Noname manuscript No.

(will be inserted by the editor)

A Comprehensive Review on Sentiment Analysis: Tasks,

Approaches and Applications

Sudhanshu Kumar∗1 · Partha Pratim Roy1 ·

Debi Prosad Dogra 2 · Byung-Gyu Kim 3
arXiv:2311.11250v1 [cs.AI] 19 Nov 2023

Received: date / Accepted: date

Abstract Sentiment analysis (SA) is an emerging field in text mining. It is the pro-
cess of computationally identifying and categorizing opinions expressed in a piece
of text over different social media platforms. Social media plays an essential role
in knowing the customer mindset towards a product, services, and the latest mar-
ket trends. Most organizations depend on the customer’s response and feedback to
upgrade their offered products and services. SA or opinion mining seems to be a
promising research area for various domains. It plays a vital role in analyzing big
data generated daily in structured and unstructured formats over the internet. This
survey paper defines sentiment and its recent research and development in different
domains, including voice, images, videos, and text. The challenges and opportunities
of sentiment analysis are also discussed in the paper.

Keywords Sentiment Analysis, Machine Learning, Lexicon-based approach, Deep

Learning, Natural Language Processing

Sudhanshu Kumar
E-mail: skumar2@cs.iitr.ac.in
Partha Pratim Roy
E-mail: partha@cs.iitr.ac.in
Debi Prosad Dogra
E-mail: dpdogra@iitbbs.ac.in
Byung-Gyu Kim
E-mail: bg.kim@sookmyung.ac.kr

1 Department of Computer Science and Engineering, IIT Roorkee, 247667, India

2 Schoolof Electrical Sciences, IIT Bhubaneswar, Odisha 752050, India.
3 Department of IT Engineering, Sookmyung Women’s University, Seoul 04310, South Korea

∗ Corresponding Author
2 Sudhanshu Kumar∗1 et al.

1 Introduction

Sentiment Analysis is the computational study of people’s opinions, attitudes, and

emotions in the form of different modalities (text, image, and speech) toward an en-
tity that represents topics, events, issues, products, services, and organizations. SA is
the branch of many fields such as machine learning, data mining, natural language
processing, and computational linguistics. Natural Language Processing (NLP) gen-
erally started back in the 1950s; little attention was paid by researchers to people’s
opinions and sentiment analysis until 2005. With the advancement of web 2.0, web
3.0, web 4.0, social media thrusts SA’s development. Social media propels the growth
of sentiment analysis. Most of the literature is on the English language, but many
publications currently tackle the multilingual issue. SA is a suitcase research prob-
lem [18] that is the combination of NLP tasks such as named entity recognition [61],
concept extraction [17], sarcasm detection [81], aspect extraction [62], and subjectiv-
ity detection. Subjective information indicates the opinions of opinion holders, while
objective texts show some objective facts. For example, ”The food is great and de-
licious.” These opinion words are subjective. Subjective texts can have a positive or
negative sentiment.
SA classification process, as shown in Figure 1 and 2 uses any classification
model to classify the reviews into positive, negative and neutral classes. There are
three levels of SA such as document level, sentence level, and aspect level. In the
document level, the whole document expresses a positive or negative opinion. For
Example, the product reviews document has either positive or negative opinions for
a product. It represents a single opinion for a document, so it comes under the doc-
ument level. Sentence level is the second category widely used in e-commerce sites
in which each sentence classifies into positive, negative, and neutral opinions. Aspect
level sentiment analysis is also called feature-based analysis. In this type of analysis,
each review categorizes into aspects and their target opinions. This level shows more
insights about the opinion that it is positive or negative for which aspect. For Exam-
ple, ‘The Food was very good at the hotel.’ It is an aspect-based SA where food is
one aspect of the review.
Different sentiment classification techniques are shown in Figure 3. It is divided
into two categories, i.e., lexicon-based approach and machine learning approach. The
Lexicon-based approach uses the dictionaries of words annotated with their seman-
tic orientation, classified into the dictionary and corpus-based approach. The second
category is the machine learning approach based on different types of learning like
supervised, unsupervised, and semi-supervised. Depending on the nature of the data,
these learning techniques are used and predict the result. Different deep learning-
based and machine learning-based techniques are the most popular ones. The total
number of research publications year-wise is as shown in Figure 4. It shows that as
the advancement of industry 4.0 and now it’s 5.0, the numbers of research papers are
increasing year by year.
The contributions of the paper are as follows:

1. A large number of literature has been reviewed in sentiment analysis process from
multiple domains and identify the pros and cons of all approaches.
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 3

Fig. 1: The process to classify the review into positive, negative, neutral using Ma-
chine learning.

2. Summarizing each of the surveyed articles in detail, including the problems ad-
dressed, dataset details and methods.
3. Analyses of existing applications in order to determine which one is most suitable
for certain application.
4. Discussing the challenges and application of sentiment analysis in order to keep
up the current research trends.

The rest of the paper is organized as follows. In Section 2, we discuss the state of
the art discussion on SA. Detailed discussion on the existing work, open issues, and
possible applications of sentiment analysis is presented in Section 3 and Section 4.
Finally, in the last Section 5 the research work has been concluded.
4 Sudhanshu Kumar∗1 et al.

Fig. 2: The process to classify the review into positive, negative and neutral using
Deep learning techniques.

Fig. 3: All sentiment analysis classification techniques from traditional to the latest
one have been shown in this figure. Initially, It is divided into machine learning and
lexicon-based approach, which further divide into different algorithms.
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 5

Fig. 4: The number of research papers with the Scopus index are published in the last
ten years from 2011 to Mid 2020. The number of publications is increasing yearly as
the advancement in technology and the evolution of industry 4.0.

2 Terminology and background concepts

Opinion, views, and feeling are often used interchangeably in the different literature
[70]. SA is also related to many terms such as emotions, moods, and feelings, which
sometimes confuse the reader with opinion or SA. Emotion is related to the percep-
tion of the stimulus and the triggering of the bodily response. For example, one person
shows an angry response when they lose the job, and in another case, they feel joy in
the same situation. Many authors [70, 93] showed that emotion is short term, while
the mood is a long-term phenomenon. They differentiate on both terms basis on their
duration.
A schematic representation of opinion and sentiment are given in Figure 5.
SA has raised a growing interest in financial and political forecasting, e-health,
e-tourism, and dialogue systems. The authors [36] proposed trust-based ranking and
the recommendation tool to improve online software services recommendation. This
system enhances the existing recommendation system (content-based and collabora-
tive filtering based) algorithm by considering the external attributes. The proposed
system result was evaluated on the Amazon marketplace review dataset and showed
a better ranking.
SA used many libraries such as TextBlob 1 and naive Bayes to classify the content
based on polarity score and subjectivity. In [35], the authors proposed a sentiment
polarity categorization process, which was the fundamental problem of sentiment
earlier. The amazon product reviews dataset’s experimental results achieved an F1
score of 0.8 and .0.73 for sentence-level and review-level categorization, respectively.
The polarity shift problem is one of the challenges in sentiment analysis to predict
1 https://textblob.readthedocs.io/en/dev/
6 Sudhanshu Kumar∗1 et al.

