and they used
CBOW as a feature extraction algorithm and SVM
for classification that gives the accuracy 96% on the
IMDB movie review dataset (Om, A. H., 2017).
(Changliang Li, 2018) builds the Chinese Sentiment
Treebank over social data and further introduces an
approach named Recursive Neural deep model for
the analysis process. The authors have been using
word vectorization to extract corpus features and
PCA to reduce dimension (Li, C., 2014). (Kumar
Ravi, 2018) performed sentiment analysis on article
citation sentences and they have been proposed an
ensemble method for deep learning. The authors
have been performing sentiment analysis by using
word embedding techniques like word2vec and
Glove as a pre-trained vector (Henry, S., 2017).
The author DoaMohey El-Din Mohamed
Hussein has done the comperative study on
sentiment mining challenges (Hussein, 2018).
(Feilong Tang, 2019) have been proposed a model
named JABST stands for joint aspect-based
sentiment topic for multi-grained aspect by using
supervised learning method to process the model.
The authors have been presented as an attention
mechanism for target-level and context-level
attention. The presented mechanism is more
effective for sentiment feature (Yang, C., 2019).
(AbinashTripathy, 2015) focuses on machine
learning techniques. They obtained the result by
using Naïve Bayes and SVM and show the
comparison on the movie review dataset. The
authors have shown the comparison between two
classifiers named Deep Recurrent Neural Network
and SVM. They concluded that the Support Vector
Machine performs much better than Deep Recurrent
Neural Network (Al-Smadi, M., 2018). (Rodrigo
Moraeset., 2013) have been presented with an
empirical comparison between ANN and SVM on
the movie review dataset and they found ANN
performs better than SVM. The authors have
presented an ensemble deep learning method for
sentiment analysis by using the IMDB movie review
dataset (Araque, O.m, 2017).
(Abinash, 2016) presented a novel technique for
sentiment mining. They used an ensemble of
classifiers named Naïve Bayes, SVM and Stochastic
Gradient Descent and achieved an accuracy of
83.33% on the IMDB movie review dataset.
(Giatsoglou Maria, 2017) have been presented with
an approach named RAE and achieved an accuracy
of 83.99%. The authors have been performed better
by using micro-blogs text (Zhang, S., 2018).
(CagatayCatal, 2017) have proposed a model for
analysis. The model has achieved the accuracy of
86.13%. The authors have proposed a model named
ML-KNN for classification. They used the
unsupervised learning method (Zhang, M. L., 2007).
The authors have presented a review on research
topics, venues and top-cited papers (Mäntylä, M. V,
2018). The authors have been presented with a
boosted ensemble-based classifier for sentiment
analysis (Athar A, 2017). Anuj Sharma et. al. have
been presented with aboosted approach based on
SVM (Sharma A, 2013; Dumoulin J, 2015). The
authors have been using a hierarchical approach for
analysis (Sharma A, 2013). (Sharma A, 2012) gave
the literature survey on the ensemble of the classifier
for sentiment mining. The authors have been using a
deep neural network for sentiment prediction (Piao
G, 2018). (Wan X., 2008) applied ensemble
techniques for unsupervised Chinese
sentimentanalysis (Piao G, 2018).
3 PROPOSED APPROACH
This section presents the detailed overview of the
proposed model to classify sentiments in movie
domain. The proposed approach uses two classifier
of deep learning i.e Long Short Term Memory and
Convolutional Neural Network. It uses word
embeddings as input and takes them to LSTM for
feature extraction and further output is given to
CNN and followed by classification layer. The
following step is followed by a proposed approach:
 Word Embedding method is used to
convert the word into featured vectors in
the given text.
 The hybrid model takes the advantages of
two deep learning approaches such as
LSTM and CNN for feature extraction.
 The classification layer uses the Softmax
activation function to compute the
predictive probability.
3.1 Long Short Term Memory
Architecture
Long Short Term Memory is a deep neural network
model that is used for sequential information and
proposed by (SeppHochreiter, 1998). LSTM is RNN
ICACSE 2021 - International Conference on Advanced Computing and Software Engineering
14

