Hybrid Deep Learning Model for Multilingual Sentiment Analysis

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 144
Hybrid Deep Learning Model for Multilingual Sentiment Analysis
Prof. Annapoorna B R, Akhil Rautela, Anurag Verma , Aditya Kumar Mishra, Dishant Kumawat
Department of Computer Science and Engineering
Dayananda Sagar College of Engineering
Shavige Malleshwara Hills, Kumaraswamy Layout, Bengaluru – 560078
---------------------------------------------------------------***----------------------------------------------------------
Abstract- The usefulness of understanding public
opinion across linguistic boundaries cannot be
overstated. Sentiment analysis is a great method for
learning more about users' opinions and their
opinions on social networks, such as movie review
sites. Natural language processing (NLP) challenges
jeopardize sentiment analysis' efficiency and accuracy.
It has been demonstrated that hybrid models can
utilize some of the advantages of classical models as
long as deep learning models are capable of
overcoming the obstacles connected with NLP. We
want to look at this hybridization in this study.
Keywords- DeepLearning, Sentiment Analysis, LSTM,
CNN
INTRODUCTION
The approach for multilingual sentiment analysis is based
on employing a sentiment dictionary to translate words for
words in any native language. A text is analyzed in three
stages: morphologically, using a sentiment dictionary to
extract verbal sentiment from each word, and finally
utilizing word sentiments to analyze the text. On tweets in
English and Hindi, we did a sentiment categorization
experiment. We used the assessment standards
"Accuracy," "Precision," "Recall," and "F1 score" to
evaluate our classifier's performance to that of different
preceding classifiers. The experimental results suggest
that our classifier may be utilized for sentiment analysis in
multilingualism since its performance is unaffected by
language variations.
Deep learning hybrid models have been suggested for
analyzing social network data for sentiment analysis. We
investigate the performance of combining SVM, CNN, and
LSTM on eight datasets of tweets and reviews using two-
word embedding approaches, Word2vec and BERT.
Following that, we contrasted four hybrid models that had
been created with single models that had been reviewed.
These tests were carried out to explore if hybrid models
could be applied to a wide range of dataset types and sizes.
The impacts of various datasets, feature extraction
methods, and deep learning models on sentiment polarity
analysis were investigated. The findings of our sentiment
polarity analysis studies demonstrate that hybrid models
beat all other models examined. Deep learning models
paired with SVM give better sentiment analysis findings
than when used alone. The dependability of hybrid models
employing SVM is greater than that of models without it in
most of the datasets evaluated; however, the
computational time for hybrid models using SVM is
substantially longer. We also discovered that the
algorithms' efficacy is heavily influenced by the datasets'
features and quality. We recognise that the context of the
dataset has a substantial impact on the sentiment analysis
algorithms we employ. We intend to examine the efficacy
of hybrid approaches for sentiment analysis on hybrid
datasets and multiple or hybrid settings in order to get a
deeper knowledge of a particular topic, such as business,
marketing, or medicine. The capacity to relate feelings to
relevant context in order to give consumers specific
individualized feedback and suggestions drives its
adoption.
We provide a multilingual sentiment analysis method that
uses a sentiment dictionary to do word-for-word
translation in any native language. The phases in this
technique are as follows: text morphological analysis,
sentiment dictionary-based word sentiment extraction,
and sentiment-based text sentiment extraction. On English
and Hindi tweets, we conduct a sentiment classification
experiment. Using the assessment standards "Accuracy,"
"Precision," "Recall," and "F1 score," we compare the
performance of our classifier in the experiment to the
performance of several preceding classifiers. The
experimental findings show that our classifier is suitable
for sentiment analysis in multilingualism since its
performance is unaffected by language variations.
Using social network data, hybrid deep learning models for
sentiment analysis were built. The performance of
integrating SVM, CNN, and LSTM with two-word
embedding methods, Word2vec and BERT, was assessed
on eight textual datasets comprising tweets and reviews.
Then we compared four hybrid models to single models
that had previously been studied. These research are being
conducted to determine if hybrid models and hybrid