Fig. 5: Schematic structure of sentiments [70], which further divide into sentiment
holder, emotional disposition and object of the review. The emotion is short-term,
while the mood is long-term disposition.

user reviews. Xia et al. [114] proposed dual sentiment analysis model to address
the polarity shift problem. The model trained for sentiment-reversed review for both
training and testing. The result of the multi-domain and Chinese datasets showed the
effectiveness of the model.
In [25], the authors proposed a map reducing paradigm to collect the user’s data
from Facebook to understand brand reviews. They refined their approach through
the iterative process of data pre-processing. In [6], the authors proposed an auto-
matic feedback technique based on Twitter data. Different classifiers like SVM, Naive
Bayes, and maximum entropy are used on Twitter comments. Out of these classifiers,
SVM-based performance was the highest. In [102], the authors proposed a random
walk algorithm for domain-oriented sentiment lexicon based on utilizing sentiment
words and documents from both the old and target domains. The proposed algorithm
reflects four kinds of relationships (words to documents, words to words, documents
to words, and documents to documents) between words and documents. Experimen-
tal results indicate improvements in identifying the polarities of sentiment words.
Day et al. [26] presented that analytical methods use deep learning in financial news
sources to forecast stock price trends. The authors found that financial news media
sources can reveal investment information. Sentiment analysis aims to classify text in
positive and negative polarity scores useful in quantifying different affective states of
a user [83]. Cambria et al. [18] have developed an NLP approach that leverages both
data and theory-driven methods to understand natural language.
Many approaches have simple categorization problems; however, sentiment analysis
is a big suitcase problem requiring multiple polarity detection tasks. NLP problems
divide into three layers: syntactic, semantics, and pragmatics. Each layer has a differ-
ent subtask to process each layer’s text output as input for the next layer. In [81], the
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 7

authors have developed a pre-trained model for extracting emotion, sentiment, and
personality features from sarcastic tweets using CNN. Experiments were conducted
on a dataset consisting of both sarcastic and non-sarcastic tweets. Results computed
on three datasets with F1 scores of 87%, 92.32%, and 93.30%, respectively.

2.1 Sentiment in Text

Text is mainly an important medium to express the user’s state of mind in reviews
and comments on the internet. It was the primary mode of communication in early
1990 when e-commerce company amazon was the first company to do business on-
line. In 1992, the authors [44] proposed an approach based on the sentence’s direc-
tionality. The approach is based on the semantic orientation of the sentence to de-
termine the directionality of the text. Another researcher [89] whose theory is based
on the information’s subjective point of view. In the early 2000s, many researchers
[23, 69, 77, 106, 107, 112] worked on sentiments analysis and opinions mining. Na-
sukawa et al. [73] showed the high precision result on customer reviews and news ar-
ticles available over web pages. They classified the specific subjects from a document
in positive or negative polarity. This paper’s result rise in interest to other researchers
in this domain. The influential 2008 review of Pang and Lee [76] covers techniques
and approaches that promise to directly enable opinion-oriented information-seeking
systems on benchmark datasets in recent research. Here, we discuss sentiment analy-
sis in NLP, including its different methods, such as supervised and unsupervised.

2.1.1 Supervised Approach

It is based on the annotated dataset (labeled data) to build a prediction model. This
approach builds a feature vector of the text, either aspect or word frequency, then the
model learns (training) on the dataset and gives prediction for unseen data in testing.
The first paper [111] used this approach to classify the text as subjective or objective
on the gold standard dataset. It achieves 81.5% accuracy on the probabilistic classi-
fier. There are different approaches in machine learning, like the supervised and unsu-
pervised approaches. The supervised approach was used on stock trading [24] domain
to find the sentiment analysis. Further development focus on user comments available
on the E-commerce site. Many machine learning algorithms like SVM, Naive Bayes,
and linear regression solve the problem related to a different domain. SVM was the
most suitable model for product reviews in supervised sentiment analysis.
The authors [30] proposed BERT (bidirectional encoder representations from
transformers) model designed to pre-trained deep bidirectional from the unlabelled
text by jointly conditioning on both left and right context. Bidirectional means that
BERT learns information from both the left and the right side of a token’s context
during the training phase. The model is implemented on eleven NLP tasks such as
GLUE, MultiNLI, SWAG, SQuAD v1.1, etc., and shown the impressive state-of-the-
art result in the SA field. Unlike recent language representation models [80, 86], this
model is used for a wide range of tasks like question answering and language infer-
ence.
8 Sudhanshu Kumar∗1 et al.

2.1.2 Unsupervised Learning Approach

In this approach, the labeled data is not present, allowing an estimation based on ex-
pert knowledge. The most popular method in unsupervised learning is cluster analysis
to find the data’s hidden pattern. In sentiment analysis, the lexicon plays an essential
role in classifying the text into positive, negative, and neutral depending on the lexi-
con method, a combination of words or phrases. The most popular lexicon is General
Inquirer [100], it is a corpus of positive and negative terms.
The aim to improve sentence-level classification, recently few methods are per-
forming well such as SentiStrength [104], Valence Aware Dictionary and sEntiment
Reasoner (VADER) [46], and Umigon [59]. VADER is a lexicon and rule-based sen-
timent analysis method used to find the sentiment of the reviews available on dif-
ferent social media platforms. The reviews are shared by a user from different age
groups and gender, so these reviews are not available in the form that can be directly
processed by any method. It converted into a normalized form after pre-processing
of these reviews. VADER evaluates the words and their context on pre-processed re-
views based on a predefined dictionary with many words (sentiment lexicon) and their
corresponding numeric score. The method produces four metrics for each review; the
first three are positive, neutral, and negative. The last metric is a compound score
used to identify these reviews’ sentiment. VADER is popular among other methods
to analyze social media posts with slang, emoticons, and acronyms.

Du et al. [32] proposed an attention mechanism for news categories in Chinese

(NLPCC201) and English (REV1-v) datasets. The experimental result shows that
this mechanism is beneficial to assign a score for keywords. The keywords that have a
higher score in the corpus mean that these keywords are more important to the dataset
than non-key words, so it improves the classifier’s accuracy compared to recurrent
neural network [66] and long short term memory [45]. This mechanism showed an
effective result in many research papers [122, 121, 117] for a different domain such as
document classification (yelp reviews, IMDB reviews, Yahoo answers, and amazon
reviews), understand human communication (language, vision, and acoustic modal-
ity), and video captioning, etc.