architecture that REMEMBERS values over
arbitrary intervals and used to resolve a problem of
vanishing gradient problem (Schuster M, 1997;
Hochreiter S., 1998; Zhou C, 2015). LSTM enables
RNN’s to remember their inputs over a long period
of time. It uses input gate, forget gate and output
gate as gates. The input gate is used for new input in,
forget gate for whether information delete or not and
the output gate for output at the current time step
(Liu P, 2016; Tang D, 2015). The LSTM
Architecture is depicted in figure 3.
Input Output

Input Gate  Output Gate 










Forget Gate
 C t
O
t
u
t
𝑁
𝑡
Figure 3.1: The LSTM gate architecture
The sigmoid function gives the output value between
0 or 1. The equation is given below:
Input gate equation
𝐼

𝜎𝑤

ℎ
,
𝑥

𝑏

(3.1)
Forget gate equation
𝐹

𝜎𝑤

ℎ
,
𝑥

𝑏

(3.2)
Output gate equation
𝑂

𝜎𝑤

ℎ
,
𝑥

𝑏

 (3.3)
where I
g
represents the input gate, F
g
as forget gate,
O
g
as output gate, σ denotes the sigmoid function, w
x
shows the weight for the respective gate (x) neurons,
h
(g-1)
represents the output of previous LSTM block,
x
g
shows input of current timestamp and b
x
shows
the biases for relative gates. The equations are given
as follows:
Cell state equation
ć

𝑡𝑎𝑛ℎ𝑤

ℎ
,
𝑥

𝑏

 (3.4)
Candidate cell equation
𝑐

𝑓

∗𝑐
,
𝑖

∗ć

(3.5)
Final output equation
ℎ

𝑜

∗𝑡𝑎𝑛ℎ𝑐

) (3.6)
where ct denotes the cell state at t timestamp and ćt
shows candidate for cell state at t timestamp. we
evaluate the above equation that our cell state knows
that what it needs to forget from previous state
(ft*ct-1) at any timestamp and what it should include
from current timestamp (it* ćt) and * represents the
element-wise multiplication. Next, we refine the cell
state and pass it to the activation function (Nowak J,
2017; Nabil M, 2016).
3.2 Convolutional Neural Network
Architecture
Convolutional Neural Network was initially
developed in the neural network image processing
community. CNN involves basically two operations
for text classification such as convolution and
pooling as feature extractors (Lai S, 2015). CNN
uses two kinds of pooling as feature extractors such
as max-pooling and average-pooling. The max-
pooling elects a maximum number of values in the
input feature map and the other selects the average
number of values in the region (Lee JY, 2016).
Convolutional Neural Network is also applied to
the text in Natural Language Processing. When we
use CNN for text instead of images, then we use the
1-D array to represent context (Yin W, 2017).
Mostly in Natural Language Processing (Nasukawa
T, 2003) task, CNN is used in sentiment analysis
which means classifying a sentence into a set of
predetermined categories. In order to perform text
classification, each sentence is known as matrix.
Every row of the matrix shows the one token,
typically a word. We can say that each row is vector
and impersonate a word (Guggilla C, 2016).
In the NLP task (Kao A, 2007), we have used
filters over full rows of matrices. The following
model is as follows:
An Efficient Approach for Sentiment Analysis using Convolutional Neural Network
15