techniques can adapt to a broad variety of dataset types
and sizes. Using a range of datasets, feature extraction
approaches, and deep learning models, we assessed the
reliability of sentiment polarity analysis. Combining deep
learning models with the SVM approach produces better
results than using a single model when performing
sentiment analysis. In terms of dependability, hybrid
models that use SVM outperform models that do not in the
majority of datasets tested; nevertheless, the
computational time for hybrid models that include SVM is
much longer. Furthermore, we discovered that the quality
and quantity of the datasets had a significant influence on
the performance of the algorithms. The context of the
dataset has a significant impact on sentiment analysis
methodologies. We propose to examine the usage of hybrid
methodologies for sentiment analysis on hybrid datasets
and numerous or hybrid contexts in order to gain a more
comprehensive knowledge of a particular topic. Its
usefulness arises from its capacity to relate attitudes to
relevant facts in order to deliver Clients are provided with
individualized feedback and suggestions. Sentiment
analysis is difficult to do without considering a range of
semantic and syntactic limitations as well as the
terminology of the input text. This research led to the
creation of a sentiment analysis deep learning model that
consists of one-layer CNN architecture and two layers of
LSTMs. Word embedding models can be used as the input
layer in this approach. Research indicates that the
proposed model can enhance accuracy by up to 11.6
percent and outperform existing techniques on many
benchmarks. The proposed model makes use of CNN for
feature extraction and LSTM for recurrence.
MOTIVATION AND BACKGROUND
Sentiment analysis, which aims to extract subjective
information from texts, is an important machine learning
issue. Sentiment analysis is intrinsically tied to text mining
and natural language processing. This helps you
understand the overall polarity of the review or how the
reviewer felt about a specific topic. Based on sentiment
analysis, we might be able to determine whether the
reviewer felt "glad," "sad," "angry," etc. when writing the
review. In assessing the performance of a film, movie
reviews are crucial.
Although numerical/star ratings provide a quantitative
indication of a film's success or failure, film reviews
provide a more qualitative analysis of the film. A textual
movie review educates us about the film's good and bad
points, and a more in-depth evaluation of a film review
may reveal if the picture meets the reviewer's overall
expectations. As a result of the project, sentiment analysis
will be available in a range of languages, giving it a strong
analytical tool that is not limited by linguistic limitations.
In comparison to standard models, the hybrid model
surpasses them by combining aspects from other models.
RELATED WORK
[1] This paper discusses the core of deep learning models
and associated approaches used in sentiment analysis for
social network data. Before feeding input data to a deep
learning model, the words embedding are employed. To
perform sentiment analysis, we examined the
architectures of DNN, CNN, and RNN based on word
embedding . A number of experiments have been
conducted to test DNN, CNN, and RNN models using
datasets ranging from tweets to reviews. There is a model
comparison as well as some associated field studies. This
data, together with the results of other model tests, paints
a complete picture of the usage of deep learning models for
sentiment analysis and their integration with text
preparation.
[2] LSTM should be utilized to examine stock investor
sentiments on stock swings, according to the paper. The
author defines a model with seven separate phases for
sentiment analysis. Participants are expected to perform
tasks including data collection, data cleansing, manual
sorting, feature extraction, LSTM model training,
sentiment classification, and sentiment analysis.
The positive group's accuracy is 80.53 percent, while the
negative group's accuracy is 72.66 percent, both of which
are greater than the accuracy of the Cheng and Lin model,
which used a sentiment dictionary to assess investor
sentiment and stock return in the Chinese stock market.
The paper expressly states the benefits of LSTM over a
vanilla RNN network in sentiment analysis, and it is
proposed that LSTM be utilized instead to get higher
accuracy in analysis. Because the computing cost and time
required for the analysis are not factored into the model,
the result of using LSTM for sentiment analysis is
confusing. Following the examination, DNN, CNN, and
hybrid approaches were discovered to be the most often
utilized models for sentiment polarity analysis. The study
also discovered that typical techniques like CNN, RNN, and
LSTM were separately analyzed and judged to have the
best accuracy.
[3] Based on our findings, we show that convolutional
neural networks can outperform data mining in stock
sentiment analysis. Documents are represented as bag-of-
word vectors in the standard data mining approach for
text categorization. These vectors represent the existence
of words in a text but not their order in a sentence. In