2.1.3 Word Embedding

The development of deep learning techniques in sentiment analysis shows promising

results in most real-world problems. Word embedding is the dominant approach in
NLP problems compare to one-hot encoding. If the words are present in the vocab-
ulary in one-hot encoding, then assign one else zero. The issue in one hot encoding
is a computational issue. When you increase your vocabulary by size n, the feature
size vector also increases by length n, requiring more computational time to train the
model. A word embedding is a learned representation for text data where words or
phrases with the same meaning have a similar representation mapped further either
in vector or real numbers. The strategy typically includes a mathematic concept from
a high-dimensional vector space to a lower-dimensional vector space. The vectors
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 9

encoding is related to linguistic regularities and patterns, each dimension related to

the word’s feature. The learning of word embedding is done by neural network [13]
from the text.
The most common word embedding system is word2vec, in which the words related
to each other, like king-queen and man-women, are represented in the vector space
near each other. The word2vec model approach is based on two models i.e. continu-
ous bag-of-words [67] and skip-gram model [68]. Another frequent word embedding
technique is Glove Vector [79] (GloVe), which utilizes both global statistics and local
statistics to train word vector fast and scalable. The word2vec captures local statistics
to do works well on analogy tasks. The author[7] proposed ensemble techniques that
were the combination of word embeddings and a linear algorithm on seven public
datasets extracted from the microblogging and movie reviews domain. This paper
showed that word embedding techniques enhance the proposed model’s performance
and work well in a smaller dataset. The deep learning algorithm does not perform
very well when the dataset is small because it requires a large amount of data to train
the model, so the word embedding algorithm is used in this case. Pre-trained word
embedding is used to solve many research problems [87, 103, 39].

2.1.4 Others Techniques

Qazi et al. [85] proposed assessing users’ opinions on multiple topics like a social get-
together, promoting efforts, and item inclinations. This study aims to find the users’
expectations and satisfaction at the post-purchase stage. They surveyed a question-
naire comprising seven sections, and the data was collected through LinkedIn and the
university mail servers. The authors utilized a disconfirmation hypothesis, a set of
seven theories, confirmatory factor analysis, and primary conditions to break down
the users’ information and assess the model. The model’s consequences demonstrated
that regular, comparative, and interesting assessments positively raise users’ desires.
The author presumed that a wide range of sentiments is a rich wellspring of data that
at last influences the customer loyalty level. Wang et al. [108] proposed a SentiRe-
lated algorithm to fill the gap between different domains. The traditional supervised
classification algorithm is performed well for a given domain but does not work well
on different domains. The SentiRelated algorithm is based on the Sentiment Related
Index to improve the model’s performance when tested on other domains. This al-
gorithm was validated on two datasets with different domains such as a computer,
Education, Hotel, Movie, Music, and Book reviews and showed 80% accuracy for
short texts. Social media platforms like Twitter are trending to become a common
platform for exchanging raw data and online text, providing a vast platform for sen-
timent analysis.
The author proposed [75] a novel metaheuristic method based on Cuckoo Search
and K-means (called CSK). It enlightens the clustering-based methods for analysing
Twitter tweets to find the user’s viewpoints and the sentiment pertained while mak-
ing such a tweet. The method proposed outlines to find the optimum cluster-heads
from the Twitter dataset’s sentimental contents. The model tested its efficacy on var-
ious Twitter datasets and then compared it with the existing methods such as particle
swarm optimization, differential evolution, cuckoo search, improved cuckoo search,
10 Sudhanshu Kumar∗1 et al.

etc. This research work performed a basis for designing a system that quickly pro-
vides conclusive reviews on any social issues.
The authors [90] discussed an approach where a publicized tweets from the Twitter
site are processed and classified based on their polarity. In this paper, a new tool is
called ”SENTICIRCLE” is used a lexicon-based approach. The word’s semantics are
extracted from its co-occurrence pattern, and the strength is updated in the lexicon
accordingly. The basic idea of the approach is that the group of word accompany-
ing it decides the semantic of the word in any text. The force of movement from the
static word sentiment orientation approach to this contextual approach derived from
the dictum ”YOU SHALL KNOW THE WORD BY THE COMPANY IT KEEPS.”
It is different from the traditional lexicon, where the words are given fixed static se-
mantics regardless of the context.

2.1.5 Microblogging Data of Non-English Text

Many studies have been conducted in sentiment analysis on English texts, while other
languages have less attention than Arabic, Hindi, Bangla, etc. Many researchers have
worked on SA in different languages after the rise of Web 2.0. The author in [3] intro-
duced an overview on Arabic assessment analysis [8, 4, 97] in which they examined
various tools and applications pertinent to it. The study additionally included both
corpus-based and dictionary-based approaches for different datasets. Microblogs like
tweets are trending rapidly for online users to share their experiences and opinions
daily. In contrast to the online reviews and blogs, these microblogs contain very dis-
persed and incomplete data. Unlike English-based microblogs, Chinese microblogs
such as Sina Weibo have less sentiment analysis. The reason being that Chinese tex-
tual analysis is more challenging than English as its grammar of expression is differ-
ent. The same length of Chinese sentences may contain more data than English, and
the separation of words in those texts is relatively obscure. In totality, textually ana-
lyzing Chinese blogs has three primary research goals: First, the new words mining
and their sentiment inference; second, how to extract other media modules and third,
establish a hierarchical sentiment detection method based on Sina Weibo linguistics.
The authors [110] proposed three primary goals for the analysis of these Chinese
microblogs. They visualize the sentiment analysis’s result, depicting the relationship
between social network sentiments and real-life events.
The researchers already working on Chinese microblogs tend to analyze the topic fo-
cussing on a single attribute while neglecting others. The model design is multilevel
in single-level features keeping all the aspects under consideration. Chen et al. [21]
used to extract the text’s sentiment using sentence-level sentiment analysis, but un-
like other traditional approaches where the same technique was used in all types of
sentences. The sentences are classified into three groups based upon their opinion tar-
gets. There are other ways to classify the sentences that have been previously used in
other research papers. For example, The sentence can be subjective or objective based
upon the subjectivity of the sentences. The subjective sentences express the opinions,
while objective sentences implicate opinions or sentiments. The opinionated targets
focused on the primary sentence classification. This opinion target can be any en-
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 11

tity on which opinion is expressed. These opinionated sentences can give an opinion
without mentioning the target on three different types of sentences: non-target, one-
target, and multi-target. The Bi-LSTM and CNN deep learning approaches were used
to classify the sentences and extract the text’s syntactic and semantic features.

2.2 Sentiment in Speech

Analysis of speech in search of emotional and affective cues has a comparably long
tradition [28]. This paper proposed statistical pattern recognition techniques to clas-
sify 1000 utterances according to their emotional content. Meanwhile, several kinds
of literature have been established, including a range of recent surveys in emotions
and affect in speech [95]. However, targeting sentiment explicitly exclusively from
spoken utterances is a comparably new field than text-based sentiment analysis. Fo-
cusing on the acoustic side of spoken language, the border between sentiment and
emotion analysis is often fragile, as discussed in [22]. Mairesse et al. [63] focused
on pitch-related features and observed that pitch contains information on sentiment
without textual cues. The authors collected short-spoken reviews from 84 speakers,
and the result outperformed a majority class baseline. This paper attracted other re-
searchers to explore this area to solve real-world problems. The authors [34] created
Arabic Speech Act and Sentiment (ArSAS) dataset. The dataset consisted of 21,064
tweets annotated for two tasks: speech act recognition and SA. Further, the tweets
are annotated for four different sentiment categories: positive, negative, neutral, and
mixed.
Ahmed et al. [2] showed the sentiment in phone calls by first using speech recogni-
tion to extract the text in the call and then use typical text-based SA techniques. The
goal was to measure agent productivity in call centers.