Figure 3.2: Convolutional Neural Network Model Structure
If we analyze the above model, then we see that
the first layer embed words into low dimensional
vector after that next layer performs convolutions
over the embedded word vectors using multiple filter
sizes and classify the result using a softmax layer
(Conneau A, 2016). We have used CNN for text
analysis for sentiment classification on the IMDB
movie review dataset (www.kaggle.com).
3.3 The Embedding Layer
This layer of network changes the words into real-
valued featured vectors. Our model takes input in the
form of vectors. In order to convert words into real-
valued featured vectors, we have used the word
embedding method. Word embedding is the process
of representing the word or phrase into a vector. The
word is stored in vocabulary and arranged
sequentially. We have used distributed
representation to overcome the dimensionality
problem.
3.4
The Proposed Approach
In this subsection, we have discussed the detailed
framework architecture of our proposed approach.
The proposed approach ensemble the advantages of
two deep learning classifiersi.e LSTM and CNN.
Previously, we provideda detailed architectural
explanation of LSTM and CNN. If we look upon
RNN, the Long Short Term Memory approach
performs efficiently related to feature extraction. In
this approach, the convolutional layer uses Max-
pooling. The ensemble approach uses the embedding
layer to take the input in the form of words and pass
it to the multi-layer LSTM model. The multilayer
LSTM generates the output and it is further passed
to the convolutional layer as an input for further
process. After the output of the convolutional layer
passed to the classification layer for the
classification process. The convolutional layer
extracts the features of text sequences.
LSTM model gives the output; L= [L
1,
L
2,
L
3,
…,
L
t
]
T
, L
t
denotes the t
th
words of the n-dimensional
vector in a given sequence. The number of LSTM
hidden layers and the vector length both are equal.
C= [C
0,
C
1,
C
2,…,
C
n-1
] will produce one value at t time
step as follow:
U
Ct
= ReLU [
∑


𝑂

T
C
i
) +b] (3.7)
where, b denotes the bias value and combination of
b and C are used RELU activation function( C(y) =
max (U, y)). It shows the single convolutional filter
to extract the value of features from a given text
sequence. The proposed approachis used multiple
convolutional filters to extract variousfeatures. Next,
the max-pooling layer formed and passed to the fully
connected layer. The classification layer uses the
Softmax activation function to calculate the
predictive probability for all categories. The
following equation shows the probability y as
category w:
P(x
(i)
= w|y
(i)
; θ) =


 

∑




 

(3.8)
ICACSE 2021 - International Conference on Advanced Computing and Software Engineering
16



Embedding
Layer





LSTM
















CONVOLUTIONALLAYER

MAX‐POOLINGLAYER
SoftMaxFunction
Out
p
ut
Input
Figure 3.3: The Framework of the Proposed Approach
4 RESULT AND DISCUSSION
This section briefly discusses the experimental setup
and measures the result of the approach. The
performance was evaluated by using different
parameters. The following subsection contains
detailed information about datasets, experimental
setup, confusion matrix, etc. These are as follows:
4.1 Dataset Description
In order to evaluate the performance, we have
used the IMDB dataset. It includes 25000numbers of
data on movie reviews from the Kagglewebsite that
contain binary values named positive and negative
sentiment. This paper uses the IMDB movie review
dataset for the purpose of experimental work, the
dataset contains 25000 numbers of data in which a
75% number of data for the training set and 25%
number of data for the validation set. After the split
the dataset, further we perform dataset preprocessing
tasks to clean the raw data and break the sentences
into words and words into text. The detailed process
is given in the introduction section.
4.2 Environmental Setup and Param
Setting
Anaconda is a package provider for machine
learning models by using python language.
Tensorflow is the framework that provides the
environment for machine learning models. We have
used python version 3.6.5, jupyter notebook and
Keras for implementing deep neural network
models. Keras is the higher-level API that uses
TensorFlow in backend and it is used for sequential
modeling. In this experiment work, we used
categorical cross-entropy for loss, Adam optimizer
with learning rate 0.001, the batch size is 32 and the
hidden layer of LSTM is 128 with dropout 0.2.
4.3 Performance Measure
If we look to the performance of the proposed
model, a confusion matrix has been used that
contains some parameters such as tp as true positive,
tn as true negative, fp as false positive, and fn as
false negative on test data. The confusion matrix is
given in Table I as follows:
It is used to calculate the accuracy of the proposed
approach by using the following formula:
Accuracy=