certain circumstances, it's clear that word order influences
the tone of a comment. You can utilise n-grams to
efficiently extract a document's sentiment and solve this
problem with CNN. According to the investigation in this
article examining and comparing a number of standard
Deep Learning techniques, convolutional neural networks
outperform logistic regression for sentiment analysis.
Convolutional neural networks outperform other models
in terms of accuracy. The study fails to discuss the use of a
combination of models to increase overall performance by
leveraging the characteristics of each model.
[4] As a benchmark for Sentiment Analysis on Twitter, the
authors used data from the SemEval-2013 competition.
Both terms and messages were classified in this project.
Subtask b, which consisted of classifying a message into
three categories: good, negative, and neutral, received the
most attention (message-level polarity classification).
Using lexical and machine learning approaches to evaluate
Twitter sentiment, this study examines how negation
handling impacts performance and how SWN-based
feature computation compliments it. To avoid
misclassification by the SVM classifier, we created
exceptions in cases where a negation cue is present but no
negation sense is present. The research asserts
unequivocally that hybrid deep learning models are
employed for improved analysis, and that utilizing hybrid
models considerably increases analysis accuracy.
[5] The author proposed a model that incorporated
characteristics from different social context levels in the
paper. This approach evolved from previous systems that
only categorized users at the individual level. It also used
community detection to find weak ties between users who
were not directly connected in the network. The
combination is projected to outperform at various levels of
confidence in the labels in the context and with diverse
degrees of social network sparsity. In a variety of
situations, the proposed method has been proved to work
for both types of categorisation. Several datasets are
examined in order to evaluate the model. The amount of
datasets that could be included in the evaluation was
limited due to the requirement for a social context.
Because it comprised a more densely linked network of
users, the RT Mind dataset appeared to be the most suited
of the three. When CRANK was compared to other baseline
models in that dataset, the findings were mixed. The model
was created by combining data from many levels of social
settings, and it was not confined to user-level
categorization for community identification, and it also
discovered weak relationships between users. There is no
comparison between various community models and
various graph traversal strategies, leaving the topic of
model efficacy with other community techniques wide
open and ambiguous.
[6] This article's authors describe numerous deep learning
models and suggest that hybrid models are the best option
for sentiment analysis. Among the hybrid models, the
authors highlight the CNN-LSTM hybrid.
CNN can assist you in determining how to extract data
characteristics. Long-term interactions, on the other hand,
necessitate the inclusion of more convolution layers, and
capturing dependencies gets more challenging as the
duration of an input sequence in a neural network
increases. The convolutional neural network layer is quite
deep in practice. Associations between word sequences
can be represented using LSTM models.
A CNN-LSTM hybrid model for sentiment analysis is
presented in the study. The suggested hybrid CNN-LSTM
model outperformed CNN and LSTM models
independently in two benchmark movie review datasets.
By 91 percent, the suggested Hybrid CNN-LSTM model
outperformed traditional machine learning and deep
learning models. The study examines a number of
individual and hybrid sentiment analysis models, such as
CNN-LSTM, which are projected to be more accurate than
their rivals.
[7] Based on a one-layer CNN architecture and two LSTM
layers, the study developed an Arabic sentiment analysis
deep learning model. Based on this architecture, FastText
supports word embedding. The results of a multi-domain
dataset show that it is quite successful, with scores of
89.10 percent precision, 92.14 percent recall, 92.44
percent F1-Score, and 90.75 percent accuracy,
respectively. The effect of word embedding methodologies
on Arabic sentiment categorization was comprehensively
explored in this work, revealing that the FastText model is
a more acceptable choice for extracting semantic and
syntactic information. The performance of the proposed
model is also evaluated using NB and KNN classifiers. SVM
is the top performing classifier, according to the statistics,
with an accuracy improvement of up to 3.92 percent. Due
to the effectiveness of CNNs in feature extraction and
LSTM's recurrent nature, the proposed model
outperformed state-of-the-art approaches on a number of
benchmarks with a +11.6 percent accuracy increase. The
study specifies explicitly how accurate and effective the
hybrid deep learning model is with layers and compact
architecture. Multidomain datasets make the model more
adaptable and accurate.