2.3 Image based Sentiment Analysis

Vision-based emotion recognition [123, 92, 19] is a relatively recent area of research.
Users share millions of images and videos over social media platforms like Twit-
ter, Tumblr, Flickr, and Instagram. These are the most popular sites where celebri-
ties from sports, entertainment, and politics field share information in images. In the
image-based sentiment analysis, opinions depict in the form of cartoons or memes.
In most cases, the information conveyed through images is more effective compared
to other modalities. Multiple techniques and algorithms such as SVM, naive Bayes,
maximum entropy, and deep learning have been proposed in the image-based senti-
ment area to get significant results. The first work introduced by [65] to classify the
images into positive and negative. The author showed that there is a strong correlation
between sentiment images and their visual content.
Further, the SentiWordNet lexicon [74] was used to find the text’s numerical scores
associated with the image. This lexicon is used in WorldNet databases to identify
the positive and negative sentiment of the word. Emotions are difficult to identify
and pin down when discussing the state of the emotion that differentiates from other
12 Sudhanshu Kumar∗1 et al.

emotional states. To find the scientific approach regarding the emotional state of the
human being. The database of different photos was collected and validated against
the specific emotional response of the viewers. This database is called International
Affective Picture System (IAPS). Mikels et al. [65] studied eight emotion output cat-
egories: awe, anger, amusement, contentment, excitement, disgust, sadness, and fear.
The author showed that each emotional state is different as different emotions have
other cognitive and behavioral consequences. This paper adds some new dimensions
of data for IAPS.

2.4 Multimodal Sentiment Analysis

Multimodal sentiment analysis [56] performs sentiment analysis from multiple data
such as audio, video, and text. It is the new dimension of traditional text-based
sentiment analysis. Poria et al. [82] proposed a new multimodal sentiment analysis
methodology, which outperformed state of the art by more than 20%. The proposed
system used feature-based fusion techniques on text, visual and audio data from the
youtube dataset. Different classifiers such as Naive Bayes, SVM, and extreme learn-
ing machines are implemented on the youtube dataset. The results showed that the
extreme learning classifier is better than other classifiers. Extreme learning classifiers
have single layer or multiple layers of hidden nodes. In most cases, the weights of the
hidden nodes are learned in a single step so the overall processing time to classifying
the result is less.
Kumar et al. [54] proposed a multimodal rating prediction framework for prod-
ucts to improve customer satisfaction. The forty participants were participated in this
study to collect EEG data of the product. The text’s reviews from the product are
processed through NLP techniques. The customer’s rating from EEG and the prod-
uct’s reviews fused through optimization techniques. The experiment result showed
that the ABC optimization approach was better than the unimodal scheme. There
are many languages other than English, where researchers are working to predict the
sentiment [116, 120, 12, 125, 113]. In [52], the authors proposed a hybrid approach
on the Arabic dataset (Text and audio). Two machine learning approaches were used
on this dataset to find the polarity. The bagging and boosting algorithms were used
to enhance the proposed system further. The summary of the reviewed articles is as
shown in Table 1.
There are many public datasets [41, 105] available in SA. For different application
such as text, images, audio and video, we use different datasets or create a dataset like
Sanskrit dataset [53]. The tools and software library is also depend on the multimodal
data. Open CV is a open source library used in computer vision tasks for object
dection, face recognition and image segmentation. NLTK is a python library used for
understanding the text or speech. Below is some popular database as shown in Table
2.
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 13

Table 1: Summary of the publication’s details included author, approach, data set, and
accuracy.
Author & Year Approach Dataset Accuracy (%)
Hearst et al. [14], 1992 Cognitive Linguistics User’s Query -
Wiebet et al. [21], 1999 Probabilistic Classifier Gold-standard -
Camera reviews,
Nasukawa et al. [22], 2003 Sentiment Lexicon 75-95
news articles
Supervised and Unsupervised
Pang et al. [23], 2008 - -
approach (Survey paper)
Electronics reviews,
Domain-Oriented
Tan et al. [11], 2011 Stock reviews 82.9
sentiment lexicon
and Hotel review
sentiment polarity amazon product
Fang et al. [9], 2015 80
categorization process reviews
Amazon
Trust-based ranking
Gallege et al. [8], 2016 marketplace -
and the recommendation
review
Naive Bayes, linear SVM, Multi-Domain and
Xia et al. [10], 2015 90
logistic regression Chinese dataset
1500 Arabic
Khasawneh et al. [12], 2015 Bagging and Boosting comments and -
Twitter reviews
Campos et al. [65], 2017 CNN Twitter images -
Poria et al. [13], 2016 ELM classifier YouTube Dataset -
Penn Treebank,
Cambria et al. [1], 2017 Top-Down and Bottom-Up -
LIWC
EEG data and
Kumar et al. [4], 2019 ABC optimization -
product reviews
Expectancy disconfirmation LinkedIn and
Qazi et al. [47], 2017 theory and Confirmatory the university mail -
factor analysis servers’ groups
Raw Data and
Wang et al. [48], 2018 SentiRelated 80
Douban Data
intermediate-level
Williams et al. [48], 2018 MOSI dataset 74.0
feature fusion
Accuracy 93.5
book reviews
Precision 93
Yang et al. [48], 2020 SLCABG collected from
Recall 93.6
Dangdang dataset
F1 93.3
Precision 91.54
15000 hotel
Xu et al. [48], 2019 Seninfo+TF-IDF Recall 92.82
comment texts
F1 92.18
Multimodal Corpus of
Lakomkin et al. [48], 2019 ASR model 73.6
Sentiment Intensity
CNN-BiGRU-CTC + Aishell-1
Guo et al. [48], 2022 94.5
ERNIE-BiLSTM and NLPCC 2014
Kumar et al. [48], 2022 BiLSTM + GloVe IIT-R STSA 92.83
BERT-LARGE + A-KVMN LAP14, REST14,
Tian et al. [48], 2021 with second-order REST15, REST16 92.48
word dependencies and Twitter
Movie review,
Behera et al. [48], 2021 Co-LSTM model Airline dataset 98.40
Self driving car GOP
Facebook corpus Precision 57.95
Attention-based
Zhao et al. [48], 2020 containing user Recall 65.78
LSTM model
personality tag F1 72.2
English dataset
average 56.24
Derakhshan et al. [48], 2019 LDA-POS model English and Persian
Persian dataset
average 55.33

3 Usage and Application of Sentiment Analysis

This section covers the wide range of applications of sentiment analysis in various
emerging areas.

3.1 Reviews from E-commerce and Microblogging Sites

We have an extensive collection of data sets available on almost everything over

the internet. It includes user comments, reviews, feedback on various topics, opin-
ions drawn using surveys, products on e-commerce websites [43], customer ser-
vices [50], and recently Twitter data in the form of tweets on ongoing COVID-19
14 Sudhanshu Kumar∗1 et al.

