Received 15 October 2024, accepted 3 December 2024, date of publication 9 December 2024,

date of current version 27 December 2024.

Digital Object Identifier 10.1109/ACCESS.2024.3513321

A Review on Text Sentiment Analysis With

Machine Learning and Deep Learning Techniques
YONATAN MAMANI-COAQUIRA AND EDWIN VILLANUEVA
Engineering Department, Pontificia Universidad Católica del Perú, San Miguel, Lima 15088, Peru
Corresponding author: Yonatan Mamani-Coaquira ([email protected])
This work was supported by Fondo Nacional de Desarrollo Científico, Tecnológico y de Innovación Tecnológica (Fondecyt) of Concytec,
which funded the doctoral studies.

ABSTRACT Automating sentiment analysis in texts has become an important task in recent years due
to the exponential growth of user-generated content, including comments and opinions on products and
services. This represents a valuable opportunity for businesses to glean insights into customer sentiment
and, in turn, to refine their offerings. Motivated by this, the machine learning field has witnessed a surge of
innovation, with an introduction of models and tools being introduced to streamline sentiment analysis. This
paper offers a thorough review of the recent advancements in machine learning and deep learning approaches
for text sentiment analysis. We propose a novel framework for studying these models, distinguishing them
by their structural intricacies. Additionally, we delve into the challenges, prospects, and emerging directions
in research, as illuminated by our framework. Consequently, this paper equips researchers with a detailed
panorama of the cutting-edge machine learning methodologies for dissecting text sentiment, easing the way
for future explorations in this vibrant field.

INDEX TERMS Machine learning, deep learning, word embedding, text encoding, sentiment analysis, text
classification.

I. INTRODUCTION to identify the polarity of sentiments (positive, negative or

In recent years we have seen a dizzying growth of neutral) expressed in texts [2], [3], [4]. We can also find
user-generated content on different platforms on the inter- methods focused on detecting emotions, such as joy, surprise,
net, such as social networks, e-comerce sites, e-learning anger, disgust, fear and sadness [3], [5], [6].
platforms, blogs, forums, etc.. These contents have valuable Most methods for sentiment analysis proposed in recent
information that can help companies and government make years are based on Machine Learning (ML) and Deep
better decisions and offer better products and services [1]. Learning (DL) models. A key step in the development
Much of this data contains information about opinions, of such models was the identification of relevant and
feelings, emotions and appreciations about products and predictive features, task known as feature extraction (FE).
services consumed by users and that end up influencing the The first features proposed for SA tasks were based on
behavior of other users. Bag of Word (BoW) [7], Term Frequency-Inverse Document
Motivated by the need to automate the extraction of this Frequency (TF-IDF) [8], and N-gram vectorization [9], [10].
information, the academic community has been actively These features were used in conjunction with classical ML
proposing a variety of methods and models for the identifica- algorithm for the sentiment classification, such as Naive
tion of emotions and sentiments from a variety of data types, Bayes (NB), Support Vector Machine (SVM), Logistic
such as texts, audios, images, videos, emoticons, among Regression (LR), Decision Tree (DT), Random Forest (RF),
others. For example, we can find a plethora of methods, among others [11], [12], [13], [14]. However, due to the
referred to as Sentiment Analysis (SA) methods, proposed limited representativeness of the aforementioned features,
several scholars focused their attention on proposing ways to
The associate editor coordinating the review of this manuscript and learn more representative features for SA and other textual
approving it for publication was Ikramullah Lali. analisys tasks. By instance, new ways of finding semantic
2024 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License.
VOLUME 12, 2024 For more information, see https://creativecommons.org/licenses/by/4.0/ 193115
 Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

representations of words (referred to as word embeddings) In the present study we present a comprehensive review of
were proposed, such as Word2Vec [15], Glove [16], and SA models and techniques. For this purpose, we introduce
FastText [17]. Along with these new word representations, a framework to study all these methods. This framework
new models based on deep learning architectures were tried facilitates the categorization and organization of different
for SA tasks, such as Recurrent Neural Networks (RNN) processes, techniques, and algorithms from recent research.
[18], Convolutional Neural Networks (CNN) [19], and neural The main contributions of this study are the following:
networks architectures with Attention Mechanisms (AM) • A framework that provides a useful categorization of
[20], [21], [22], [23]. This new wave of text representations machine learning and deep learning sentiment analysis
and models has shown promise in SA and related tasks. models.
The field of sentiment analysis has seen rapid growth, • A description and discussion of the combination of
with a large number of methods introduced recently, not only recent models in deep learning for sentiment polarity
to improve accuracy but also to address emerging forms of classification.
textual communication, such as jargon, hyper-abbreviated • Discussion and suggestions for future directions in
texts and short words, emoticons, etc. Due to this vibrant Sentiment Analysis.
wave of innovations, it is difficult for newcomer scholars in
the area to have a clear panorama of the methodologies, chal- II. CONCEPTUAL BACKGROUND OF SENTIMENT
lenges and research opportunities in the field. There are some ANALYSIS
reviews of the SA literature that may help new researchers Sentiment Analysis is primarily focused on determining the
in the area. Jain et al. [11] presented a systematic review on polarity or emotion that the person expresses in different data
ML techniques for consumer Sentiment Analysis, focusing sources [21], [23], [28], [29].
on sentiment classification, predictive recommendation, and
fake hotel and tourism reviews. Minaee et al. [21] conducted A. MODALITIES OF DATA FOR SENTIMENT ANALYSIS
a comprehensive review on DL models for text classification, Different modalities of data have been used to determine or
describing different hybrid and independent models used for classify the sentiments or emotions expressed by individu-
Sentiment Analysis and listing the 40 most popular datasets als [22], [29], [30], [31], such as:
in this area. Zhang et al. [1] published a survey of DL
• Text
models for SA, showing different architectures of hybridized
• Image (body gesture or facial expression)
or independent deep neural networks and sharing information
• Video (body gesture or facial expression)
from research spanning document, sentence, and aspect levels
• Audio (voice)
for opinion processing. Zarandi and Mirzaei [24] reviewed
• Emoticon
different DL, ML, and Graph Neural Network (GNN) models
for sentiment classification, including GAN models.
B. SENTIMENT ANALYSIS TASKS
While these reviews can help to better understand the
Tasks for SA can be classified into two general categories:
progress in the area, they focus on limited aspects of
opinion and emotion analysis [12], [23], [32]. However,
the sentiment analysis process, such as FE techniques or
many expressed opinions do not manifest emotion; therefore,
classification models. Notwithstanding, several steps are
emotions and opinions are not equivalent [3].
involved in the pipeline of SA model development, such as
the selection of the training dataset, the data cleaning and pre-
1) OPINION MINING
procesing steps, the FE method, the selection of performance
metrics, the training and evaluation of ML/DL models and the This field of study aims to analyze expressions manifested
model deployment and application. by people or organizations. The opinion can be about an
Additionaly, on the last few years have appeared several object, product, service, event, organization, person, etc. The
novel SA methods that have not been covered by previous polarity of opinions is usually classified into three classes:
reviews. During our literature review on sentiment analysis positive, negative, or neutral [3], [11], [12], [32]. According
studies we found that only a few studies covers novel deep to Yadollahi et al. [32], the opinion mining field studies the
learning models for SA tasks [1], [11], [25], [26], [27]. following problems:
Furthermore, we observed that the scope of the reviewed • Subjectivity detection
SA applications focuses on a few application domains, like • Opinion polarity classification
hospitals, tourism [11], and e-commerce [24]. On the other • Opinion spam detection
hand, with respect to the process of text representation and • Opinion summary
attribute extraction, most review studies analyze classical • Detection of expressed argument
methods, like TF-IDF [25], [26], Word2vec, and Glove [1],
[11]. We believe that there is a need to delve deeper into most 2) EMOTION MINING
recent text representation methods, such as BERT, Roberta, Emotion mining aims to identify the emotions expressed
FastText, etc., in applications with ML/DL models to improve in texts or other modalities [32]. Emotions are thoughts or
FE. feelings that can be subjective [3]. People have six primary

193116 VOLUME 12, 2024

Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

emotions such as joy, surprise, anger, disgust, sadness, and TABLE 1. The SR research questions and contributions.
fear; these emotions can also be divided into secondary and
tertiary and even have emotions according to the degree of
intensity [3], [5], [6]. According to Yadollahi et al. [32], the
emotion mining field addresses the following problems:
• Emotion detection
• Emotion polarity detection
• Emotion classification
• Emotion cause detection

C. LEVELS OF SENTIMENT ANALYSIS

The analysis of sentiments or opinions can be performed at
different levels, such as: document-level, sentence-level, and
aspect-level [12], [22], [30], [33], [34].

1) DOCUMENT-LEVEL
This level allows for determining the polarity of the opinion
expressed in entire documents, which can be positive,
negative or neutral. The analyzed entity is the whole content
TABLE 2. Keywords used in the search strings.
of the text. It does not allow for comparison or evaluate the
polarity of multiple entities.

2) SENTENCE-LEVEL
At this level the aim is to determine the polarity of the
sentiment expressed in the sentences of a text (which usually
are positive, negative, or neutral). Neutral opinion means
no opinion because many sentences contain information
that do not represent any sentiment. By analyzing the
sentences separately instead of the entire document, a better
understanding of the feelings expressed in the text can be
inferred. This level of study has two main subtasks, which
are 1) subjectivity classification, which determines whether
the text is subjective or objective 2) sentence-level sentiment
classification, if it is subjective, it determines the positive,
negative, or neutral polarity [35].

3) ASPECT-LEVEL
The objective of this level is to determine the sentiment
or polarity about aspects contained in a text. A text can
have several aspects about which feelings are expressed. For later. We followed the review workflow shown in Figure 1,
example, in the following text: ‘‘the new iPhone 5’s screen based on the Preferred Reporting Items for Systematic
size is amazing, but its battery life is short’’, we can see that Review and Meta-Analysis (PRISMA) guidelines [37].
there are opinions expressed on two aspects, for the iPhone
5 screen it is a positive opinion, but for the battery it is a A. QUESTIONS AND GOALS
negative opinion [29]. Therefore, this level has two main The present review aims to find studies based on ML, DL and
subtasks: 1) Aspect-extraction, which aims to identify the word embedding models used or proposed for sentiment
aspects mentioned in the text, and 2) Aspect-level Sentiment classification tasks. For this review we pose four research
Analysis (ALSA), which aims to determine the polarity of the questions (RQ), presented in Table 1.
sentiment associated with the aspect [33], [36].
B. REVIEW PROTOCOL
III. REVIEW METHODOLOGY The search of articles was carried out using Web of Science
We performed a systematic review (SR) of the scientific (WoS) database. This indexes high quality journals and
literature to identify relevant research works related to Senti- conferences. We follow the PRISMA guidelines to perform
ment Analysis. The results of this review led us to propose a the review. This involved four steps: identification, screening,
Framework to Study Sentiment Analysis methods, described eligibility and Inclusion steps (Figure 2).

VOLUME 12, 2024 193117

 Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

FIGURE 1. Workflow for systematic review.

TABLE 3. Search strings.