[8] The author's major focus in this study is on
comprehending the text's mixed linguistic character, which
includes both formal and informal textual material.
Because user-generated material frequently incorporates
localized slangs as a means of expressing genuine
emotions. The author employs a variety of methods to
examine the data's subjectivity and polarity. This study
considered not just formal languages, but also informal
and limited-resource languages. The author also suggested
using a hybrid architecture to create sentiment analysis
resources. The final methodologies applied in the different
components, as well as the quality of resources accessible,
are critical to the correctness of the suggested framework.
[9] Bing, Google, and Moses are three distinct machine
translation (MT) systems used by this author. The authors
discovered that SMT systems can collect training data for
languages other than English, and sentiment analysis
systems can perform similarly to English. Here, the author
carried out a number of tests. In the first experiment, the
data is translated from English to three other languages:
German, Spanish, and French, and the model is trained
separately for each language. In the second experiment,
the author combined all of the translations of the training
data obtained for the same language using the three
different MT systems. SMO, AdaBoost M2, and Bagging
classifiers are used by the author. However, combining all
of the translated training data significantly raises the
training data's noise level.
[10] The author tries to do sentiment analysis on
languages with the greatest datasets, such as English, and
then reuse the model for languages with low resources.
The authors use RNN to build a sentiment analysis model
based on English reviews. The robust technique of utilizing
a single model trained on English reviews beats the
baseline considerably in various languages. As a result, it
can deal with a variety of languages. The author created a
sentiment analysis model tailored to a particular domain
just for this article.
[11] The author of this article examines the feasibility of
building sentiment detection and classification models in
languages with fewer/no resources than English,
emphasizing the importance of translation quality on
sentiment classification performance, using machine
translation and supervised methods. In the training phase,
the author could see that improper translations result in
an increase in features, sparseness, and greater difficulty
in recognising a hyperplane that separates the positive and
negative instances. The extracted features are
insufficiently informative for the classifier to learn when
the translation quality is poor. The results of the testing
can be used to assess the quality of the translation. In this
case, poor translation quality causes a decline in
performance.
[12] The author introduces a unique hybrid deep learning
model in this research that purposefully integrates
multiple word embedding strategies (Word2Vec, FastText,
character-level embedding) with various deep learning
methods (LSTM, GRU, BiLSTM, CNN). The proposed model
incorporates information from several deep learning word
embedding algorithms and classifies texts based on their
emotional content. Several deep learning models known as
basic models were also built in order to conduct a series of
experiments to evaluate the proposed model's
performance. When comparing the suggested model's
performance to that of previous research, it is clear that
the new model outperforms the others. To test the
suggested model's performance. The author carried out
two separate trials. In the first experiment, twelve basic
deep learning models were created. There are four types of
deep learning. Twelve basic deep learning models were
built in the first experiment. Four deep learning models
(CNN, LSTM, BiLSTMGRU) were combined with three
distinct text representation methods (Word2Vec, FastText,
and character-level embedding). FastText and Word2Vec
embedding, both word representation approaches,
performed better with RNNs models in the trial. The
combination of BiLSTM and FastText in a word embedding
technique yielded the highest classification accuracy of
80.44 percent. With a mix of CNN and character-level
representation, accuracy was 75.67 percent. With our
suggested combination of CNN and BiLSTM with fastText
and character embedding, we achieved an accuracy of
82.14 percent classification success. As a consequence, the
suggested model's performance is superior to that of
existing basic models. The suggested model's performance
was compared to that of a prior research on the same
dataset in the second experiment. The M-Hybrid author
obtained 82.14 percent, compared to 69.25 percent in
earlier research on the dataset. The author categorized the
dataset used in this work with various significant deep
learning algorithms in order to confirm the correctness of
our model. In light of the presented strategy, the author
suggests combining several text representation methods
for a higher classification accuracy rate.
[13] CNN and LSTM are used in this study to build
information channels for Vietnamese sentiment analysis.
This scenario gives an innovative and efficient method for
combining the benefits of CNN with LSTM. The author also
assessed their method on the corpus and the VLSP corpus.
On the two datasets, the suggested model outperforms
SVM, LSTM, and CNN, according to the experimental
findings. The author also gathered 17,500 reviews from
Vietnamese social media platforms and labeled them as