Table 2: List of most popular public datasets available in the sentiment analysis field
in different languages and modalities. The details included the source and uses of the
dataset like the movie reviews, product reviews, social media data, etc.
Research Paper Dataset Use of this dataset Volume
Bai et al. [10] IMDB Movie To analysis movie review 50,000 movie reviews
Google 640917,
To do the sentiment analysis Microsoft 161292
Li et al. [60] Twitter
of tweets of different domain and Sony 141529
tweets
Sentiment analysis of tweets
Araque et al. [7] Sentiment140 1.6 million tweets
for a product or brand
Amazon Product To classify user review in
Dredze et al. [15] 142.8 million review
Reviews positive and negative
To find the aspect based 3 million restaurant
Qian et al. [58] Restaurant Reviews
sentiment analysis reviews
To do the sentiment analysis million Facebook users
Karyotis et al. [51] Facebook
of Facebook post and their posts
470 positive tweets
Chen et al. [115] Flicker images To classify the image and 133 negative tweets,
Tumblr 1179
IAPS 395,
Yang et al. [119] IAPS, Instagram visual sentiment prediction
Instagram 23308
Restaurants 3841,
Yang et al. [118] SemEval 2014 Aspect-based sentiment analysis
Laptops 3845
Positive 3094,
Zhang et al. [124] SemEval 2016 Sentiment analysis track Negative 2043
and Neutral 863 tweets
Detecting sentiment,
Schmitt et al. [94] SemEval 2017 8000-10000 tweets
humour, and truth
Joshi et al. [48] Hindi Movie Reviews Sentiment analysis in Hindi 250 Hindi Movie Reviews
Reviews of hotel
Chinese Text Sentiment
Xu et al. [116] clothes, fruit 2,50000 reviews
Analysis
digital etc
The Multimodal
Stappen et al. [99] MuSe-CAR Sentiment Analysis 15 GB Audio, Video, Text
in Car Reviews
4 emotions: angry,
Latif et al. [57] URDU-Dataset 0.072 GB Audio
happy, neutral, and sad.
6 emotions provides
single-word utterances
Duville et al. [33] MESD 0,097 GB Audio
for anger, disgust, fear
happiness, neutral, and sadness.

[1, 64, 5, 20, 72] pandemic. Therefore, there is a severe demand for a system based
on sentiment analysis that can extract sentiments about a particular product, item, or
service. It will help us to automate the user feedback or customer rating for the given
product, services, etc. This would help to improve the product and offered services
and eventually serve both the buyer and seller’s requirements.

3.2 Business Intelligence

Nowadays, consumers are getting more intelligent [98], quality-conscious, and tech-
nical savvy; therefore, they tend to seek out the reviews and ratings of online prod-
ucts and services before buying them. Many companies like Uber [11], Oyo [96],
and zomato [42] use digital transformation models to take feedback from the cus-
tomer. The online customer opinion decides the success or failure of their offered
services and products. The companies demand to extract sentiment from the online
user reviews to enhance their offered products and services. It also helps companies to
launch their new products and services in the new market for target customers. There-
fore, It is evident that sentiment analysis plays a vital role in getting the customer and
competition insights, which help companies make corrective and preventive actions
to sustain and grow their businesses in the digital era.
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 15

3.3 Global Financial Market

SA is also helpful in the share market and the Federal open market committee (FOMC
2
) statement [101, 14, 31] to extract the meaningful information for traders through
which they understand the global financial markets 3 . Some interesting trends reveal
in Figure 6, Figure 7 and Figure 8 through sentiment analysis of FOMC statements.

Fig. 6: The federal open market committee (FOMC) controls the monetary policy of
the central bank. The FOMC’s statement lexical frequency list (most popular word in
the report) of July and September 2017.

3.4 Applications in Smart Homes

Smart homes are an emerging technology, and in the near future, the entire home will
be more secure and better connected with other home appliances. The people would
control and manage any part of the house using smart wearable devices such as apple
watch, intelligent assistant devices such as Alexa [16, 37], Google Home [91, 78],
etc. Recently there has been a lot of research going on in the Internet of Things (IoT)
and the SA. The SA also found its way in IoT, e.g., the connected home using smart
devices such as smart bulbs, smart music devices could alter its ambiance to create a
calming and comfortable environment based on the sentiment or emotion of the user.

2 https://www.federalreserve.gov/monetarypolicy/fomccalendars.htm
3 https://www.forbes.com/sites/alapshah/2017/09/22/sentiment-analysis-of-fomc-statements-reveals-a-
more-hawkish-fed/?sh=4079d5a8632e
16 Sudhanshu Kumar∗1 et al.

Fig. 7: In the list of positive and negative words in the FOMC’s statement of July
and September 2017, the words in bold font denote the negative word while the other
indicates the positive word.

3.5 Detection of Hate Speech

Bigotry speech [40, 9, 88] is used to express repugnance towards a specifically in-
tended community, group, or person that can cause a dangerous situation to the vic-
tim. It can also be used to demean or offend particular community members or groups
on any social media. SA based detection system would help the social media compa-
nies such as Twitter [47], Instagram [71], etc., instant messaging companies such as
WhatsApp [27], Telegram and local enforcement and government to suppress hate
speech and fake news towards a specific person, sex, religion, race, country, etc.
which in turn improve their reputation and bring harmony in the community.

3.6 Emotion detection in suicide notes

In modern society, suicides are rising rapidly in recent times; it is critical to find a
faster way to fine-grained emotion detection [38, 29, 84] and anxiety in online posts,
microblogging text in the form of tweets by these troubled individuals. The SA-based
detection and analysis system may help to detect such tendencies upfront and prevent
suicides.

3.7 Stress Detection

On the flip side of excessive competition, improving the living style in a fast-moving
world, people typically face many changes from their work environment, eating habits,
etc. The body reacts to these stress changes, influencing an individual’s emotional,
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 17

Fig. 8: Most commonly used words in the FOMC statements since 2012, where n is
the occurrence of words. In this chart, the most frequent word has shown from top to
bottom; the top word in the chart has a higher occurrence than the last word.

mental, and physical health. The SA based detection system may help to detect stress
symptoms [109, 49] upfront and prevent any adverse impact due to this.
18 Sudhanshu Kumar∗1 et al.

4 Challenges and Perspectives

The SA is a particularly challenging task for human behaviors and subjective senti-
ments. Below are a few of the challenges-

4.1 Recognizing Subjective Parts of The Phrase

The English language can sometimes be tricky. Homonyms, or multiple-meaning

words, have the same spelling and usually sound alike but have different meanings.
Subjective parts in the phrase or sentence epitomize sentiment-related content. The
Homonyms in the phrase might be treated as subjective in one case or objective in
some other. It brings it challenging to identify the subjective portions of the phrase.
For example: 1. The new lamp had good light for reading. 2. Magnesium is a light
metal. The word light is used to mean a particular quality or type of light in the first
phase, whereas the light word objectively means having a relatively low density in
the second phrase. Users share views or opinions over the internet on different social
media platforms. Different age groups and gender share information or opinion in
their way, recent study [55] prove that older people people share their opinion in a
better way instead of young ones.