FIGURE 2. PRISMA methodology.

a total of 2,810 articles. Figure 3 shows the distribution of the

articles for each search string without using filters.
In the Screening step we applied the inclusion and
FIGURE 3. Number of research articles found.
exclusion criteria of Table 4. These criteria are related to
relevance, recentness (year of publication) and knowledge
area (computer science, technology or linguistics). With this
filtering, it was reduced to 1026 articles.
For the Identification step we employed four search strings In the eligibility phase we read the summary of the articles
(Table 3) in the WoS database, one for each research question. to identify those that are truly relevant to the present study and
Table 2 presents the keywords used. In this stage we identified that allow us to meet our review objectives. We identified a

193118 VOLUME 12, 2024

Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

TABLE 4. Inclusion & exclusion criteria.

FIGURE 4. The percentage of research articles by the search string.

TABLE 5. Most used dataset for sentiment analysis.

A. DATA COLLECTION
In the initial phase of the Framework’s data collection,
several datasets were identified for sentiment classification,
total of 217 relevant articles. Figure 4 shows the distribution as detailed in Table 5. The majority of research articles
of the selected articles along the search strings. String utilized datasets from tweet extraction, along with IMDB and
2 resulted in 48.8% of the total relevant articles, followed by Amazon datasets, to determine positive and negative polarity.
String 1 with 30.0%. This suggests that most recent articles Figure 6 shows the representativeness of languages in the
on sentiments analysis are based on deep learning models. datasets used in various sentiment classification models. It is
In addition, the articles we found also demonstrate the use evident that English is the most representative language
of pre-trained word embedding models to enhance sentiment compared to other languages for training and validating
polarity classification. different models.

IV. RESULTS B. DATA CLEANING

This section elaborates on our findings and framework During this phase, various tasks are carried out to clean
proposal using the 217 studies found in section III. Below, the data. However, the specific cleaning tasks to be used
we outline the seven phases of the proposal (see Figure 5) depend on the type of analysis and the dataset used for model
which are based on the results of the systematic review and development. We identified 16 common cleaning techniques
address the research questions. during the review, listed in Table 6.

VOLUME 12, 2024 193119

 Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

FIGURE 5. Framework proposal for sentiment analysis.

TABLE 6. Data clean techniques for sentiment classification.

FIGURE 6. Representativeness of languages in common datasets for

sentiment analysis.

TABLE 7. Preprocessing techniques.

techniques: stemming, lemmatization, and tokenization (see
Table 7). The choice of the technique depend on the objectives
defined for developing the model.

D. TEXT ENCODING
Text encoding, is an important step in sentiment classifica-
tion. It aims to find numerical representations of words and
texts. It can be considered as a form of feature extraction for
the subsequent phase of model construction. Table 8 shows
the most common text encoding techniques we found in
papers proposing non-neural models for sentiment classifi-
cation. It can be seen that TF-IDF and BoW are the most
frequently used text encoding techniques in such approaches.
In papers that propose neural-network based models, the
text encoding techniques are usually different. We list such
techniques in Table 9. It can be seen that Word2vec is the
most common text-encoding technique in neural-network
C. PREPROCESSING based models. It is also commonly used Glove, FastText,
This phase aims to prepare the data for efficient fea- and BERT techniques as word embedding techniques. Most
ture extraction. We identified three frequent pre-procesing authors indicate that such encodings contribute to improve

193120 VOLUME 12, 2024

Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

TABLE 8. Common text encoding techniques found in non-neural models TABLE 9. Common text encoding techniques found in neural models for
for sentiment classification. sentiment classification.

the performance of sentiment polarity classifiers due to their

rich numerical representations of words, that are obtained by
pre-training the encoders with vast amounts of data.

E. MACHINE LEARNING TECHNIQUES

Machine learning techniques are at the heart of the process of
building sentiment detection models. In the present review we
group these techniques into those that are not based on neural
networks and those that are. Table 10 lists the non-neural
network-based techniques, with the most popular techniques
being NB, SVM, LR, and ensemble mmodels (RF, Bagging,
Voting, Stacking). Table 11 shows neural-networks based
techniques. The most common neural network architectures
are CNN, Long Short Term Memory (LSTM) networks,
Gated Recurrent Unit (GRU) networks, and AM networks.
These architectures are used independently or hybridized
with other neural or non-neural models. The techniques
with the greatest performance gains are usually hybrid
models.

F. POLARITY CLASSIFICATION GRANULARITY

The target categories in sentiment polarity classification
is normally determined in the definition of the model
development project, since it defines the data labeling
requirements for the training and validation phase. Most
polarity classification studies target two clases: positive and
negative sentiments. It is also common to find three-polarity
classification models: positive, negative, and neutral senti-
ments. Figure 7 shows the relative proportion of the studies FIGURE 7. Percentage of sentiment polarity most used for sentiment
for each classification granularity. We can observe that there classification.

are a few works targeting five polarities: higly positive,

positive, neutral, negative, and higly negative classes. This
small number of studies is probably due to the small G. EVALUATION METRICS
number of available datasets that have more than 3 labeled Table 12 shows the evaluation metrics we found in the
classes. reviewed studies. The most commonly used metrics are

VOLUME 12, 2024 193121

 Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

TABLE 10. Machine learning techniques not based on neural networks TABLE 11. Machine learning techniques based on neural networks for
for sentiment classifier induction. sentiment classifier induction.

Accuracy, Precision, Recall, and F1-score. However, other

evaluation metrics have been used for model evaluation, like
AUC, Kappaca score, etc. Many studies use more than one
metric. This is because each metric evaluates or emphasizes
some aspect of the results, so using multiple metrics generates
a broader view of the model’s capabilities.

V. DISCUSSION
In this study we describe a framework for studying articles
that propose sentiment classification models. We discuss here
some relevant aspects that we found throughout our review.
1) Sentiment Analysis: The primary tasks in Sentiment
Analysis involves opinion and emotion mining. In our
literature review, the most commonly used levels of
Sentiment Analysis are sentence and document. How-
ever, a sentence can contain multiple sentiments, so it
is important to identify and analyze the sentence’s final

193122 VOLUME 12, 2024

Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

TABLE 12. Evaluation metrics most commonly used. This demonstrates that learning good text encodings
is the key to improving sentiment identification.
They are some interesting studies that use knowledge
graphs in the process of text encoding [162], [226]
showing promising results. Probably, the incorporation
of semantic knowledge of words through knowledge
graphs helps to have better numerical representations
of texts.
4) Machine and Deep Learning Models: Recent stud-
ies have proposed several hybrid methods, both
neural network-based and non-neural network-based.
classification; in any case, it’s very complex to identify In Table 10 we list non-neural network-based. It is
polarity because the text could have multiple opinions. evident that NB, SVM, RF, LR, and DT are the most
Therefore, recent deep learning models focuses on the commonly used. However, studies utilizing hybrid
aspect level analysis (see [34], [119], [147], [187], models such as Bagging, Voting, and Stacking have
[191], [207], [211], [215], [239]). In those works, the shown promising results for sentiment classification
sentiment of each entity is determined. This approach tasks in some cases. On the other hand, models
may be more beneficial than document-level analysis based on neural networks have demonstrated better
since since it can show how the product or service is results than traditional algorithms like NB, SVM, and
appreciated at different aspects. Most reviewed studies RF. Table 11 shows the use of Recurrent Neural
are typically evaluated in the SemEval dataset (see Networks (RNN) models such as LSTM, CNN, Bidi-
Table 5). rectional Long short-term Memory (BiLSTM), GRU,
2) Preprocessing: In the present review we have iden- and Bidirectional Gated Recurrent Unit (BiGRU) in
tified preprocessing techniques that have shown their numerous applications for Sentiment Analysis due
relevance in recent studies. Tokenization is essential to their ability to identify complex patterns within
for extracting the best features, while Lemmatization text more effectively. Furthermore, recent studies have
and Stemming techniques contribute to improving shown that hybrid methods using CNN, RNN, AM, and
classification results. It’s important to note that these traditional ML algorithms have produced better results
techniques are not universally implemented in all than standalone models.
languages and datasets. It is important to analyze and
identify the appropriate library to use based on the VI. CHALLENGES AND OPPORTUNITIES
dataset characteristics. While some studies use these Sentiment Analysis has recently made significant progress in
techniques in the cleaning phase, we have observed psychology, politics, marketing, economics, and education.
that Stemming is more commonly used compared to From the reviewed studies, we identified several novel ideas.
Lemmatization (see Table 7). We suggest using both However, there are some challenges and gaps that need to be
techniques in the preprocessing phase to maintain an addressed in future researches. Below, we summarize some
organized workflow. challenges that could that could help the progress of the
3) Text encoding: The performance of a sentiment Sentiment Analysis field.
classification model depends depends heavily on the
choice of text encoding. For ML models not based
on neural networks, the TF-IDF and BoW techniques A. REDUCTION OF COMPUTATIONAL COST
have been the most popularly used coding techniques Most models require high memory capacities for training
(see Table 8). However, recent studies have shown that and inference. Some techniques, like model distillation and
Word2Vec and Glove techniques have demonstrated model quantization can help reducing the number of trainable
promising results in non neural-networks models. parameters, which in turn would help to accelerate the model
On the other hand, in recent years there has been training and improve model performance.
an explosion of methods based on artificial neural
networks. In the field of sentiment classification, these B. NEW DATASETS
methods have frequently shown superior predictive In our review, we found that most studies use a few datasets to
performances than traditional ML methods. This is train and evaluate their proposals (see Table 5). In 66 studies
probably due to their non-linear characteristics and we found that the dataset are opinions extracted from Twitter
ability to digest large amounts of data through the social network; 32 studies use IMDB movie reviews dataset,
fitting of millions of parameters. It is common that 28 studies use Amazon reviews dataset, and 17 studies use
recent sentiment classification models also use neural Sentiment140 dataset. This shows the need to collect new
network-based text encoders in their pipeline, like datasets with more records, more diversity of fields, more
Glove, Word2Vec, FastText, and BERT (Table 9). polarity/sentiment classes and more languages. There is also
VOLUME 12, 2024 193123
 Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