X 100%
The parameter accuracy is used to validate the
proposed hybrid model by using the test set and
validate set. The Table II depicted the comparative
result of the proposed approach with deep learning
approaches:
An Efficient Approach for Sentiment Analysis using Convolutional Neural Network
17

Table 1: Confusion Matrix
Label 1 (Predicted) Label 2 (Predicted)
el 1 (Actual) tn fp
el 2 (Actual) Fn tp
Table 2: Accuracy Comparison of Ensemble Approach with Deep Neural Network Model for Sentiment Analysis
Deep Learning Models Time (per
second)
Test Accuracy Valid
Accuracy
1.
Bidirectional LSTM 2119 88% 84.85%
2.
LSTM 1164 89% 85%
3.
CNN (max- pooling) 575 97% 85.52%
4.
CNN+GRU 551 95% 85.61%
5.
Bidirectional GRU 2676 89% 85.64%
6.
DeLC model 1026 91% 85.78%
Figure 3.4: Comparative study of proposed approach over Traditional Approaches
Here, we compare the outcome of proposed
approach over traditional approaches. The
experimental work give the result analysis as
follows, Bidirectional LSTM gives the accuracy
84.85%, LSTM 85%, CNN 85.52%, CNN-GRU
85.61%, Bidirectional GRU 85.64% and DeLC
hybrid model gives the valid accuracy 85.78%. We
can see that the proposed approach gives the
outstanding result over other deep learning
approaches. The figure 3.4 shows the comparative
result of the proposed approach with deep learning
ICACSE 2021 - International Conference on Advanced Computing and Software Engineering
18