good, neutral, or negative. This method combines the
benefits of CNN and LSTM into a single model. In
comparison to other models, which produces the best
results? There is room for improvement in performance on
unambiguous scenarios as well as cases with both positive
and negative attitudes.
[14] The purpose of this research is to develop a semi-
automated sentiment analysis learning system that can be
changed in response to language changes in order to
remain current. This is a hybrid strategy that employs both
lexicon-based and machine learning methods to detect the
polarity of tweets. Several datasets were chosen to put the
suggested approach to the test. The accuracy for a 3-class
classification challenge was 73.67 percent, while the
accuracy for a 2-class classification problem was 83.73
percent. The semi-automated learning component was
proven to improve accuracy by 17.55 percent. The author
offers a hybrid approach for Arabic Twitter sentiment
analysis that gives high accuracy and dependable
performance in a dynamic setting where language analysis
systems must intelligently cope with these ongoing
changes.
[15] The author presents a hybrid sentiment classification
model that incorporates a Manhattan LSTM (MaLSTM)
based on a recurrent neural network (RNN), also known as
long-short term memory (LSTM), and support vector
machines (SVM). The suggested technique uses LSTM to
learn the hidden representation before employing SVM to
identify attitudes. SVM-based representations of the LSTM
are used to determine attitudes from an IMDB movie
review dataset based on the learned representations. In
comparison to existing hashtag-based models, the
proposed model outperforms them. The proposed system
(MaLSTM) shows that merging Siamese LSTM with SVM
yields an outstanding text classification model. When there
are a lot of sentences, a pooling layer is created, and then
the SVM is used to categorize them. Because it can detect
hidden unit representations of sentences, the model may
be employed in real-time applications. According to the
findings, the approach is competitive with state-of-the-art
algorithms on a variety of text classification tasks.
Regularization strategies can improve classification in
neural network models with dropout.
[16] In this research, the author presents a novel hybrid
deep learning architecture for sentiment analysis in
resource-constrained languages. Convolutional Neural
Networks were also used to train sentiment embedding
vectors (CNNs). To choose the improved qualities, a multi-
objective optimization (MOO) framework is employed.
Finally, the sentiment enhanced vector acquired is utilized
to train the SVM for sentiment classification. An
assessment of the suggested approach for coarse-grained
(sentence level) and fine-grained (aspect level) sentiment
analysis on four independent Hindi datasets. He also
validated the proposed technique on two benchmark
English datasets. According to a performance evaluation,
Across all datasets, the suggested technique outperforms
existing state-of-the-art systems. When a sentence lacks a
distinct sentiment identifier, the algorithm has a more
difficult time properly predicting the sentiment.
[17] This study presents a hybrid learning system that
incorporates both deep and shallow learning aspects. Not
only can a hybrid technique classify single language text
sentiment, but it can also classify bilingual text sentiment.
Recurrent neural networks with long short term memory
(RNNs with LSTM), Naive Bayes Support Vector Machine
(NB-SVM), word vectors, and bag-of-words are among the
models investigated. Experiments demonstrate that
accuracy may reach 89 percent, and that the hybrid
technique outperforms any other method alone. The
hybrid technique for multi-language text sentiment
classification is a novelty in this work since it includes both
generative and discriminative models.
[18] The author suggests utilizing convolutional neural
networks to classify sentiments. Experiments with three
well-known datasets show that employing consecutive
convolutional layers for relatively long texts works well,
and that our network beats state-of-the-art deep learning
models. The author showed in his research that using
many convolutional layers in a row increased performance
on reasonably long texts. The suggested CNN models
obtained around 81 percent and 68 percent accuracy for
binary and ternary classification, respectively. Despite the
lengthy content, the convolutional layers helped to
improve performance. The suggested CNN models had a
ternary classification accuracy of 68 percent, which isn't
extremely good.
[19] The author of this study discusses and contrasts
previous work in the subject of multilingual sentiment
analysis. The goal is to see if the approaches allow for
precise implementation and reliable replication of the
claimed result, and whether the precision seen by the
author is less than that defined in the original
methodology publications. Author used several ways using
SVM classifierI, such as lexicon-based, corpus-based,
hybrid, supervised learning, and so on, and compared the
reported accuracy with the accuracy that author obtained.
The paper offers insights into the authors' various
methodologies as well as his research, which paints a clear
picture of the accuracies of various approaches.