a need to build datasets oriented toward specific domains that based on neural-network word embeddings and deep learning
are not represented in existing datasets. classification methods yield state of the art results. This
review can help new researchers in this area to have a
C. ASPECT-LEVEL SENTIMENT CLASSIFICATION panoramic view of the area, as well as to identify challenges,
Most research has focused on the sentence and document current gaps and new research opportunities.
level due to the availability of datasets such as IMDB,
Amazon reviews, Twitter US airline, Sentiment140, and Yelp ACKNOWLEDGMENT
Sentiment dataset. However, the aspect level is a relatively The authors would like to express their gratitude to the Arti-
new and understudied area. For instance, Table 5 shows ficial Intelligence Laboratory, Pontifical Catholic University
that SemEval is the only dataset related to aspect labels. of Peru, for providing them with access to their facilities and
Therefore, there is a need for new studies using datasets servers.
and models related to aspect-based and domain-specific
sentiment analysis. REFERENCES
[1] L. Zhang, S. Wang, and B. Liu, ‘‘Deep learning for sentiment analysis: A
D. MODELS BASED ON CONCEPT UNDERSTANDING survey,’’ WIREs Data Mining Knowl. Discovery, vol. 8, no. 4, p. e1253,
Sophisticated text encoders, such as Word2Vec (see Table 9) Mar. 2018.
can help develop better sentiment classification models. [2] T. Nasukawa and J. Yi, ‘‘Sentiment analysis: Capturing favorability using
natural language processing,’’ in Proc. 2nd Int. Conf. Knowl. Capture,
These techniques consider the semantics, syntax, mor- New York, NY, USA, 2003, pp. 70–77.
phology, and context of the text. However, in order to [3] B. Liu, Sentiment Analysis and Opinion Mining, 1st ed. USA: Morgan &
better comprehend the entire meaning of a verbal language Claypool, 2012, doi: 10.1007/978-3-031-02145-9.
sentence, it is necessary to address the conceptual level. This [4] R. Tokuhisa, K. Inui, and Y. Matsumoto, ‘‘Emotion classification
using massive examples extracted from the web,’’ in Proc. 22nd Int.
can allow us to handle words that may not explicitly convey Conf. Comput. Linguistics (COLING), vol. 1, Manchester, U.K., 2008,
an emotion but are associated with concepts that do. pp. 881–888.
Deep learning models with CNN, RNN, and AM have [5] P. Shaver, J. Schwartz, D. Kirson, and C. O’Connor, ‘‘Emotion
knowledge: Further exploration of a prototype approach,’’ J. Personality
shown promising results for Sentiment Analysis. However, Social Psychol., vol. 52, pp. 1061–1086, Jun. 1987.
there is a need to develop novel architectures with more [6] W. G. Parrott, Emotions in Social Psychology: Essential Readings.
advanced linguistic, lexical, semantic, and morphologi- Philadelphia, PA, USA: Psychology Press, 2001.
cal features and concept understanding capabilities. Some [7] Z. S. Harris, ‘‘Distributional structure,’’ WORD, vol. 10, nos. 2–3,
pp. 146–162, Aug. 1954.
promising studies in this line are those that try to incorporate [8] K. S. Jones, ‘‘A statistical interpretation of term specificity and its
knowledge graphs [162], [226]. Using word concepts is application in retrieval,’’ J. Documentation, vol. 28, pp. 11–21, Jan. 1972.
a promising idea because a word may have significant [9] C. E. Shannon, ‘‘A mathematical theory of communication,’’ Bell Syst.
conceptual relations with others through edges, with the Tech. J., vol. 27, no. 3, pp. 379–423, Jul. 1948.
[10] A. A. Markov, ‘‘An example of statistical investigation of the text Eugene
nodes being words. Onegin concerning the connection of samples in chains,’’ Sci. Context,
vol. 19, no. 4, pp. 591–600, Dec. 2006.
E. NEW DEEP LEARNING ARCHITECTURES [11] P. K. Jain, R. Pamula, and G. Srivastava, ‘‘A systematic literature review
on machine learning applications for consumer sentiment analysis using
DL models have demonstrated promising results for Senti- online reviews,’’ Comput. Sci. Rev., vol. 41, Aug. 2021, Art. no. 100413.
ment Analysis. However, these models require assistance in [12] K. Cortis and B. Davis, ‘‘Over a decade of social opinion mining: A
comprehending the symbols and words within the text. There systematic review,’’ Artif. Intell. Rev., vol. 54, no. 7, pp. 4873–4965,
is a necessity to design architectures based on deep neural Oct. 2021.
[13] F. Hemmatian and M. K. Sohrabi, ‘‘A survey on classification techniques
networks or apply graph knowledge with linguistic, lexical, for opinion mining and sentiment analysis,’’ Artif. Intell. Rev., vol. 52,
semantic, and morphological features or knowledge in order no. 3, pp. 1495–1545, Oct. 2019.
to fully understand the text. [14] S. Osorio Angel, A. Peña Pérez Negrón, and A. Espinoza-Valdez,
‘‘Systematic literature review of sentiment analysis in the Span-
ish language,’’ Data Technol. Appl., vol. 55, no. 4, pp. 461–479,
VII. CONCLUSION Aug. 2021.
This article identified more than a thousand scientific [15] T. Mikolov, K. Chen, G. Corrado, and J. Dean, ‘‘Efficient estimation of
articles on machine learning (ML) models for sentiment word representations in vector space,’’ 2013, arXiv:1301.3781.
classification. After thoroughly reviewing and analyzing the [16] J. Pennington, R. Socher, and C. Manning, ‘‘Glove: Global vectors for
word representation,’’ in Proc. Conf. Empirical Methods Natural Lang.
articles, 217 studies were selected for this review based Process. (EMNLP), Doha, Qatar, 2014, pp. 1532–1543.
on their relevance and contribution to the field. The paper [17] P. Bojanowski, E. Grave, A. Joulin, and T. Mikolov, ‘‘Enriching
proposes a framework to study and disecting Sentiment word vectors with subword information,’’ Trans. Assoc. Comput.
Linguistics, vol. 5, pp. 135–146, Jun. 2017. [Online]. Available: https://
Analysis methods. The framework comprises the following direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00051/1567442/tacl_
aspects: data collection, data cleaning, pre-processing, text- a_00051.pdf
encoding, ML techniques, polarity classification, and [18] J. Elman, ‘‘Finding structure in time,’’ Cognit. Sci., vol. 14, no. 2,
pp. 179–211, Jun. 1990.
evaluation metrics. In each aspect, we consider algorithms
[19] Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, ‘‘Gradient-based
and techniques that can be used according to the objective learning applied to document recognition,’’ Proc. IEEE, vol. 86, no. 11,
of the study. Recent studies have demonstrated that methods pp. 2278–2324, Nov. 1998.