models including some parameters such as time (per
second), test accuracy and valid accuracy.
The figure 3.4 shows the comparison of existing
approaches with the proposed approach in terms of
time, test accuracy and valid accuracy. If we look
forward to the analysis process, we found that the
proposed approach may perform better in relation to
other deep learning approaches. The experimental
result shows that the proposed approach performs
effectively with an accuracy of 85.78%.
5 CONCLUSION
Sentiment classification is the method of extracting a
user’sview as positive or negative for a specific task.
We have introduced an efficient approach for
sentiment analysis that ensemble the advantages of
two deep learning models name as Long Short Term
Memory and Convolutional Neural Network. LSTM
overcomes the vanishing gradient problem and
preserves historical information of long term text
dependencies. Further, CNN extracts the feature of
context. In this paper, the proposed ensemble
approach efficiently improves the accuracy of
sentiment classification. The proposed ensemble
approach gives an accuracy of 85.78% on IMDB
movie review data. It is found that the proposed
approach performs better than other deep learning
approaches.
ACKNOWLEDGEMENT
I would like to extend my deep gratitude to Dr.
Arvind Kumar Tiwari, Associate Professor, KNIT,
Sultanpur for being so helpful, galvanizing,
generous guidance and also continuous
encouragement and support.
REFERENCES
Alessia D, Ferri F, Grifoni P, Guzzo T. Approaches, tools
and applications for sentiment analysis
implementation. International Journal of Computer
Applications. 2015 Jan 1;125(3).
Alshari EM, Azman A, Doraisamy S, Mustapha N,
Alkeshr M. Effective Method for Sentiment Lexical
Dictionary Enrichment Based on Word2Vec for
Sentiment Analysis. In2018 Fourth International
Conference on Information Retrieval and Knowledge
Management (CAMP) 2018 Mar 26 (pp. 1-5). IEEE.
Al-Smadi, M., Qawasmeh, O., Al-Ayyoub, M., Jararweh,
Y., & Gupta, B. Deep Recurrent neural network vs.
support vector machine for aspect-based sentiment
analysis of Arabic hotels’ reviews. Journal of
Computational Science, 27, 2018 (pp. 386–393).
https://doi.org/10.1016/j.jocs.2017.11.006.
Alwehaibi A, Roy K. Comparison of Pre-Trained Word
Vectors for Arabic Text Classification Using Deep
Learning Approach. In2018 17th IEEE International
Conference on Machine Learning and Applications
(ICMLA) 2018 Dec 17 (pp. 1471-1474). IEEE.
Araque, O., Corcuera-Platas, I., Sánchez-Rada, J. F., &
Iglesias, C. A. Enhancing deep learning sentiment
analysis with ensemble techniques in social
applications. Expert Systems with Applications, 77,
2017 (pp. 236–246).
https://doi.org/10.1016/j.eswa.2017.02.002.
Athar A, Butt WH, Anwar MW, Latif M, Azam F.
Exploring the Ensemble of Classifiers for Sentimental
Analysis: A Systematic Literature Review.
InProceedings of the 9th International Conference on
Machine Learning and Computing 2017 Feb 24 (pp.
410-414). ACM.
Bandana R. Sentiment Analysis of Movie Reviews Using
Heterogeneous Features. In2018 2nd International
Conference on Electronics, Materials Engineering &
Nano-Technology (IEMENTech) 2018 May 4 (pp. 1-
4). IEEE.
Cai J, Li J, Li W, Wang J. Deep learning Model Used in
Text Classification. In2018 15th International
Computer Conference on Wavelet Active Media
Technology and Information Processing
(ICCWAMTIP) 2018 Dec 14 (pp. 123-126). IEEE.
Catal, C., &Nangir, M. A sentiment classification model
based on multiple classifiers. Applied Soft omputing
Journal, 50, 2017 (pp.135–141).
https://doi.org/10.1016/j.asoc.2016.11.022.
Chen S, Peng C, Cai L, Guo L. A Deep Neural Network
Model for Target-based Sentiment Analysis. In2018
International Joint Conference on Neural Networks
(IJCNN) 2018 Jul 8 (pp. 1-7). IEEE.
Conneau A, Schwenk H, Barrault L, Lecun Y. Very deep
convolutional networks for text classification. arXiv
preprint arXiv:1606.01781. 2016 Jun 6.
Day MY, Lin YD. Deep learning for sentiment analysis on
google play consumer review. In2017 IEEE
International Conference on Information Reuse and
Integration (IRI) 2017 Aug 4 (pp. 382-388). IEEE.
Deho BO, Agangiba AW, Aryeh LF, Ansah AJ. Sentiment
Analysis with Word Embedding. In2018 IEEE 7th
International Conference on Adaptive Science &
Technology (ICAST) 2018 Aug 22 (pp. 1-4). IEEE.
Dragoni M, Petrucci G. A neural word embeddings
approach for multi-domain sentiment analysis. IEEE
Transactions on Affective Computing. 2017 Oct
1;8(4):457-70.
Dumoulin J, Affi D, Mugellini E, AbouKhaled O, Bertini
M, Del Bimbo A. Affect recognition in a realistic
movie dataset using a hierarchical approach.
InProceedings of the 1st International Workshop on
An Efficient Approach for Sentiment Analysis using Convolutional Neural Network
19