[20] The author of this study mixes many models into one.
We've observed that the hybrid method beats all other
models, whether single or multiple languages, or both, and
to evaluate each model's contribution, we combine two
models at a time to build the hybrid model. Time.
Experiments have shown that each model works in a
distinct way. Although different languages are handled
differently, they all contribute to the overall image. This
model is a mix of the two. As the penalty coefficient drops,
over-fitting becomes more prevalent and dangerous. The
accuracy of the models is affected by many iterations. As
the number of iterations grows, the hybrid model that
matches the training data gets increasingly accurate. The
accuracy of training data, on the other hand, significantly
improves. The precision of test results is deteriorating.
CONCLUSION
We conclude that using TF-IDF rather than combining
deep learning with word embedding is optimal. Moreover,
CNN provides a good balance between accuracy and
processing time, compared to other models. The reliability
of RNN is higher than that of CNN on most datasets, but its
computation time is much longer. At long last, a decision
was reached. The study found that the algorithm's
effectiveness is highly reliant on the input data. The
datasets' characteristics allow testing deep learning
algorithms with bigger datasets simpler. When compared
to solo CNN and LSTM models in two benchmark movie
review datasets, the suggested Hybrid CNN-LSTM model
performed surprisingly well.
REFERENCES
[1]N. C. Dang, M. N. Moreno-García, and F. De la Prieta,
“Sentiment analysis based on deep learning: a comparative
study,” Electronics, vol. 9, no. 3, p. 483, 2020.
[2]G. Wang, G. Yu, and X. Shen, “The effect of online
investor sentiment on stock movements: an LSTM
approach,” Complexity, vol. 2020, Article ID 4754025, 11
pages, 2020.
[3] S. Sohangir, D. Wang, A. Pomeranets, and T. M.
Khoshgoftaar, “Big data: deep learning for financial
sentiment analysis,” Journal of Big Data, vol. 5, no. 1, p. 3,
2018.
[4]I.Gupta and N. Joshi, “Enhanced twitter sentiment
analysis using hybrid approach and by accounting local
contextual semantics,” Journal of Intelligent Systems, vol.
29, no. 1, pp. 1611–1625, 2019.
[5]J. F. Sánchez-Rada and C. A. Iglesias, “CRANK: a hybrid
model for user and content sentiment classification using
social context and community detection,” Applied Sciences,
vol. 10, no. 5, p. 1662, 2020.
[6]A. U. Rehman, A. K. Malik, B. Raza, and W. Ali, “A hybrid
CNN-LSTM model for improving accuracy of movie
reviews sentiment analysis,” Multimedia Tools and
Applications, vol. 78, no. 18, pp. 26597–26613, 2019.
[7]A. H. Ombabi, W. Ouarda, and A. M. Alimi, “Mining, “deep
learning CNN–LSTM framework for Arabic sentiment
analysis using textual information shared in social
networks,” Social Network Analysis and Mining, vol. 10, no.
1, pp. 1–13, 2020.
[8] S. L. Lo, E. Cambria, and R. Chiong, “Multilingual
sentiment analysis: from formal to informal and scarce
resource languages,” Artificial Intelligence Review 48(4),
pp. 499-527, 2017.
[9] A. Balahur and M. Turchi, “Multilingual Sentiment
Analysis using Machine Translation?,” Proceedings of the
3rd Workshop in Computational Approaches to
Subjectivity and Sentiment Analysis, pp. 52-60, 2012.
[10] E. F. Can, A. Ezen-Can, and F. Can, “Multilingual
Sentiment Analysis: An RNN-Based Framework for Limited
Data,” arXiv preprint arXiv:1806.04511, 2018
[11]A. Balahur and M. Turchi, “Comparative experiments
using supervised learning and translation for multilingual
sentiment analysis,” Computer Speech and Language, vol.
28, no. 1, January 2014.
[12] M. U. Salur and I. Aydin, “A novel hybrid deep learning
model for sentiment classification,” IEEE Access, vol. 8, pp.
58080–58093, 2020.
[13] Q.-H. Vo, H.-T. Nguyen, B. Le, and M.-L. Nguyen, “Multi-
channel LSTM-CNN model for Vietnamese sentiment
analysis,” in 2017 9th international conference on
knowledge and systems engineering (KSE), IEEE, Hue,
Vietnam, October 2017
[14] K. Elshakankery and M. F. Ahmed, “HILATSA: a hybrid
Incremental learning approach for Arabic tweets
sentiment analysis,” Egyptian Informatics Journal, vol. 20,
no. 3, pp. 163–171, 2019.
[15] H. Srinidhi, G. Siddesh, and K. Srinivasa, “A hybrid
model using MaLSTM based on recurrent neural networks
with support vector machines for sentiment analysis,”