193124 VOLUME 12, 2024

Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

[20] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, [44] S. Chirgaiya, D. Sukheja, N. Shrivastava, and R. Rawat, ‘‘Analysis of
L. Kaiser, and I. Polosukhin, ‘‘Attention is all you need,’’ in Proc. Adv. sentiment based movie reviews using machine learning techniques,’’
Neural Inf. Process. Syst., 2017, pp. 1–11. J. Intell. Fuzzy Syst., vol. 41, no. 5, pp. 5449–5456, Nov. 2021.
[21] S. Minaee, N. Kalchbrenner, E. Cambria, N. Nikzad, M. Chenaghlu, [45] M. Azam, T. Ahmed, R. Ahmad, A. Ur Rehman, F. Sabah, and
and J. Gao, ‘‘Deep learning–based text classification: A comprehensive R. Muhammad Asif, ‘‘A two-step approach for improving sentiment
review,’’ ACM Comput. Surv., vol. 54, p. 62, Apr. 2021. classification accuracy,’’ Intell. Autom. Soft Comput., vol. 30, no. 3,
[22] A. Yadav and D. K. Vishwakarma, ‘‘Sentiment analysis using deep pp. 853–867, 2021.
learning architectures: A review,’’ Artif. Intell. Rev., vol. 53, no. 6, [46] M. M. Ali, ‘‘Arabic sentiment analysis about online learning to mitigate
pp. 4335–4385, Aug. 2020. COVID-19,’’ J. Intell. Syst., vol. 30, no. 1, pp. 524–540, Jan. 2021.
[23] Z. Kastrati, F. Dalipi, A. S. Imran, K. Pireva Nuci, and M. A. Wani, [47] U. Younis, M. Z. Asghar, A. Khan, A. Khan, J. Iqbal, and
‘‘Sentiment analysis of Students’ feedback with NLP and deep learning: N. Jillani, ‘‘Applying machine learning techniques for performing
A systematic mapping study,’’ Appl. Sci., vol. 11, no. 9, p. 3986, comparative opinion mining,’’ Open Comput. Sci., vol. 10, no. 1,
Apr. 2021. pp. 461–477, Dec. 2020.
[24] A. K. Zarandi and S. Mirzaei, ‘‘A survey of aspect-based sentiment [48] Z. M. Zain, A. A. Alhajji, N. S. Alotaibi, N. B. Almutairi, A. D. Aldawas,
analysis classification with a focus on graph neural network methods,’’ M. M. Almazrua, and A. M. Alhammad, ‘‘Ranking beauty clinics in
Multimedia Tools Appl., vol. 83, no. 19, pp. 56619–56695, Dec. 2023. Riyadh using lexicon-based sentiment analysis and multiattribute-utility
[25] P. Nandwani and R. Verma, ‘‘A review on sentiment analysis and emotion theory,’’ Int. J. Adv. Comput. Sci. Appl., vol. 11, no. 10, pp. 1–10, 2020.
detection from text,’’ Social Netw. Anal. Mining, vol. 11, no. 1, p. 81, [49] S. Mihi, B. Ait, I. El, S. Arezki, and N. Laachfoubi, ‘‘MSTD: Moroccan
Aug. 2021. sentiment Twitter dataset,’’ Int. J. Adv. Comput. Sci. Appl., vol. 11, no. 10,
[26] D. Tiwari, B. Nagpal, B. S. Bhati, A. Mishra, and M. Kumar, ‘‘A pp. 363–372, 2020.
systematic review of social network sentiment analysis with comparative [50] F. M. Plaza-Del-Arco, M. T. Martín-Valdivia, L. A. Ureña-López, and
study of ensemble-based techniques,’’ Artif. Intell. Rev., vol. 56, no. 11, R. Mitkov, ‘‘Improved emotion recognition in Spanish social media
pp. 13407–13461, Nov. 2023. through incorporation of lexical knowledge,’’ Future Gener. Comput.
[27] A. Ullah, S. N. Khan, and N. M. Nawi, ‘‘Review on sentiment analysis Syst., vol. 110, pp. 1000–1008, Sep. 2020.
for text classification techniques from 2010 to 2021,’’ Multimedia Tools [51] M. Y. Khan and K. N. Junejo, ‘‘Exerting 2D-space of sentiment Lexicons
Appl., vol. 82, no. 6, pp. 8137–8193, Mar. 2023. with machine learning techniques: A hybrid approach for sentiment
[28] B. Liu, Sentiment Analysis: Mining Opinions, Sentiments, and Emotions. analysis,’’ Int. J. Adv. Comput. Sci. Appl., vol. 11, no. 6, pp. 1–10, 2020.
Cambridge, U.K.: Cambridge Univ. Press, 2015.
[52] L. B. Shyamasundar and P. Jhansi Rani, ‘‘A multiple-layer machine
[29] E. Cambria, B. Schuller, Y. Xia, and C. Havasi, ‘‘New avenues in learning architecture for improved accuracy in sentiment analysis,’’
opinion mining and sentiment analysis,’’ IEEE Intell. Syst., vol. 28, no. 2, Comput. J., vol. 63, no. 1, pp. 395–409, Jan. 2020.
pp. 15–21, Mar. 2013.
[53] M. Bilgin and H. Köktas, ‘‘Sentiment analysis with term weighting and
[30] M. Birjali, M. Kasri, and A. Beni-Hssane, ‘‘A comprehensive survey on
word vectors,’’ Int. Arab J. Inf. Technol., vol. 16, pp. 953–959, Jan. 2019.
sentiment analysis: Approaches, challenges and trends,’’ Knowl.-Based
[54] M. B. Alvi, N. A. Mahoto, M. A. Unar, and M. A. Shaikh, ‘‘An effective
Syst., vol. 226, Aug. 2021, Art. no. 107134.
framework for tweet level sentiment classification using recursive text
[31] Y. Wang, W. Song, W. Tao, A. Liotta, D. Yang, X. Li, S. Gao, Y.
pre-processing approach,’’ Int. J. Adv. Comput. Sci. Appl., vol. 10, no. 6,
Sun, W. Ge, W. Zhang, and W. Zhang, ‘‘A systematic review on
p. 572, 2019.
affective computing: Emotion models, databases, and recent advances,’’
Inf. Fusion, vols. 83–84, pp. 19–52, Jul. 2022. [55] D. Gamal, M. Alfonse, E.-S. M. El-Horbaty, and A.-B. M. Salem,
‘‘Analysis of machine learning algorithms for opinion mining in different
[32] A. Yadollahi, A. G. Shahraki, and O. R. Zaiane, ‘‘Current state of text
domains,’’ Mach. Learn. Knowl. Extraction, vol. 1, no. 1, pp. 224–234,
sentiment analysis from opinion to emotion mining,’’ ACM Comput.
Dec. 2018.
Surv., vol. 50, no. 2, pp. 1–33, May 2017.
[56] M. Ghosh and G. Sanyal, ‘‘Analysing sentiments based on multi feature
[33] B. Liu, Sentiment Analysis: Mining Opinions, Sentiments, and Emotions
combination with supervised learning,’’ Int. J. Data Mining, Model.
(Studies in Natural Language Processing), 2nd ed. Cambridge, U.K.:
Manage., vol. 11, no. 4, pp. 391–416, 2019.
Cambridge Univ. Press, 2020.
[34] K. W. Trisna and H. J. Jie, ‘‘Deep learning approach for aspect-based [57] M. Ghosh and G. Sanyal, ‘‘An ensemble approach to stabilize the features
sentiment classification: A comparative review,’’ Appl. Artif. Intell., for multi-domain sentiment analysis using supervised machine learning,’’
vol. 36, no. 1, Dec. 2022, Art. no. 2014186. J. Big Data, vol. 5, no. 1, p. 44, Nov. 2018.
[35] B. Liu and L. Zhang, A Survey of Opinion Mining and Sentiment Analysis. [58] S. Alowaidi, M. Saleh, and O. Abulnaja, ‘‘Semantic sentiment analysis of
Boston, MA, USA: Springer, 2012, pp. 415–463. Arabic texts,’’ Int. J. Adv. Comput. Sci. Appl., vol. 8, no. 2, pp. 1–7, 2017.
[36] S. Poria, E. Cambria, G. Winterstein, and G.-B. Huang, ‘‘Sentic patterns: [59] A. B. Nassif, A. M. Darya, and A. Elnagar, ‘‘Empirical evaluation of
Dependency-based rules for concept-level sentiment analysis,’’ Knowl.- shallow and deep learning classifiers for Arabic sentiment analysis,’’
Based Syst., vol. 69, pp. 45–63, Oct. 2014. ACM Trans. Asian Low-Resource Lang. Inf. Process., vol. 21, no. 1,
[37] M. J. Page et al., ‘‘The PRISMA 2020 statement: An updated guideline for pp. 1–25, Nov. 2021.
reporting systematic reviews,’’ BMJ, vol. 372, pp. 790–799, Mar. 2021. [60] S. Ranjan and S. Mishra, ‘‘Perceiving university students’ opinions from
[38] T. Nijhawan, G. Attigeri, and T. Ananthakrishna, ‘‘Stress detection Google app reviews,’’ Concurrency Comput., Pract. Exper., vol. 34,
using natural language processing and machine learning over social no. 10, p. e6800, May 2022.
interactions,’’ J. Big Data, vol. 9, no. 1, p. 33, Mar. 2022. [61] A. K. Durairaj and A. Chinnalagu, ‘‘Transformer based contextual model
[39] A. Chinnalagu and A. K. Durairaj, ‘‘Context-based sentiment analysis on for sentiment analysis of customer reviews: A fine-tuned BERT,’’ Int.
customer reviews using machine learning linear models,’’ PeerJ Comput. J. Adv. Comput. Sci. Appl., vol. 12, no. 11, pp. 1–7, 2021.
Sci., vol. 7, p. e813, Dec. 2021. [62] S. M. Mohamed, E. H. Mohamed, and M. A. Belal, ‘‘Polarity detection
[40] A. Khattak, M. Z. Asghar, Z. Ishaq, W. H. Bangyal, and I. A. Hameed, of dialectal Arabic using deep learning models,’’ Int. J. Adv. Comput. Sci.
‘‘Enhanced concept-level sentiment analysis system with expanded Appl., vol. 12, no. 11, pp. 212–218, 2021.
ontological relations for efficient classification of user reviews,’’ Egyptian [63] T. E. Trueman and E. Cambria, ‘‘A convolutional stacked bidirectional
Informat. J., vol. 22, no. 4, pp. 455–471, Dec. 2021. LSTM with a multiplicative attention mechanism for aspect category and
[41] A. H. Hussein, I. F. Moawad, and R. M. Badry, ‘‘Arabic sentiment analysis sentiment detection,’’ Cognit. Comput., vol. 13, no. 6, pp. 1423–1432,
for multi-dialect text using machine learning techniques,’’ Int. J. Adv. Nov. 2021.
Comput. Sci. Appl., vol. 12, no. 12, pp. 693–700, 2021. [64] P. K. Jain, V. Saravanan, and R. Pamula, ‘‘A hybrid CNN-LSTM: A
[42] H. Ali, H. Farman, H. Yar, Z. Khan, S. Habib, and A. Ammar, ‘‘Deep deep learning approach for consumer sentiment analysis using qualitative
learning-based election results prediction using Twitter activity,’’ Soft user-generated contents,’’ ACM Trans. Asian Low-Resource Lang. Inf.
Comput., vol. 26, no. 16, pp. 7535–7543, Aug. 2022. Process., vol. 20, no. 5, pp. 1–15, Jul. 2021.
[43] B. Gaye, D. Zhang, and A. Wulamu, ‘‘Sentiment classification for [65] S. Sagnika, B. S. P. Mishra, and S. K. Meher, ‘‘An attention-based
employees reviews using regression vector- stochastic gradient descent CNN-LSTM model for subjectivity detection in opinion-mining,’’ Neural
classifier (RV-SGDC),’’ PeerJ Comput. Sci., vol. 7, p. e712, Sep. 2021. Comput. Appl., vol. 33, no. 24, pp. 17425–17438, Dec. 2021.