Affect & Sentiment in Multimedia 2015 Oct 30 (pp.
15-20). ACM.
Ge, X., Jin, X., Miao, B., Liu, C., & Wu, X. Research on
the Key Technology of Chinese Text Sentiment
Analysis. Proceedings of the IEEE International
Conference on Software Engineering and Service
Sciences, ICSESS, 2018-November,2019(pp. 395–
398).
Giatsoglou, M., Vozalis, M. G., Diamantaras, K., Vakali,
A., Sarigiannidis, G., &Chatzisavvas, K. C. Sentiment
analysis leveraging emotions and word embeddings.
Expert Systems with Applications, 69, 2017 (pp.214–
224).
Guggilla C, Miller T, Gurevych I. CNN-and LSTM-based
claim classification in online user comments.
InProceedings of COLING 2016, the 26th
International Conference on Computational
Linguistics: Technical Papers 2016 Dec (pp. 2740-
2751).
Hassan A, Mahmood A. Convolutional recurrent deep
learning model for sentence classification. Ieee
Access. 2018;6:13949-57.
Hassan, A., &Mahmood, A. Deep Learning approach for
sentiment analysis of short texts. 2017 3rd
International Conference on Control, Automation and
Robotics, ICCAR 2017, (pp. 705–710).
https://doi.org/10.1109/ICCAR.2017.7942788
Henry, S. A New Story: Purposeful storytelling and
designing with data | Sarah Henry - Academia.edu.
Harmony Institute, 2017(pp. 534–539).
Hochreiter S. The vanishing gradient problem during
learning recurrent neural nets and problem solutions.
International Journal of Uncertainty, Fuzziness and
Knowledge-Based Systems. 1998 Apr;6(02):107-16.
https://www.kaggle.com/oumaimahourrane/sentiment-
analysis-cnn-vs-lstm/data.
Hussein, D. M. E. D. M. A survey on sentiment analysis
challenges. Journal of King Saud University -
Engineering Sciences, 30(4),2018 (pp.330–338).
https://doi.org/10.1016/j.jksues.2016.04.002 .
Kao A, Poteet SR, editors. Natural language processing
and text mining. Springer Science & Business Media;
2007 Mar 6.
Kim SM, Hovy E. Determining the sentiment of opinions.
InProceedings of the 20th international conference on
Computational Linguistics 2004 Aug 23 (p. 1367).
Association for Computational Linguistics.
Lai S, Xu L, Liu K, Zhao J. Recurrent convolutional
neural networks for text classification. InTwenty-ninth
AAAI conference on artificial intelligence 2015 Feb
19.
Lee JY, Dernoncourt F. Sequential short-text classification
with recurrent and convolutional neural networks.
arXiv preprint arXiv:1603.03827. 2016 Mar 12.
Li, C., Xu, B., Wu, G., He, S., Tian, G., &Hao, H.
Recursive deep learning for sentiment analysis over
social data. Proceedings - 2014 IEEE/WIC/ACM
International Joint Conference on Web Intelligence
and Intelligent Agent Technology - Workshops,WI-
IAT 2014, 2, 1388–1429.