4.2 Dependence on The Domains

The same phrase might have different interpretations in different domains in which it
is being used. For Example, the word ’unpredictable’ is positive in entertainment and
theater, etc., but if the same word is used in the context of an automobile’s break, it
has a negative opinion. Still, this is challenging to identify the domain from which any
word is related correctly. Different pre-trained word embedding corpus domains such
as IMDB movie reviews corpus and customer reviews dataset classify the sentence
correctly. This challenge is still not solved completely, and researchers are continu-
ously working on this problem.

4.3 Detection of Sarcasm in The Phrase

Sarcastic sentences express a negative opinion about a person or thing using positive
words in unique. Often, people use it to say the opposite of what’s true to make
someone look or feel foolish. For Example: -” Good perfume. You must marinate
in it for long”. The sentence has only positive words, but it expresses a negative
sentiment.

4.4 Dependence on The Order

Discourse Structure analysis is essential for opinion mining and sentiment analysis.
For Example, A is better than B conveys the exact opposite opinion from B is better
than A. For finding SA for these kind of sentence is quite challenging.
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 19

4.5 Idioms

ML programs are designed so that they don’t understand a figure of speech. For ex-
ample, language such as ”not my cup of tea” will disrupt the algorithm because it
understands the things literally. When any user uses idioms in a comment or review,
the sentence interpretation is not correctly map by the algorithm. The situation is even
more difficult if the comment is multilingual.

4.6 Multilingual sentiment analysis

User share their opinion in different languages like Hinglish which is the combina-
tion of Hindi and English. Every language has its own lemmatizer, POS tagger and
grammatical constructs so ML or Deep learning algorithm understand the context
and classify the comment in positive and negative. The real challenges is that we can
not translate multiple language into one base language. Usually in micro-blogging or
chatting, user share their feeling in multilingual.
Despite different challenges in sentiment analysis still, it is an emerging field
among customers for decision-making. Figure 9 and 10 presents users’ most popular
topic and query search from 2004 to 2020.

Fig. 9: The top 25 topics search by the user worldwide from 2004 to 2020 in the
sentiment analysis field. The most frequent topics search by the users is analysis and
opinion.
20 Sudhanshu Kumar∗1 et al.

Fig. 10: The most frequently searched query worldwide from 2004 to 2020 in the
sentiment analysis field. The sentiment, sentiment analysis, and Twitter sentiment
are the top three search queries worldwide.

5 Discussion Towards ML and DL Techniques on Sentiment Analysis Field

In the last decade, the paradigm shifted from machine learning to deep learning tech-
niques. In-text data, the context problem is a big challenge to understand the sen-
tence’s meaning through the ML algorithm correctly. This problem solves through
pre-trained word embedding and the VADER approach even we have a smaller train-
ing dataset. However, the pre-trained word embedding corpus was trained on the
google news dataset (100 billion words) and IMDB movie dataset. It shows a good
result when the data are related to the pre-trained corpus domain; otherwise, it will
not predict the result as expected. The BERT model is the start of the art model in
NLP. It uses the bidirectional training of the input, which provides a more profound
sense of the language context. However, it is very compute-intensive and takes time
to predict the result. ML techniques are also predicted good results, depending on the
dataset and nature of data. DL techniques predict a good outcome for a large dataset.
The present study covered different domains like text, speech, image, and video
to analyze sentiment. In all domains, the start of the art algorithms and papers were
discussed in the study.
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 21

6 Conclusion and Future Work

With the advancement of technology in machine learning and deep learning, the SA
plays a vital role in analyzing data available on the internet in text, image, and speech.
The SA is computationally identifying the polarity of text into a positive, negative,
and neutral review. In this survey paper, we have investigated the history of the SA
and its impact on the research community from the years 2000 to current trends. In the
last five years, most articles are related to social media such as Facebook, Instagram,
and Twitter. Most articles are related to the application area of health, restaurant,
travel, spam, and politics. We have also included the top-cited paper and discuss the
research challenges and perspectives suitable for new researchers who want to start
research in the ML, NLP, and SA fields. We also cover in detail about global fina-
cial market (FOMC), different languages like Sanskrit, Hindi, Arabic, Chinese etc
and modalities in which many authors used SA. In future work, the SA is combined
with network traffic to detect fake opinion or news, which creates a serious prob-
lem, resulting in mob violence. The method to do the SA will also improve with the
continuous advancement of the NLP and ML fields.

Compliance with ethical standards

Conflict of interest The authors declared that they have no conflicts of interest to this
work.

References

1. Abd-Alrazaq A, Alhuwail D, Househ M, Hamdi M, Shah Z (2020) Top con-

cerns of tweeters during the covid-19 pandemic: infoveillance study. Journal of
medical Internet research 22(4):e19016
2. Ahmed A, Toral S, Shaalan K (2016) Agent productivity measurement in call
center using machine learning. In: International Conference on Advanced Intel-
ligent Systems and Informatics, Springer, pp 160–169
3. Al-Ayyoub M, Khamaiseh AA, Jararweh Y, Al-Kabi MN (2019) A comprehen-
sive survey of arabic sentiment analysis. Information processing & management
56(2):320–342
4. Al-Radaideh QA, Al-Qudah GY (2017) Application of rough set-based feature
selection for arabic sentiment analysis. Cognitive Computation 9(4):436–445
5. Alamoodi AH, Zaidan BB, Zaidan AA, Albahri OS, Mohammed K, Malik RQ,
Almahdi EM, Chyad MA, Tareq Z, Albahri AS, et al (2021) Sentiment analysis
and its applications in fighting covid-19 and infectious diseases: A systematic
review. Expert systems with applications 167:114155
6. Anto MP, Antony M, Muhsina KM, Johny N, James V, Wilson A (2016) Product
rating using sentiment analysis. In: Proceedings of the International Conference
on Electrical, Electronics, and Optimization Techniques (ICEEOT), pp 3458–
3462
22 Sudhanshu Kumar∗1 et al.