VOLUME 12, 2024 193125

 Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

[66] N. Jamal, C. Xianqiao, F. Al-Turjman, and F. Ullah, ‘‘A deep learning– [87] Y. Didi, A. Walha, and A. Wali, ‘‘COVID-19 tweets classification based
based approach for emotions classification in big corpus of imbalanced on a hybrid word embedding method,’’ Big Data Cognit. Comput., vol. 6,
tweets,’’ ACM Trans. Asian Low-Resource Lang. Inf. Process., vol. 20, no. 2, p. 58, May 2022.
no. 3, pp. 1–16, Mar. 2021. [88] O. Iparraguirre-Villanueva, V. Guevara-Ponce, F. Sierra-Linan,
[67] A. R. Pathak, M. Pandey, and S. Rautaray, ‘‘Topic-level sentiment analysis S. Beltozar-Clemente, and M. Cabanillas-Carbonell, ‘‘Sentiment
of social media data using deep learning,’’ Appl. Soft Comput., vol. 108, analysis of tweets using unsupervised learning techniques and the
Sep. 2021, Art. no. 107440. k-means algorithm,’’ Int. J. Adv. Comput. Sci. Appl., vol. 13, no. 6,
[68] Ö. Çoban, S. A. Özel, and A. Inan, ‘‘Deep learning-based sentiment pp. 571–578, 2022.
analysis of Facebook data: The case of Turkish users,’’ Comput. J., vol. 64, [89] I. Steinke, J. Wier, L. Simon, and R. Seetan, ‘‘Sentiment analysis of online
no. 1, pp. 473–499, Nov. 2019. movie reviews using machine learning,’’ Int. J. Adv. Comput. Sci. Appl.,
[69] M. E. Basiri, S. Nemati, M. Abdar, E. Cambria, and U. R. Acharya, vol. 13, no. 9, pp. 1–7, 2022.
‘‘ABCDM: An attention-based bidirectional CNN-RNN deep model for [90] M. Purba, E. Ermatita, A. Abdiansah, H. Noprisson, V. Ayumi,
sentiment analysis,’’ Future Gener. Comput. Syst., vol. 115, pp. 279–294, H. Setiawan, U. Salamah, and Y. Yadi, ‘‘Effect of random splitting and
Feb. 2021. cross validation for Indonesian opinion mining using machine learning
[70] H. Elfaik and E. H. Nfaoui, ‘‘Deep bidirectional LSTM network learning- approach,’’ Int. J. Adv. Comput. Sci. Appl., vol. 13, no. 9, pp. 1–7, 2022.
based sentiment analysis for Arabic text,’’ J. Intell. Syst., vol. 30, no. 1,
[91] A. Alsanad, ‘‘An improved Arabic sentiment analysis approach using
pp. 395–412, Dec. 2020.
optimized multinomial Naïve Bayes classifier,’’ Int. J. Adv. Comput. Sci.
[71] R. Ranjan and A. K. Daniel, ‘‘Proposed hybrid model for sentiment
Appl., vol. 13, no. 8, pp. 1–10, 2022.
classification using CovNet-DualLSTM techniques,’’ ADCAIJ: Adv.
[92] W. Aljedaani, F. Rustam, M. W. Mkaouer, A. Ghallab, V. Rupapara,
Distrib. Comput. Artif. Intell. J., vol. 10, no. 4, pp. 401–418, Feb. 2022.
[72] G. Singh Chauhan, Y. Kumar Meena, D. Gopalani, and R. Nahta, P. B. Washington, E. Lee, and I. Ashraf, ‘‘Sentiment analysis on
‘‘A two-step hybrid unsupervised model with attention mechanism Twitter data integrating TextBlob and deep learning models: The case
for aspect extraction,’’ Expert Syst. Appl., vol. 161, Dec. 2020, of U.S. airline industry,’’ Knowl.-Based Syst., vol. 255, Nov. 2022,
Art. no. 113673. Art. no. 109780.
[73] M. Umer, I. Ashraf, A. Mehmood, S. Kumari, S. Ullah, and G. Sang [93] M. Arbane, R. Benlamri, Y. Brik, and A. D. Alahmar, ‘‘Social media-
Choi, ‘‘Sentiment analysis of tweets using a unified convolutional neural based COVID-19 sentiment classification model using bi-LSTM,’’ Expert
network-long short-term memory network model,’’ Comput. Intell., Syst. Appl., vol. 212, Feb. 2023, Art. no. 118710.
vol. 37, no. 1, pp. 409–434, Feb. 2021. [94] M. Venugopalan and D. Gupta, ‘‘A reinforced active learning approach
[74] M. Atzeni and D. Reforgiato Recupero, ‘‘Multi-domain sentiment for optimal sampling in aspect term extraction for sentiment analysis,’’
analysis with mimicked and polarized word embeddings for human– Expert Syst. Appl., vol. 209, Dec. 2022, Art. no. 118228.
robot interaction,’’ Future Gener. Comput. Syst., vol. 110, pp. 984–999, [95] V. Jain and K. L. Kashyap, ‘‘Ensemble hybrid model for Hindi COVID-
Sep. 2020. 19 text classification with Metaheuristic optimization algorithm,’’
[75] H.-T. Nguyen and L.-M. Nguyen, ‘‘ILWAANet: An interactive lexicon- Multimedia Tools Appl., vol. 82, no. 11, pp. 16839–16859, Oct. 2022.
aware word-aspect attention network for aspect-level sentiment classifi- [96] R. Dutta, N. Das, M. Majumder, and B. Jana, ‘‘Aspect based sentiment
cation on social networking,’’ Expert Syst. Appl., vol. 146, May 2020, analysis using multi-criteria decision-making and deep learning under
Art. no. 113065. COVID-19 pandemic in India,’’ CAAI Trans. Intell. Technol., vol. 8, no. 1,
[76] S.-T. Park, D.-Y. Kim, and G. Li, ‘‘An analysis of environmental pp. 219–234, Mar. 2023.
big data through the establishment of emotional classification system [97] S. Aslan, ‘‘A novel TCNN–Bi-LSTM deep learning model for predicting
model based on machine learning: Focus on multimedia contents for sentiments of tweets about COVID-19 vaccines,’’ Concurrency Comput.,
portal applications,’’ Multimedia Tools Appl., vol. 80, nos. 26–27, Pract. Exper., vol. 34, no. 28, p. e7387, Dec. 2022.
pp. 34459–34477, Nov. 2021. [98] D. C. Neagu, A. B. Rus, M. Grec, M. A. Boroianu, N. Bogdan, and A. Gal,
[77] R. Piryani, B. Piryani, V. K. Singh, and D. Pinto, ‘‘Sentiment analysis ‘‘Towards sentiment analysis for Romanian Twitter content,’’ Algorithms,
in nepali: Exploring machine learning and lexicon-based approaches,’’ vol. 15, no. 10, p. 357, Sep. 2022.
J. Intell. Fuzzy Syst., vol. 39, no. 2, pp. 2201–2212, Aug. 2020. [99] A. Pimpalkar and J. R. Raj R, ‘‘MBiLSTMGloVe: Embedding GloVe
[78] S. Soubraylu and R. Rajalakshmi, ‘‘Hybrid convolutional bidirectional knowledge into the corpus using multi-layer BiLSTM deep learning
recurrent neural network based sentiment analysis on movie reviews,’’ model for social media sentiment analysis,’’ Expert Syst. Appl., vol. 203,
Comput. Intell., vol. 37, no. 2, pp. 735–757, May 2021. Oct. 2022, Art. no. 117581.
[79] F. Es-Sabery, A. Hair, J. Qadir, B. Sainz-De-Abajo, [100] A. Diwali, K. Dashtipour, K. Saeedi, M. Gogate, E. Cambria, and
B. García-Zapirain, and I. D. L. Torre-Díez, ‘‘Sentence-level A. Hussain, ‘‘Arabic sentiment analysis using dependency-based rules
classification using parallel fuzzy deep learning classifier,’’ IEEE and deep neural networks,’’ Appl. Soft Comput., vol. 127, Sep. 2022,
Access, vol. 9, pp. 17943–17985, 2021. Art. no. 109377.
[80] M. Divyapushpalakshmi and R. Ramalakshmi, ‘‘An efficient sentimental [101] R. E. Saudy, A. E. D. M. El-Ghazaly, E. S. Nasr, and M. H. Gheith,
analysis using hybrid deep learning and optimization technique for ‘‘A novel hybrid sentiment analysis classification approach for mobile
Twitter using parts of speech (POS) tagging,’’ Int. J. Speech Technol., applications Arabic slang reviews,’’ Int. J. Adv. Comput. Sci. Appl.,
vol. 24, no. 2, pp. 329–339, Jun. 2021. vol. 13, no. 8, pp. 1–11, 2022.
[81] R. K. Behera, M. Jena, S. K. Rath, and S. Misra, ‘‘Co-LSTM:
[102] J.-W. Lin, T. D. Thanh, and R.-G. Chang, ‘‘Multi-channel word
Convolutional LSTM model for sentiment analysis in social big data,’’
embeddings for sentiment analysis,’’ Soft Comput., vol. 26, no. 22,
Inf. Process. Manage., vol. 58, no. 1, Jan. 2021, Art. no. 102435.
pp. 12703–12715, Nov. 2022.
[82] S. H. Janjua, G. F. Siddiqui, M. A. Sindhu, and U. Rashid, ‘‘Multi-
level aspect based sentiment classification of Twitter data: Using hybrid [103] K. L. Tan, C. P. Lee, K. M. Lim, and K. S. M. Anbananthen, ‘‘Sentiment
approach in deep learning,’’ PeerJ Comput. Sci., vol. 7, p. e433, Apr. 2021. analysis with ensemble hybrid deep learning model,’’ IEEE Access,
[83] H. S. Sharaf Al-Deen, Z. Zeng, R. Al-Sabri, and A. Hekmat, vol. 10, pp. 103694–103704, 2022.
‘‘An improved model for analyzing textual sentiment based on a deep [104] V. Kumar, ‘‘Spatiotemporal sentiment variation analysis of geotagged
neural network using multi-head attention mechanism,’’ Appl. Syst. COVID-19 tweets from India using a hybrid deep learning model,’’ Sci.
Innov., vol. 4, no. 4, p. 85, Oct. 2021. Rep., vol. 12, no. 1, p. 1849, Feb. 2022.
[84] C. Singla, F. N. Al-Wesabi, Y. Singh Pathania, B. Sulaiman Alfurhood, [105] A. Londhe and P. P. Rao, ‘‘Incremental learning based optimized
A. Mustafa Hilal, M. Rizwanullah, M. Ahmed Hamza, and M. Mahzari, sentiment classification using hybrid two-stage LSTM-SVM classifier,’’
‘‘An optimized deep learning model for emotion classification in tweets,’’ Int. J. Adv. Comput. Sci. Appl., vol. 13, no. 6, pp. 611–619, 2022.
Comput., Mater. Continua, vol. 70, no. 3, pp. 6365–6380, 2022. [106] E. Omara, M. Mousa, and N. Ismail, ‘‘Character gated recurrent neural
[85] K. L. Tan, C. P. Lee, K. S. M. Anbananthen, and K. M. Lim, ‘‘RoBERTa- networks for Arabic sentiment analysis,’’ Sci. Rep., vol. 12, no. 1, p. 9779,
LSTM: A hybrid model for sentiment analysis with transformer and Jun. 2022.
recurrent neural network,’’ IEEE Access, vol. 10, pp. 21517–21525, 2022. [107] K. Shanmugavadivel, S. H. Sampath, P. Nandhakumar, P. Mahalingam,
[86] R. A. Saleh, M. Driss, and I. Almomani, ‘‘CBiLSTM: A hybrid deep M. Subramanian, P. K. Kumaresan, and R. Priyadharshini, ‘‘An analysis
learning model for efficient reputation assessment of cloud services,’’ of machine learning models for sentiment analysis of Tamil code-mixed
IEEE Access, vol. 10, pp. 35321–35335, 2022. data,’’ Comput. Speech Lang., vol. 76, Nov. 2022, Art. no. 101407.