Liu P, Qiu X, Huang X. Recurrent neural network for text
classification with multi-task learning. arXiv preprint
arXiv:1605.05101. 2016 May 17.
Mäntylä, M. V., Graziotin, D., &Kuutila, M. The
evolution of sentiment analysis—A review of research
topics, venues, and top cited papers. Computer Science
Review, 27, 2018(pp.16–32).
https://doi.org/10.1016/j.cosrev.2017.10.002
Medhat W, Hassan A, Korashy H. Sentiment analysis
algorithms and applications: A survey. Ain Shams
engineering journal. 2014 Dec 1;5(4):1093-113.
Mejova Y. Sentiment analysis: An overview.
Comprehensive exam paper. Computer Science
Department. 2009:1-34.
Mohammad S, Dunne C, Dorr B. Generating high-
coverage semantic orientation lexicons from overtly
marked words and a thesaurus. InProceedings of the
2009 Conference on Empirical Methods in Natural
Language Processing: Volume 2-Volume 2 2009 Aug
6 (pp. 599-608). Association for Computational
Linguistics.
Moraes, R., Valiati, J. F., &GaviãoNeto, W. P. Document-
level sentiment classification: An empirical
comparison between SVM and ANN. Expert Systems
with Applications, 40(2), 2013 (pp.621–633).
https://doi.org/10.1016/j.eswa.2012.07.059.
Nabil M, Atyia A, Aly M. Cufe at semeval-2016 task 4: A
gated recurrent model for sentiment classification.
InProceedings of the 10th International Workshop on
Semantic Evaluation (SemEval-2016) 2016 (pp. 52-
57).
Nasukawa T, Yi J. Sentiment analysis: Capturing
favorability using natural language processing.
InProceedings of the 2nd international conference on
Knowledge capture 2003 Oct 23 (pp. 70-77). ACM.
Nguyen HQ, Nguyen QU. An Ensemble of Shallow and
Deep Learning Algorithms for Vietnamese Sentiment
Analysis. In2018 5th NAFOSTED Conference on
Information and Computer Science (NICS) 2018 Nov
23 (pp. 165-170). IEEE.
Nowak J, Taspinar A, Scherer R. LSTM recurrent neural
networks for short text and sentiment classification.
InInternational Conference on Artificial Intelligence
and Soft Computing 2017 Jun 11 (pp. 553-562).
Springer, Cham.
Om, A. H. CNN for Users Interests Classification. 2017.
Patel A, Tiwari AK. Sentiment Analysis by using
Recurrent Neural Network. Available at SSRN
3349572. 2019 Feb 8.
Piao, G., & Breslin, J. G. (2018, April). Financial aspect
and sentiment predictions with deep neural networks:
an ensemble approach. In Companion Proceedings of
the The Web Conference 2018 (pp. 1973-1977).
Poria, S., Chaturvedi, I., Cambria, E., &Hussain, A.
Convolutional MKL based multimodal emotion
recognition and sentiment analysis. Proceedings -
ICACSE 2021 - International Conference on Advanced Computing and Software Engineering
20