Engineering and Applied Science Research, vol. 47, no. 3,
pp. 232–240, 2020.
[16] M. S. Akhtar, A. Kumar, A. Ekbal, and P. Bhattacharyya,
“A hybrid deep learning architecture for sentiment
analysis,” in Proceedings of COLING 2016, the 26th
International Conference on Computational Linguistics:
Technical Papers, Osaka, Japan, December 2016.
[17] G. Liu, X. Xu, B. Deng, S. Chen, and L. Li, “A hybrid
method for bilingual text sentiment classification based on
deep learning,” in Proceedings of the 2016 17th IEEE/ACIS
International Conference on Software Engineering,
Artificial Intelligence, Networking and Parallel/Distributed
Computing (SNPD), IEEE, Shanghai, China, May 2016.
[18] H. Kim and Y.-S. Jeong, “Sentiment classification using
convolutional neural networks,” Applied Sciences, vol. 9,
no. 11, p. 2347, 2019.
[19] K. Dashtipour et al., “Multilingual Sentiment Analysis:
State of the Art And Independent Comparison of
Techniques,” Cognitive Computation,vol. 8, pp. 757-771,
2016.
[20]Liu, Guolong; Xu, Xiaofei; Deng, Bailong; Chen, Siding;
Li, Li (2016). [IEEE 2016 17th IEEE/ACIS International
Conference on Software Engineering, Artificial Intelligence,
Networking and Parallel/Distributed Computing (SNPD) -
Shanghai, China (2016.5.30-2016.6.1)] 2016

Hybrid Deep Learning Model for Multilingual Sentiment Analysis

More Related Content

Similar to Hybrid Deep Learning Model for Multilingual Sentiment Analysis (20)

More from IRJET Journal (20)

Recently uploaded (20)

Hybrid Deep Learning Model for Multilingual Sentiment Analysis