193126 VOLUME 12, 2024

Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

[108] M. Karim, M. M. Saad Missen, M. Umer, A. Fida, A. A. Eshmawi, [129] S. Aslan, ‘‘A deep learning-based sentiment analysis approach
A. Mohamed, and I. Ashraf, ‘‘Comprehension of polarity of articles by (MF-CNN-BILSTM) and topic modeling of tweets related to the
citation sentiment analysis using TF-IDF and ML classifiers,’’ PeerJ Ukraine–Russia conflict,’’ Appl. Soft Comput., vol. 143, Aug. 2023,
Comput. Sci., vol. 8, p. e1107, Dec. 2022. Art. no. 110404.
[109] N. Aslam, K. Xia, F. Rustam, E. Lee, and I. Ashraf, ‘‘Self voting [130] Y. Zahidi, Y. Al-Amrani, and Y. El Younoussi, ‘‘Improving Arabic
classification model for online meeting app review sentiment analysis and sentiment analysis using LSTM based on word embedding models,’’
topic modeling,’’ PeerJ Comput. Sci., vol. 8, p. e1141, Dec. 2022. Vietnam J. Comput. Sci., vol. 10, no. 3, pp. 391–407, Aug. 2023.
[110] A. Karanikola, G. Davrazos, C. M. Liapis, and S. Kotsiantis, ‘‘Financial [131] J. Cao, J. Li, M. Yin, and Y. Wang, ‘‘Online reviews sentiment analysis
sentiment analysis: Classic methods vs. Deep learning models,’’ Intell. and product feature improvement with deep learning,’’ ACM Trans.
Decis. Technol., vol. 17, no. 4, pp. 893–915, Nov. 2023. Asian Low-Resource Lang. Inf. Process., vol. 22, no. 8, pp. 1–17,
[111] S. Ounacer, D. Mhamdi, S. Ardchir, A. Daif, and M. Azzouazi, Aug. 2023.
‘‘Customer sentiment analysis in hotel reviews through natural language [132] S. Angamuthu and P. Trojovský, ‘‘Integrating multi-criteria decision-
processing techniques,’’ Int. J. Adv. Comput. Sci. Appl., vol. 14, no. 1, making with hybrid deep learning for sentiment analysis in recommender
pp. 1–11, 2023. systems,’’ PeerJ Comput. Sci., vol. 9, p. e1497, Aug. 2023.
[112] M. Sha, ‘‘A cloud based sentiment analysis through logistic regres- [133] K. K. Mohbey, G. Meena, S. Kumar, and K. Lokesh, ‘‘A CNN-
sion in AWS platform,’’ Comput. Syst. Sci. Eng., vol. 45, no. 1, LSTM-based hybrid deep learning approach for sentiment analysis on
pp. 857–868, 2023. monkeypox tweets,’’ New Gener. Comput., vol. 42, no. 1, pp. 89–107,
[113] N. E. Aoumeur, Z. Li, and E. M. Alshari, ‘‘Improving the polarity of Aug. 2023.
text through word2vec embedding for primary classical Arabic sentiment [134] N. Habbat, H. Nouri, H. Anoun, and L. Hassouni, ‘‘Sentiment analysis
analysis,’’ Neural Process. Lett., vol. 55, no. 3, pp. 2249–2264, Jun. 2023. of imbalanced datasets using BERT and ensemble stacking for deep
[114] M. Errami, M. A. Ouassil, R. Rachidi, B. Cherradi, S. Hamida, and learning,’’ Eng. Appl. Artif. Intell., vol. 126, Nov. 2023, Art. no. 106999.
A. Raihani, ‘‘Sentiment analysis on Moroccan dialect based on ML and [135] L. Dang, C. Wang, M.-H. Tsou, Y.-E. Hou, and H. Han, ‘‘Sentiment
social media content detection,’’ Int. J. Adv. Comput. Sci. Appl., vol. 14, analysis of COVID-19 related social distancing using Twitter data
no. 3, pp. 415–425, 2023. based on deep learning,’’ Multimedia Tools Appl., vol. 83, no. 11,
[115] M. A. Virgananda, I. Budi, and R. S. Ryan, ‘‘Purchase intention and pp. 32587–32612, Sep. 2023.
sentiment analysis on Twitter related to social commerce,’’ Int. J. Adv. [136] M. S. Başarslan and F. Kayaalp, ‘‘Sentiment analysis using a deep
Comput. Sci. Appl., vol. 14, no. 7, pp. 1–8, 2023. ensemble learning model,’’ Multimedia Tools Appl., vol. 83, no. 14,
[116] D. A. Musleh, I. Alkhwaja, A. Alkhwaja, M. Alghamdi, H. Abahussain, pp. 42207–42231, Oct. 2023.
F. Alfawaz, N. Min-Allah, and M. M. Abdulqader, ‘‘Arabic sentiment [137] M. Wankhade, C. S. R. Annavarapu, and A. Abraham, ‘‘CBMAFM:
analysis of Youtube comments: NLP-based machine learning approaches CNN-BiLSTM multi-attention fusion mechanism for sentiment classi-
for content evaluation,’’ Big Data Cognit. Comput., vol. 7, no. 3, p. 127, fication,’’ Multimedia Tools Appl., vol. 83, no. 17, pp. 51755–51786,
Jul. 2023. Nov. 2023.
[117] C. Yan, J. Liu, W. Liu, and X. Liu, ‘‘Sentiment analysis and topic mining [138] R. Ghasemi and S. Momtazi, ‘‘How a deep contextualized representation
using a novel deep attention-based parallel dual-channel model for online and attention mechanism justifies explainable cross-lingual sentiment
course reviews,’’ Cognit. Comput., vol. 15, no. 1, pp. 304–322, Jan. 2023. analysis,’’ ACM Trans. Asian Low-Resource Lang. Inf. Process., vol. 22,
[118] M. M. Abdelgwad, T. H. A. Soliman, and A. I. Taloba, ‘‘Arabic aspect no. 11, pp. 1–15, Nov. 2023.
sentiment polarity classification using BERT,’’ J. Big Data, vol. 9, no. 1, [139] I. N. Yulita, A. F. Al-Auza’i, A. S. Prabuwono, A. Sholahuddin,
p. 115, Dec. 2022. F. Ardiansyah, I. Sarathan, and Y. Djuyandi, ‘‘Bidirectional long short-
[119] M. M. Samia, A. Rajee, M. R. Hasan, M. O. Faruq, and P. C. Paul, term memory for analysis of public opinion sentiment on government
‘‘Aspect-based sentiment analysis for Bengali text using bidirectional policy during the COVID-19 pandemic,’’ Int. J. Adv. Comput. Sci. Appl.,
encoder representations from transformers (BERT),’’ Int. J. Adv. Comput. vol. 14, no. 11, pp. 877–885, 2023.
Sci. Appl., vol. 13, no. 12, pp. 1–9, 2022. [140] W. Chang and M. Zhu, ‘‘Sentiment analysis method of consumer
[120] M. Al-Smadi, M. M. Hammad, S. A. Al-Zboon, S. Al-Tawalbeh, and comment text based on BERT and hierarchical attention in e-commerce
E. Cambria, ‘‘Gated recurrent unit with multilingual universal sentence big data environment,’’ J. Intell. Syst., vol. 32, no. 1, Dec. 2023,
encoder for Arabic aspect-based sentiment analysis,’’ Knowl.-Based Syst., Art. no. 20230025.
vol. 261, Feb. 2023, Art. no. 107540. [141] H. Murfi, T. Gowandi, G. Ardaneswari, and S. Nurrohmah, ‘‘BERT-
[121] M. Qorich and R. El Ouazzani, ‘‘Text sentiment classification of Amazon based combination of convolutional and recurrent neural network for
reviews using word embeddings and convolutional neural networks,’’ Indonesian sentiment analysis,’’ Appl. Soft Comput., vol. 151, Jan. 2024,
J. Supercomput., vol. 79, no. 10, pp. 11029–11054, Jul. 2023. Art. no. 111112.
[122] A. Siagh, F. Z. Laallam, O. Kazar, and H. Salem, ‘‘An improved sentiment [142] I. Kanwal, F. Wahid, S. Ali, A.-U. Rehman, A. Alkhayyat, and
classification model based on data quality and word embeddings,’’ A. Al-Radaei, ‘‘Sentiment analysis using hybrid model of stacked auto-
J. Supercomput., vol. 79, no. 11, pp. 11871–11894, Jul. 2023. encoder-based feature extraction and long short term memory-based
[123] A. Alqarni and A. Rahman, ‘‘Arabic tweets-based sentiment analysis to classification approach,’’ IEEE Access, vol. 11, pp. 124181–124197,
investigate the impact of COVID-19 in KSA: A deep learning approach,’’ 2023.
Big Data Cognit. Comput., vol. 7, no. 1, p. 16, Jan. 2023. [143] P. Thiengburanathum and P. Charoenkwan, ‘‘SETAR: Stacking ensemble
[124] H. Said, B. S. Tawfik, and M. A. Makhlouf, ‘‘A deep learning approach learning for Thai sentiment analysis using RoBERTa and hybrid feature
for sentiment classification of COVID-19 vaccination tweets,’’ Int. J. Adv. representation,’’ IEEE Access, vol. 11, pp. 92822–92837, 2023.
Comput. Sci. Appl., vol. 14, no. 4, pp. 1–10, 2023. [144] C. Ahle Touate, R. El Ayachi, and M. Biniz, ‘‘Classification of sentiment
[125] N. Parveen, P. Chakrabarti, B. T. Hung, and A. Shaik, ‘‘Twitter sentiment using optimized hybrid deep learning model,’’ Comput. Informat., vol. 42,
analysis using hybrid gated attention recurrent network,’’ J. Big Data, no. 3, pp. 651–666, 2023.
vol. 10, no. 1, p. 50, Apr. 2023. [145] J. Khan, N. Ahmad, S. Khalid, F. Ali, and Y. Lee, ‘‘Sentiment and context-
[126] C. V. Loc, T. X. Viet, T. H. Viet, L. H. Thao, and N. H. Viet, ‘‘Pre- aware hybrid DNN with attention for text sentiment classification,’’ IEEE
trained language model-based deep learning for sentiment classification Access, vol. 11, pp. 28162–28179, 2023.
of Vietnamese feedback,’’ Int. J. Comput. Intell. Appl., vol. 22, no. 3, [146] B. B. Al-Onazi, A. Q. A. Hassan, M. K. Nour, M. Al Duhayyim,
Sep. 2023, Art. no. 2350016. A. Mohamed, A. Atta Abdelmageed, I. Yaseen, and G. Pasha Mohammed,
[127] D. A. J. Daniel and M. J. Meena, ‘‘Deep learning-based hybrid ‘‘Quantum particle swarm optimization with deep learning-based Arabic
sentiment analysis with feature selection using optimization tweets sentiment analysis,’’ Comput., Mater. Continua, vol. 75, no. 2,
algorithm,’’ Multimedia Tools Appl., vol. 82, no. 28, pp. 43273–43296, pp. 2575–2591, 2023.
Nov. 2023. [147] W. Ahmad, H. U. Khan, T. Iqbal, and S. Iqbal, ‘‘Attention-based
[128] A. Baqach and A. Battou, ‘‘CLAS: A new deep learning approach for multi-channel gated recurrent neural networks: A novel feature-centric
sentiment analysis from Twitter data,’’ Multimedia Tools Appl., vol. 82, approach for aspect-based sentiment classification,’’ IEEE Access,
no. 30, pp. 47457–47475, Dec. 2023. vol. 11, pp. 54408–54427, 2023.