IEEE International Conference on Data Mining,
ICDM,2017 (pp. 439–448).
Ravi K, Ravi V. A survey on opinion mining and
sentiment analysis: tasks, approaches and applications.
Knowledge-Based Systems. 2015 Nov 1;89:14-46.
Ravi, K., Setlur, S., Ravi, V., &Govindaraju, V. Article
Citation Sentiment Analysis Using Deep Learning.
Proceedings of 2018 IEEE 17th International
Conference on Cognitive Informatics and Cognitive
Computing,ICCI*CC2018, (pp.78–85).
Ruangkanokmas, P., Achalakul, T., &Akkarajitsakul, K.
Deep Belief Networks with Feature Selection for
Sentiment Classification. Proceedings - International
Conference on Intelligent Systems, Modelling and
Simulation, ISMS, 2017 (pp. 9–14).
Schuster M, Paliwal KK. Bidirectional recurrent neural
networks. IEEE Transactions on Signal Processing.
1997 Nov;45(11):2673-81.
Sharfuddin AA, Tihami MN, Islam MS. A Deep Recurrent
Neural Network with BiLSTM model for Sentiment
Classification. In2018 International Conference on
Bangla Speech and Language Processing (ICBSLP)
2018 Sep 21 (pp. 1-4). IEEE.
Sharma A, Dey S. A boosted svm based ensemble
classifier for sentiment analysis of online reviews.
ACM SIGAPP Applied Computing Review. 2013 Dec
1;13(4):43-52.
Sharma A, Dey S. A boosted SVM based sentiment
analysis approach for online opinionated text.
InProceedings of the 2013 research in adaptive and
convergent systems 2013 Oct 1 (pp. 28-34). ACM.
Sharma A, Dey S. An artificial neural network based
approach for sentiment analysis of opinionated text.
InProceedings of the 2012 ACM Research in Applied
Computation Symposium 2012 Oct 23 (pp. 37-42).
ACM. Sharma A, Dey S. An artificial neural network
based approach for sentiment analysis of opinionated
text. InProceedings of the 2012 ACM Research in
Applied Computation Symposium 2012 Oct 23 (pp.
37-42). ACM.
Sun, B., Tian, F., & Liang, L. Tibetan Micro-Blog
Sentiment Analysis Based on Mixed Deep Learning.
ICALIP 2018 - 6th International Conference on
Audio, Language and Image Processing,2018(pp.
109–112).
https://doi.org/10.1109/ICALIP.2018.8455328.
Tang D, Qin B, Feng X, Liu T. Effective LSTMs for
target-dependent sentiment classification. arXiv
preprint arXiv:1512.01100. 2015 Dec 3.
Tang D, Qin B, Liu T. Document modeling with gated
recurrent neural network for sentiment classification.
InProceedings of the 2015 conference on empirical
methods in natural language processing 2015 (pp.
1422-1432).
Tang, F., Fu, L., Yao, B., &Xu, W. Aspect based fine-
grained sentiment analysis for online reviews.
Information Sciences, 488, 2019(pp.190–204).
https://doi.org/10.1016/j.ins.2019.02.064.
Tripathy, A., Agrawal, A., &Rath, S. K. Classification of
Sentimental Reviews Using Machine Learning
Techniques. Procedia Computer Science, 57, 2015
(pp.821–829).
https://doi.org/10.1016/j.procs.2015.07.523.
Tripathy, A., Agrawal, A., &Rath, S. K. Classification of
sentiment reviews using n-gram machine learning
approach. Expert Systems with Applications, 57, 2016
(pp.117–126).
https://doi.org/10.1016/j.eswa.2016.03.028.
Wan X. Using bilingual knowledge and ensemble
techniques for unsupervised Chinese sentiment
analysis. InProceedings of the conference on empirical
methods in natural language processing 2008 Oct 25
(pp. 553-561). Association for Computational
Linguistics.
Yang G, He H, Chen Q. Emotion-Semantic-Enhanced
Neural Network. IEEE/ACM Transactions on Audio,
Speech, and Language Processing. 2019
Mar;27(3):531-43.
Yang, C., Zhang, H., Jiang, B., & Li, K. Aspect-based
sentiment analysis with alternating coattention
networks. Information Processing and Management,
56(3), 2019 (pp.463–478).
https://doi.org/10.1016/j.ipm.2018.12.004.
Yenter A, Verma A. Deep CNN-LSTM with combined
kernels from multiple branches for IMDB Review
Sentiment Analysis. In2017 IEEE 8th Annual
Ubiquitous Computing, Electronics and Mobile
Communication Conference (UEMCON) 2017 Oct 19
(pp. 540-546). IEEE.
Yin W, Kann K, Yu M, Schütze H. Comparative study of
cnn and rnn for natural language processing. arXiv
preprint arXiv:1702.01923. 2017 Feb 7.
Zhang Z, Lan M. Learning sentiment-inherent word
embedding for word-level and sentence-level
sentiment analysis. In2015 International Conference
on Asian Language Processing (IALP) 2015 Oct 24
(pp. 94-97). IEEE.
Zhang, M. L., & Zhou, Z. H. ML-KNN: A lazy learning
approach to multi-label learning. Pattern Recognition,
40(7), 2007 (pp. 2038–2048).
Zhang, S., Wei, Z., Wang, Y., & Liao, T. Sentiment
analysis of Chinese micro-blog text based on extended
sentiment dictionary.Future Generation Computer
Systems, 81, 2018 (pp. 395–403).
https://doi.org/10.1016/j.future.2017.09.048.
Zharmagambetov AS, Pak AA. Sentiment analysis of a
document using deep learning approach and decision
trees. In2015 Twelve International Conference on
Electronics Computer and Computation (ICECCO)
2015 Sep 27 (pp. 1-4). IEEE.
Zhou C, Sun C, Liu Z, Lau F. A C-LSTM neural network
for text classification. arXiv preprint
arXiv:1511.08630. 2015 Nov 27.
An Efficient Approach for Sentiment Analysis using Convolutional Neural Network
21