7. Araque O, Corcuera-Platas I, Sánchez-Rada JF, Iglesias CA (2017) Enhancing

deep learning sentiment analysis with ensemble techniques in social applica-
tions. Expert Systems with Applications 77:236–246
8. Badarneh O, Al-Ayyoub M, Alhindawi N, Jararweh Y, et al (2018) Fine-grained
emotion analysis of arabic tweets: A multi-target multi-label approach. In: 2018
IEEE 12th International Conference on Semantic Computing (ICSC), IEEE, pp
340–345
9. Badjatiya P, Gupta S, Gupta M, Varma V (2017) Deep learning for hate speech
detection in tweets. In: Proceedings of the 26th international conference on
World Wide Web companion, pp 759–760
10. Bai X (2011) Predicting consumer sentiments from online text. Decision Sup-
port Systems 50(4):732–742
11. Baj-Rogowska A (2017) Sentiment analysis of facebook posts: The uber case.
In: 2017 Eighth International Conference on Intelligent Computing and Infor-
mation Systems (ICICIS), IEEE, pp 391–395
12. Behera RK, Jena M, Rath SK, Misra S (2021) Co-lstm: Convolutional lstm
model for sentiment analysis in social big data. Information Processing & Man-
agement 58(1):102435
13. Bengio Y, Ducharme R, Vincent P, Jauvin C (2003) A neural probabilistic lan-
guage model. Journal of machine learning research 3(Feb):1137–1155
14. Bhandari P (2022) Sentiment analysis of fomc meeting transcripts: Pre and post
mexican pesos crisis
15. Blitzer J, Dredze M, Pereira F (2007) Biographies, bollywood, boom-boxes and
blenders: Domain adaptation for sentiment classification. In: Proceedings of the
45th annual meeting of the association of computational linguistics, pp 440–447
16. Bogdan R, Tatu A, Crisan-Vida MM, Popa M, Stoicu-Tivadar L (2021) A
practical experience on the amazon alexa integration in smart offices. Sensors
21(3):734
17. Cambria E, Poria S, Bajpai R, Schuller B (2016) Senticnet 4: A semantic re-
source for sentiment analysis based on conceptual primitives. In: Proceedings
of COLING 2016, the 26th international conference on computational linguis-
tics: Technical papers, pp 2666–2677
18. Cambria E, Poria S, Gelbukh A, Thelwall M (2017) Sentiment analysis is a big
suitcase. IEEE Intelligent Systems 32(6):74–80
19. Campos V, Jou B, Giro-i Nieto X (2017) From pixels to sentiment: Fine-tuning
cnns for visual sentiment prediction. Image and Vision Computing 65:15–22
20. Chakraborty K, Bhatia S, Bhattacharyya S, Platos J, Bag R, Hassanien AE
(2020) Sentiment analysis of covid-19 tweets by deep learning classifiers—a
study to show how popularity is affecting accuracy in social media. Applied
Soft Computing 97:106754
21. Chen T, Xu R, He Y, Wang X (2017) Improving sentiment analysis via sentence
type classification using bilstm-crf and cnn. Expert Systems with Applications
72:221–230
22. Crouch S, Khosla R (2012) Sentiment analysis of speech prosody for dialogue
adaptation in a diet suggestion program. ACM SIGHIT Record 2(1):8–8
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 23

23. Das S, Chen M (2001) Yahoo! for amazon: Extracting market sentiment from
stock message boards. In: Proceedings of the Asia Pacific finance association
annual conference (APFA), Bangkok, Thailand, vol 35, p 43
24. Das SR, Chen MY (2007) Yahoo! for amazon: Sentiment extraction from small
talk on the web. Management science 53(9):1375–1388
25. Dasgupta SS, Natarajan S, Kaipa KK, Bhattacherjee SK, Viswanathan A (2015)
Sentiment analysis of facebook data using hadoop based open source technolo-
gies. In: Proceedings of the IEEE International Conference on Data Science and
Advanced Analytics (DSAA), pp 1–3
26. Day MY, Lee CC (2016) Deep learning for financial sentiment analysis on fi-
nance news providers. In: Proceedings of the IEEE/ACM International Confer-
ence on Advances in Social Networks Analysis and Mining (ASONAM), pp
1127–1134
27. Deb K, Paul S, Das K (2020) A framework for predicting and identifying radi-
calization and civil unrest oriented threats from whatsapp group. In: Emerging
Technology in Modelling and Graphics, Springer, pp 595–606
28. Dellaert F, Polzin T, Waibel A (1996) Recognizing emotion in speech. In: Pro-
ceeding of Fourth International Conference on Spoken Language Processing.
ICSLP’96, IEEE, vol 3, pp 1970–1973
29. Desmet B, Hoste V (2013) Emotion detection in suicide notes. Expert Systems
with Applications 40(16):6351–6358
30. Devlin J, Chang MW, Lee K, Toutanova K (2018) Bert: Pre-training of
deep bidirectional transformers for language understanding. arXiv preprint
arXiv:181004805
31. Doh T, Kim S, Yang SK (2021) How you say it matters: Text analysis of fomc
statements using natural language processing. Economic Review-Federal Re-
serve Bank of Kansas City 106(1):25–40
32. Du C, Huang L (2018) Text classification research with attention-based recur-
rent neural networks. International Journal of Computers Communications &
Control 13(1):50–61
33. Duville MM, Alonso-Valerdi LM, Ibarra-Zarate DI (2021) The mexican emo-
tional speech database (mesd): elaboration and assessment based on machine
learning. In: 2021 43rd Annual International Conference of the IEEE Engineer-
ing in Medicine & Biology Society (EMBC), IEEE, pp 1644–1647
34. Elmadany A, Mubarak H, Magdy W (2018) Arsas: An arabic speech-act and
sentiment corpus of tweets. OSACT 3:20
35. Fang X, Zhan J (2015) Sentiment analysis using product review data. Journal
of Big Data 2(1):5
36. Gallege LS, Raje RR (2016) Towards selecting and recommending online soft-
ware services by evaluating external attributes. In: Proceedings of the 11th An-
nual Cyber and Information Security Research Conference (CISRC), pp 23:1-
23:4
37. Gao Y, Pan Z, Wang H, Chen G (2018) Alexa, my love: analyzing reviews of
amazon echo. In: 2018 IEEE SmartWorld, Ubiquitous Intelligence & Comput-
ing, Advanced & Trusted Computing, Scalable Computing & Communications,
Cloud & Big Data Computing, Internet of People and Smart City Innovation
24 Sudhanshu Kumar∗1 et al.

(SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI), IEEE, pp 372–380

38. Ghosh S, Ekbal A, Bhattacharyya P (2022) Deep cascaded multitask framework
for detection of temporal orientation, sentiment and emotion from suicide notes.
Scientific reports 12(1):1–16
39. Giatsoglou M, Vozalis MG, Diamantaras K, Vakali A, Sarigiannidis G,
Chatzisavvas KC (2017) Sentiment analysis leveraging emotions and word em-
beddings. Expert Systems with Applications 69:214–224
40. Gitari ND, Zuping Z, Damien H, Long J (2015) A lexicon-based approach for
hate speech detection. International Journal of Multimedia and Ubiquitous En-
gineering 10(4):215–230
41. Guo H, Zhan X, Chi C (2022) Multiple scene sentiment analysis based on chi-
nese speech and text. Journal of Computers 33(1):165–178
42. Gupta R, Sameer S, Muppavarapu H, Enduri MK, Anamalamudi S (2021) Sen-
timent analysis on zomato reviews. In: 2021 13th International Conference on
Computational Intelligence and Communication Networks (CICN), IEEE, pp
34–38
43. Haque TU, Saber NN, Shah FM (2018) Sentiment analysis on large scale ama-
zon product reviews. In: 2018 IEEE international conference on innovative re-
search and development (ICIRD), IEEE, pp 1–6
44. Hearst MA (1992) Direction-based text interpretation as an information access
refinement. Text-based intelligent systems: Current research and practice in in-
formation extraction and retrieval pp 257–274
45. Hochreiter S, Schmidhuber J (1997) Long short-term memory. Neural compu-
tation 9(8):1735–1780
46. Hutto CJ, Gilbert E (2014) Vader: A parsimonious rule-based model for senti-
ment analysis of social media text. In: Eighth international AAAI conference
on weblogs and social media
47. Jiang L, Suzuki Y (2019) Detecting hate speech from tweets for sentiment anal-
ysis. In: 2019 6th International Conference on Systems and Informatics (IC-
SAI), IEEE, pp 671–676
48. Joshi A, Balamurali A, Bhattacharyya P, et al (2010) A fall-back strategy for
sentiment analysis in hindi: a case study. Proceedings of the 8th ICON
49. Jung H, Park HA, Song TM, et al (2017) Ontology-based approach to social
data sentiment analysis: detection of adolescent depression signals. Journal of
medical internet research 19(7):e7452
50. Kang D, Park Y (2014) based measurement of customer satisfaction in mobile
service: Sentiment analysis and vikor approach. Expert Systems with Applica-
tions 41(4):1041–1050
51. Karyotis C, Doctor F, Iqbal R, James A, Chang V (2018) A fuzzy computa-
tional model of emotion for cloud based sentiment analysis. Information Sci-
ences 433:448–463
52. Khasawneh RT, Wahsheh HA, Alsmadi IM, AI-Kabi MN (2015) Arabic senti-
ment polarity identification using a hybrid approach. In: 2015 6th International
Conference on Information and Communication Systems (ICICS), IEEE, pp
148–153
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 25