VOLUME 12, 2024 193127

 Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

[148] R. Vatambeti, S. V. Mantena, K. V. D. Kiran, M. Manohar, and [171] V. Rupapara, F. Rustam, A. Amaar, P. B. Washington, E. Lee, and
C. Manjunath, ‘‘Twitter sentiment analysis on online food services based I. Ashraf, ‘‘Deepfake tweets classification using stacked bi-LSTM and
on elephant herd optimization with hybrid deep learning technique,’’ words embedding,’’ PeerJ Comput. Sci., vol. 7, p. e745, Oct. 2021.
Cluster Comput., vol. 27, no. 1, pp. 655–671, Feb. 2024. [172] A. Suciati and I. Budi, ‘‘Aspect-based sentiment analysis and emotion
[149] G. Kaur and A. Sharma, ‘‘A deep learning-based model using hybrid detection for code-mixed review,’’ Int. J. Adv. Comput. Sci. Appl., vol. 11,
feature extraction approach for consumer sentiment analysis,’’ J. Big no. 9, pp. 179–186, 2020.
Data, vol. 10, no. 1, p. 5, Jan. 2023. [173] S. Sharmin and D. Chakma, ‘‘Attention-based convolutional neural
[150] R. K. Dey and A. K. Das, ‘‘Modified term frequency-inverse document network for Bangla sentiment analysis,’’ AI Soc., vol. 36, no. 1,
frequency based deep hybrid framework for sentiment analysis,’’ pp. 381–396, Mar. 2021.
Multimedia Tools Appl., vol. 82, no. 21, pp. 32967–32990, Sep. 2023. [174] M. L. Barrón Estrada, R. Zatarain Cabada, R. Oramas Bustillos, and
[151] M. Irfan, N. Ayub, Q. A. Ahmed, S. Rahman, M. S. Bashir, M. Graff, ‘‘Opinion mining and emotion recognition applied to learning
G. Nowakowski, S. M. Alqhtani, and M. Sieja, ‘‘AQSA: Aspect-based environments,’’ Expert Syst. Appl., vol. 150, Jul. 2020, Art. no. 113265.
quality sentiment analysis for multi-labeling with improved ResNet [175] B. Gaye, D. Zhang, and A. Wulamu, ‘‘A tweet sentiment classification
hybrid algorithm,’’ Electronics, vol. 12, no. 6, p. 1298, Mar. 2023. approach using a hybrid stacked ensemble technique,’’ Information,
[152] M. Berrimi, M. Oussalah, A. Moussaoui, and M. Saidi, ‘‘Attention vol. 12, no. 9, p. 374, Sep. 2021.
mechanism architecture for Arabic sentiment analysis,’’ ACM Trans. [176] N. Capuano, L. Greco, P. Ritrovato, and M. Vento, ‘‘Sentiment
Asian Low-Resource Lang. Inf. Process., vol. 22, no. 4, pp. 1–26, analysis for customer relationship management: An incremental learning
Mar. 2023. approach,’’ Appl. Intell., vol. 51, pp. 3339–3352, Jun. 2021.
[153] S. Soumya and K. V. Pramod, ‘‘Hybrid deep learning approach for [177] O. Gokalp, E. Tasci, and A. Ugur, ‘‘A novel wrapper feature selection
sentiment classification of Malayalam tweets,’’ Int. J. Adv. Comput. Sci. algorithm based on iterated greedy Metaheuristic for sentiment classifi-
Appl., vol. 13, no. 4, pp. 1–9, 2022. cation,’’ Expert Syst. Appl., vol. 146, May 2020, Art. no. 113176.
[154] X. Zhang and Y. Ma, ‘‘An ALBERT-based TextCNN-hatt hybrid model [178] A. Madasu and S. Elango, ‘‘Efficient feature selection techniques
enhanced with topic knowledge for sentiment analysis of sudden-onset for sentiment analysis,’’ Multimedia Tools Appl., vol. 79, nos. 9–10,
disasters,’’ Eng. Appl. Artif. Intell., vol. 123, Aug. 2023, Art. no. 106136. pp. 6313–6335, Mar. 2020.
[155] C.-H. Lin and U. Nuha, ‘‘Sentiment analysis of Indonesian datasets based [179] L. Zheng, H. Wang, and S. Gao, ‘‘Sentimental feature selection for
on a hybrid deep-learning strategy,’’ J. Big Data, vol. 10, no. 1, p. 88, sentiment analysis of Chinese online reviews,’’ Int. J. Mach. Learn.
May 2023. Cybern., vol. 9, pp. 75–84, Jan. 2018.
[156] K. B. Muhammad and S. M. A. Burney, ‘‘Innovations in Urdu sentiment [180] M. I. A. Latiffi, M. R. Yaakub, and I. S. Ahmad, ‘‘Flower pollination
analysis using machine and deep learning techniques for two-class algorithm for feature selection in tweets sentiment analysis,’’ Int. J. Adv.
classification of symmetric datasets,’’ Symmetry, vol. 15, no. 5, p. 1027, Comput. Sci. Appl., vol. 13, no. 5, pp. 1–8, 2022.
May 2023. [181] A. Nasser and H. Sever, ‘‘A concept-based sentiment analysis approach
[157] M. Mujahid, K. Kanwal, F. Rustam, W. Aljedaani, and I. Ashraf, for Arabic,’’ Int. Arab J. Inf. Technol., vol. 17, no. 5, pp. 778–788,
‘‘Arabic ChatGPT tweets classification using RoBERTa and BERT Sep. 2020.
ensemble model,’’ ACM Trans. Asian Low-Resource Lang. Inf. Process., [182] P. Q. Tran, N. T. Trieu, N. V. Dao, H. T. Nguyen, and H. X. Huynh,
vol. 22, no. 8, pp. 1–23, Aug. 2023. ‘‘Effective opinion words extraction for food reviews classification,’’ Int.
[158] A. S. Talaat, ‘‘Sentiment analysis classification system using hybrid J. Adv. Comput. Sci. Appl., vol. 11, no. 7, pp. 421–426, 2020.
BERT models,’’ J. Big Data, vol. 10, no. 1, p. 110, Jun. 2023. [183] C. Catal and M. Nangir, ‘‘A sentiment classification model based on
[159] F. Alemayehu, M. Meshesha, and J. Abate, ‘‘Amharic political sentiment multiple classifiers,’’ Appl. Soft Comput., vol. 50, pp. 135–141, Jan. 2017.
analysis using deep learning approaches,’’ Sci. Rep., vol. 13, no. 1, [184] M. Giatsoglou, M. G. Vozalis, K. Diamantaras, A. Vakali, G. Sarigianni-
p. 17982, Oct. 2023. dis, and K. C. Chatzisavvas, ‘‘Sentiment analysis leveraging emotions and
[160] H. Tang and R. Zhang, ‘‘Research on Chinese microblog sentiment word embeddings,’’ Expert Syst. Appl., vol. 69, pp. 214–224, Mar. 2017.
classification based on TextCNN-BiLSTM model,’’ J. Inf. Process. Syst., [185] B. Al sari, R. Alkhaldi, D. Alsaffar, T. Alkhaldi, H. Almaymuni,
vol. 19, no. 6, pp. 842–857, Dec. 2023. N. Alnaim, N. Alghamdi, and S. O. latunji, ‘‘Sentiment analysis for
[161] N. Singh and U. C. Jaiswal, ‘‘Sentiment analysis based on Urdu reviews cruises in Saudi Arabia on social media platforms using machine learning
using hybrid deep learning models,’’ Appl. Comput. Syst., vol. 28, no. 2, algorithms,’’ J. Big Data, vol. 9, p. 21, Feb. 2022.
pp. 258–265, Dec. 2023. [186] A. Mohasseb, M. Bader-El-Den, and M. Cocea, ‘‘A customised grammar
[162] Y. Jin and A. Zhao, ‘‘Bert-based graph unlinked embedding for sentiment framework for query classification,’’ Expert Syst. Appl., vol. 135,
analysis,’’ Complex Intell. Syst., vol. 10, no. 2, pp. 2627–2638, Apr. 2024. pp. 164–180, Nov. 2019.
[163] M. Wankhade, C. S. R. Annavarapu, and M. K. Verma, ‘‘CBVoSD: [187] R. Chiha, M. B. Ayed, and C. d. C. Pereira, ‘‘A complete framework for
Context based vectors over sentiment domain ensemble model for review aspect-level and sentence-level sentiment analysis,’’ Appl. Intell., vol. 52,
classification,’’ J. Supercomput., vol. 78, no. 5, pp. 6411–6447, Apr. 2022. pp. 17845–17863, Apr. 2022.
[164] E. H. Mohamed, M. E. Moussa, and M. H. Haggag, ‘‘An enhanced [188] H. Sadr and M. Nazari Soleimandarabi, ‘‘ACNN-TL: Attention-based
sentiment analysis framework based on pre-trained word embedding,’’ convolutional neural network coupling with transfer learning and contex-
Int. J. Comput. Intell. Appl., vol. 19, no. 4, Dec. 2020, Art. no. 2050031. tualized word representation for enhancing the performance of sentiment
[165] S. Siddiqui, M. A. Rehman, S. M. Daudpota, and A. Waqas, ‘‘Opinion classification,’’ J. Supercomput., vol. 78, no. 7, pp. 10149–10175,
mining: An approach to feature engineering,’’ Int. J. Adv. Comput. Sci. May 2022.
Appl., vol. 10, no. 3, pp. 1–7, 2019. [189] R. Yan, Y. Yu, and D. Qiu, ‘‘Emotion-enhanced classification based
[166] S. Symeonidis, D. Effrosynidis, and A. Arampatzis, ‘‘A comparative on fuzzy reasoning,’’ Int. J. Mach. Learn. Cybern., vol. 13, no. 3,
evaluation of pre-processing techniques and their interactions for Twitter pp. 839–850, Mar. 2022.
sentiment analysis,’’ Expert Syst. Appl., vol. 110, pp. 298–310, Nov. 2018. [190] N. Gozuacik, C. O. Sakar, and S. Ozcan, ‘‘Social media-based opinion
[167] F. H. Khan, U. Qamar, and S. Bashir, ‘‘Lexicon based semantic detection retrieval for product analysis using multi-task deep neural networks,’’
of sentiments using expected likelihood estimate smoothed odds ratio,’’ Expert Syst. Appl., vol. 183, Nov. 2021, Art. no. 115388.
Artif. Intell. Rev., vol. 48, no. 1, pp. 113–138, Jun. 2017. [191] S. Al-Dabet, S. Tedmori, and M. Al-Smadi, ‘‘Enhancing Arabic aspect-
[168] M. Kamyab, G. Liu, A. Rasool, and M. Adjeisah, ‘‘ACR-SA: Attention- based sentiment analysis using deep learning models,’’ Comput. Speech
based deep model through two-channel CNN and bi-RNN for sentiment Lang., vol. 69, Sep. 2021, Art. no. 101224.
analysis,’’ PeerJ Comput. Sci., vol. 8, p. e877, Mar. 2022. [192] F. Huang, X. Li, C. Yuan, S. Zhang, J. Zhang, and S. Qiao, ‘‘Attention-
[169] H. Suresh, ‘‘An innovative and efficient method for Twitter sentiment emotion-enhanced convolutional LSTM for sentiment analysis,’’ IEEE
analysis,’’ Int. J. Data Mining, Model. Manage., vol. 11, no. 1, pp. 1–18, Trans. Neural Netw. Learn. Syst., vol. 33, no. 9, pp. 4332–4345,
2019. Sep. 2022.
[170] H. Liu and M. Cocea, ‘‘Fuzzy information granulation towards [193] K. Chen and W. Ke, ‘‘A hierarchical neural model for target-based
interpretable sentiment analysis,’’ Granular Comput., vol. 2, no. 4, sentiment analysis,’’ Concurrency Comput., Pract. Exper., vol. 33, no. 10,
pp. 289–302, Dec. 2017. p. e6184, May 2021.

193128 VOLUME 12, 2024

Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

[194] J. Yang, X. Zou, W. Zhang, and H. Han, ‘‘Microblog sentiment analysis [216] M. E. Basiri, S. Nemati, M. Abdar, S. Asadi, and U. R. Acharrya,
via embedding social contexts into an attentive LSTM,’’ Eng. Appl. Artif. ‘‘A novel fusion-based deep learning model for sentiment analysis
Intell., vol. 97, Jan. 2021, Art. no. 104048. of COVID-19 tweets,’’ Knowl.-Based Syst., vol. 228, Sep. 2021,
[195] K. Ayyub, S. Iqbal, M. W. Nisar, S. G. Ahmad, and E. U. Munir, Art. no. 107242.
‘‘Stance detection using diverse feature sets based on machine learning [217] J. Tao and X. Fang, ‘‘Toward multi-label sentiment analysis: A transfer
techniques,’’ J. Intell. Fuzzy Syst., vol. 40, no. 5, pp. 9721–9740, learning based approach,’’ J. Big Data, vol. 7, p. 1, Jan. 2020.
Apr. 2021. [218] C. N. Dang, M. N. Moreno-García, and F. De la Prieta, ‘‘Using
[196] M. Kasri, M. Birjali, and A. Beni-Hssane, ‘‘Word2Sent: A new learning hybrid deep learning models of sentiment analysis and item genres
sentiment-embedding model with low dimension for sentence level in recommender systems for streaming services,’’ Electronics, vol. 10,
sentiment classification,’’ Concurrency Comput., Pract. Exper., vol. 33, no. 20, p. 2459, Oct. 2021.
no. 9, p. e6149, May 2021. [219] N. Liu and B. Shen, ‘‘Aspect term extraction via information-augmented
[197] L.-C. Chen, C.-M. Lee, and M.-Y. Chen, ‘‘Exploration of social media for neural network,’’ Complex Intell. Syst., vol. 9, no. 1, pp. 537–563,
sentiment analysis using deep learning,’’ Soft Comput., vol. 24, no. 11, Jul. 2022.
pp. 8187–8197, Jun. 2020. [220] S. Vernikou, A. Lyras, and A. Kanavos, ‘‘Multiclass sentiment analysis on
[198] F. Mehmood, M. U. Ghani, M. A. Ibrahim, R. Shahzadi, COVID-19-related tweets using deep learning models,’’ Neural Comput.
W. Mahmood, and M. N. Asim, ‘‘A precisely xtreme-multi channel Appl., vol. 34, no. 22, pp. 19615–19627, Nov. 2022.
hybrid approach for Roman Urdu sentiment analysis,’’ IEEE Access,
[221] C. Yan, J. Liu, W. Liu, and X. Liu, ‘‘Research on public opinion
vol. 8, pp. 192740–192759, 2020.
sentiment classification based on attention parallel dual-channel deep
[199] M. U. Salur and I. Aydin, ‘‘A novel hybrid deep learning model for
learning hybrid model,’’ Eng. Appl. Artif. Intell., vol. 116, Nov. 2022,
sentiment classification,’’ IEEE Access, vol. 8, pp. 58080–58093, 2020.
Art. no. 105448.
[200] C. N. Dang, M. N. Moreno-García, and F. De la Prieta, ‘‘Hybrid deep
[222] K. Shrivastava and S. Kumar, ‘‘A sentiment analysis system for the Hindi
learning models for sentiment analysis,’’ Complexity, vol. 2021, no. 1,
language by integrating gated recurrent unit with genetic algorithm,’’ Int.
Aug. 2021, Art. no. 9986920.
Arab J. Inf. Technol., vol. 17, no. 6, pp. 954–964, 2020.
[201] T. Li, H. Xu, Z. Liu, Z. Dong, Q. Liu, J. Li, S. Fan, and X. Sun,
‘‘A spatiotemporal multi-feature extraction framework for opinion [223] K. Dashtipour, M. Gogate, J. Li, F. Jiang, B. Kong, and A. Hussain,
mining,’’ Neurocomputing, vol. 490, pp. 337–346, Jun. 2022. ‘‘A hybrid Persian sentiment analysis framework: Integrating dependency
[202] F. Alotaibi and V. Gupta, ‘‘Sentiment analysis system using hybrid word grammar based rules and deep neural networks,’’ Neurocomputing,
embeddings with convolutional recurrent neural network,’’ Int. Arab vol. 380, pp. 1–10, Mar. 2020.
J. Inf. Technol., 5 vol. 19, no. 3, pp. 1–6, 2022. [224] A. Vohra and R. Garg, ‘‘Deep learning based sentiment analysis of public
[203] H. Yang, A. Alsadoon, P. W. C. Prasad, T. Al-Dala’In, T. A. Rashid, perception of working from home through tweets,’’ J. Intell. Inf. Syst.,
A. Maag, and O. H. Alsadoon, ‘‘Deep learning neural networks for vol. 60, no. 1, pp. 255–274, Aug. 2022.
emotion classification from text: Enhanced leaky rectified linear unit [225] A. Muslim, A. B. Mutiara, R. Refianti, C. M. Karyati, and G. Setiawan,
activation and weighted loss,’’ Multimedia Tools Appl., vol. 81, no. 11, ‘‘Comparison of accuracy between long short-term memory-deep learn-
pp. 15439–15468, May 2022. ing and multinomial logistic regression-machine learning in sentiment
[204] O. Alharbi, ‘‘A deep learning approach combining CNN and Bi-LSTM analysis on Twitter,’’ Int. J. Adv. Comput. Sci. Appl., vol. 11, no. 2,
with SVM classifier for Arabic sentiment analysis,’’ Int. J. Adv. Comput. pp. 1–8, 2020.
Sci. Appl., vol. 12, no. 6, pp. 1–8, 2021. [226] F. A. Lovera, Y. C. Cardinale, and M. N. Homsi, ‘‘Sentiment analysis
[205] X. Li, R. Lu, P. Liu, and Z. Zhu, ‘‘Graph convolutional net- in Twitter based on knowledge graph and deep learning classification,’’
works with hierarchical multi-head attention for aspect-level sentiment Electronics, vol. 10, no. 22, p. 2739, Nov. 2021.
classification,’’ J. Supercomput., vol. 78, no. 13, pp. 14846–14865, [227] H. Yu, G. Lu, Q. Cai, and Y. Xue, ‘‘A KGE based knowledge enhancing
Apr. 2022. method for aspect-level sentiment classification,’’ Mathematics, vol. 10,
[206] J. Zhou, J. X. Huang, Q. V. Hu, and L. He, ‘‘Is position important? Deep no. 20, p. 3908, Oct. 2022.
multi-task learning for aspect-based sentiment analysis,’’ Appl. Intell., [228] A. Naresh and P. V. Krishna, ‘‘An efficient approach for sentiment
vol. 50, pp. 3367–3378, Jun. 2020. analysis using machine learning algorithm,’’ Evol. Intell., vol. 14,
[207] N. Liu and B. Shen, ‘‘ReMemNN: A novel memory neural network for pp. 725–731, Jun. 2021.
powerful interaction in aspect-based sentiment analysis,’’ Neurocomput- [229] W. Muhammad, M. Mushtaq, K. N. Junejo, and M. Y. Khan, ‘‘Sentiment
ing, vol. 395, pp. 66–77, Jun. 2020. analysis of product reviews in the absence of labelled data using
[208] A. Onan, ‘‘Sentiment analysis on product reviews based on weighted supervised learning approaches,’’ Malaysian J. Comput. Sci., vol. 33,
word embeddings and deep neural networks,’’ Concurrency Comput., no. 2, pp. 118–132, Apr. 2020.
Pract. Exper., vol. 33, no. 23, p. e5909, Dec. 2021. [230] Z. Wang and Z. Lin, ‘‘Optimal feature selection for learning-based
[209] G. Lu, J. Gan, J. Yin, Z. Luo, B. Li, and X. Zhao, ‘‘Multi-task learning algorithms for sentiment classification,’’ Cogn. Comput., vol. 12,
using a hybrid representation for text classification,’’ Neural Comput. pp. 238–248, Jan. 2020.
Appl., vol. 32, no. 11, pp. 6467–6480, Jun. 2020.
[231] I. Safder, Z. Mahmood, R. Sarwar, S.-U. Hassan, F. Zaman,
[210] G. Lu, X. Zhao, J. Yin, W. Yang, and B. Li, ‘‘Multi-task learning using
R. M. A. Nawab, F. Bukhari, R. A. Abbasi, S. Alelyani, N. R. Aljohani,
variational auto-encoder for sentiment classification,’’ Pattern Recognit.
and R. Nawaz, ‘‘Sentiment analysis for Urdu online reviews using deep
Lett., vol. 132, pp. 115–122, Apr. 2020.
learning models,’’ Expert Syst., vol. 38, no. 8, 2021, Art. no. e12751.
[211] G. Liu, X. Huang, X. Liu, and A. Yang, ‘‘A novel aspect-based sentiment
[232] O. Appel, F. Chiclana, J. Carter, and H. Fujita, ‘‘Cross-ratio uninorms
analysis network model based on multilingual hierarchy in online social
as an effective aggregation mechanism in sentiment analysis,’’ Knowl.-
network,’’ Comput. J., vol. 63, no. 1, pp. 410–424, Jan. 2020.
Based Syst., vol. 124, pp. 16–22, May 2017.
[212] H.-J. Park, M. Song, and K.-S. Shin, ‘‘Deep learning models and datasets
for aspect term sentiment classification: Implementing holistic recurrent [233] L. Chiruzzo, M. Etcheverry, and A. Rosá, ‘‘Sentiment analysis in Spanish
attention on target-dependent memories,’’ Knowl.-Based Syst., vol. 187, tweets: Some experiments with focus on neutral tweets,’’ Procesamiento
Jan. 2020, Art. no. 104825. del Lenguaje Natural, vol. 64, pp. 109–116, Jan. 2020.
[213] S. L. Ramaswamy and J. Chinnappan, ‘‘RecogNet-LSTM+CNN: A [234] R. Dhanalakshmi and T. Sri Devi, ‘‘Adaptive cognitive intelligence in
hybrid network with attention mechanism for aspect categorization analyzing employee feedback using LSTM,’’ J. Intell. Fuzzy Syst., vol. 39,
and sentiment classification,’’ J. Intell. Inf. Syst., vol. 58, pp. 379–404, no. 6, pp. 8069–8078, Dec. 2020.
Jan. 2022. [235] Q. Cheng, Y. Ke, and A. Abdelmouty, ‘‘Negative emotion diffusion and
[214] L. Bao, P. Lambert, and T. Badia, ‘‘Improving aspect-based neural sen- intervention countermeasures of social networks based on deep learning,’’
timent classification with lexicon enhancement, attention regularization J. Intell. Fuzzy Syst., vol. 39, no. 4, pp. 4935–4945, Oct. 2020.
and sentiment induction,’’ Natural Lang. Eng., vol. 30, no. 1, pp. 1–30, [236] S. Rani, A. K. Bashir, A. Alhudhaif, D. Koundal, and E. S. Gunduz,
2022. ‘‘An efficient CNN-LSTM model for sentiment detection in #BlackLives-
[215] L.-C. Cheng, Y.-L. Chen, and Y.-Y. Liao, ‘‘Aspect-based sentiment Matter,’’ Expert Syst. Appl., vol. 193, May 2022, Art. no. 116256.
analysis with component focusing multi-head co-attention networks,’’ [237] K. Pasupa and T. S. Na Ayutthaya, ‘‘Hybrid deep learning models for Thai
Neurocomputing, vol. 489, pp. 9–17, Jun. 2022. sentiment analysis,’’ Cogn. Comput., vol. 14, pp. 167–193, Jan. 2022.

VOLUME 12, 2024 193129

 Y. Mamani-Coaquira, E. Villanueva: Review on Text Sentiment Analysis With ML and DL Techniques

[238] Y. Long, R. Xiang, Q. Lu, C.-R. Huang, and M. Li, ‘‘Improving attention EDWIN VILLANUEVA received the degree in
model based on cognition grounded data for sentiment analysis,’’ IEEE electrical engineering from the National Univer-
Trans. Affect. Comput., vol. 12, no. 4, pp. 900–912, Oct. 2021. sity of San Agustin of Arequipa, Peru, in 2000,
[239] N. Liu and B. Shen, ‘‘Aspect-based sentiment analysis with gated the master’s degree from the University of São
alternate neural network,’’ Knowl.-Based Syst., vol. 188, Jan. 2020, Paulo, Brazil, the master’s degree in electrical
Art. no. 105010. engineering, in 2007, and the Ph.D. degree in
science, in 2012
He completed postdoctoral stays with the Insti-
tute of Mathematical and Computer Sciences,
University of São Paulo, and the Institute of
YONATAN MAMANI-COAQUIRA received the Functional Genomics, Universität Regensburg, Germany. He was an
bachelor’s degree in computer and systems engi- Engineer in several public and private institutions in Peru, for more than
neering from the National University Micaela six years, accumulating experience in the telecommunications and software
Bastidas of Apurimac, Peru, in 2013, and the development sectors. He participated as a Postdoctoral Researcher with the
master’s degree in computer science from the Institute of Mathematical and Computer Sciences, University of São Paulo,
National University of San Antonio Abad of and Universität Regensburg. Since 2017, he has been a Senior Researcher
Cusco, Peru, in 2021. Currently, he is pursuing the with the Pontifical Catholic University of Peru. His research has mainly
Ph.D. degree in engineering, with a specialization focused on areas of computer science, specifically artificial intelligence,
in computer science with the Pontifical Catholic machine learning, probabilistic graphical models, signal processing, and
University of Peru (PUCP). bioinformatics. He had been leading research and technological development
He is a Conducting Researcher with the Artificial Intelligence Group, projects in multidisciplinary fields, such as modeling and prediction of
PUCP. His research interests include machine learning, deep learning, data atmospheric phenomena, air quality monitoring and modeling, and modeling
analytics, computer vision, and natural language processing. He previously and prediction of financial markets.
had an internship and collaborated as a Researcher with the Data Analytics
Science and Engineering Laboratory, Pablo de Olavide University, Seville,
Spain.

193130 VOLUME 12, 2024