53. Kumar P, Pathania K, Raman B (2022) Zero-shot learning based cross-lingual

sentiment analysis for sanskrit text with insufficient labeled data. Applied Intel-
ligence pp 1–18
54. Kumar S, Yadava M, Roy PP (2019) Fusion of eeg response and sentiment
analysis of products review to predict customer satisfaction. Information Fusion
52:41–52
55. Kumar S, Gahalawat M, Roy PP, Dogra DP, Kim BG (2020) Exploring impact
of age and gender on sentiment analysis using machine learning. Electronics
9(2):374
56. Lakomkin E, Zamani MA, Weber C, Magg S, Wermter S (2019) Incorporating
end-to-end speech recognition models for sentiment analysis. In: 2019 Interna-
tional Conference on Robotics and Automation (ICRA), IEEE, pp 7976–7982
57. Latif S, Qayyum A, Usman M, Qadir J (2018) Cross lingual speech emotion
recognition: Urdu vs. western languages. In: 2018 International Conference on
Frontiers of Information Technology (FIT), IEEE, pp 88–93
58. Lei X, Qian X, Zhao G (2016) Rating prediction based on social sentiment from
textual reviews. IEEE transactions on multimedia 18(9):1910–1921
59. Levallois C (2013) Sentiment analysis for tweets based on lexicons an heuristics
60. Li YM, Li TY (2013) Deriving market intelligence from microblogs. Decision
Support Systems 55(1):206–217
61. Ma Y, Cambria E, Gao S (2016) Label embedding for zero-shot fine-grained
named entity typing. In: Proceedings of COLING 2016, the 26th International
Conference on Computational Linguistics: Technical Papers, pp 171–180
62. Ma Y, Peng H, Cambria E (2018) Targeted aspect-based sentiment analysis via
embedding commonsense knowledge into an attentive lstm. In: Thirty-second
AAAI conference on artificial intelligence
63. Mairesse F, Polifroni J, Di Fabbrizio G (2012) Can prosody inform sentiment
analysis? experiments on short spoken reviews. In: 2012 IEEE International
Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, pp
5093–5096
64. Manguri KH, Ramadhan RN, Amin PRM (2020) Twitter sentiment analysis
on worldwide covid-19 outbreaks. Kurdistan Journal of Applied Research pp
54–65
65. Mikels JA, Fredrickson BL, Larkin GR, Lindberg CM, Maglio SJ, Reuter-
Lorenz PA (2005) Emotional category data on images from the international
affective picture system. Behavior research methods 37(4):626–630
66. Mikolov T, Karafiát M, Burget L, Černockỳ J, Khudanpur S (2010) Recurrent
neural network based language model. In: Eleventh annual conference of the
international speech communication association
67. Mikolov T, Chen K, Corrado G, Dean J (2013) Efficient estimation of word
representations in vector space. arXiv preprint arXiv:13013781
68. Mikolov T, Sutskever I, Chen K, Corrado GS, Dean J (2013) Distributed rep-
resentations of words and phrases and their compositionality. In: Advances in
neural information processing systems, pp 3111–3119
69. Morinaga S, Yamanishi K, Tateishi K, Fukushima T (2002) Mining product rep-
utations on the web. In: Proceedings of the eighth ACM SIGKDD international
26 Sudhanshu Kumar∗1 et al.

conference on Knowledge discovery and data mining, pp 341–349

70. Munezero MD, Montero CS, Sutinen E, Pajunen J (2014) Are they different?
affect, feeling, emotion, sentiment, and opinion detection in text. IEEE transac-
tions on affective computing 5(2):101–111
71. Naf’an MZ, Bimantara AA, Larasati A, Risondang EM, Nugraha NAS (2019)
Sentiment analysis of cyberbullying on instagram user comments. Journal of
Data Science and Its Applications 2(1):38–48
72. Naseem U, Razzak I, Khushi M, Eklund PW, Kim J (2021) Covidsenti: A large-
scale benchmark twitter data set for covid-19 sentiment analysis. IEEE Trans-
actions on Computational Social Systems 8(4):1003–1015
73. Nasukawa T, Yi J (2003) Sentiment analysis: Capturing favorability using nat-
ural language processing. In: Proceedings of the 2nd international conference
on Knowledge capture, pp 70–77
74. Ohana B, Tierney B (2009) Sentiment classification of reviews using sentiword-
net. In: 9th. it & t conference, vol 13, pp 18–30
75. Pandey AC, Rajpoot DS, Saraswat M (2017) Twitter sentiment analysis us-
ing hybrid cuckoo search method. Information Processing & Management
53(4):764–779
76. Pang B, Lee L (2008) Opinion mining and sentiment analysis. Foundations and
trends in information retrieval 2(1-2):1–135
77. Pang B, Lee L, Vaithyanathan S (2002) Thumbs up?: sentiment classification
using machine learning techniques. In: Proceedings of the ACL-02 conference
on Empirical methods in natural language processing-Volume 10, Association
for Computational Linguistics, pp 79–86
78. Park H, Kim JH (2018) Perception of virtual assistant and smart speaker: Se-
mantic network analysis and sentiment analysis. In: Proceedings of the Korean
Institute of Information and Commucation Sciences Conference, The Korea In-
stitute of Information and Commucation Engineering, pp 213–216
79. Pennington J, Socher R, Manning CD (2014) Glove: Global vectors for word
representation. In: Proceedings of the 2014 conference on empirical methods in
natural language processing (EMNLP), pp 1532–1543
80. Peters ME, Neumann M, Iyyer M, Gardner M, Clark C, Lee K, Zettle-
moyer L (2018) Deep contextualized word representations. arXiv preprint
arXiv:180205365
81. Poria S, Cambria E, Hazarika D, Vij P (2016) A deeper look into sarcastic
tweets using deep convolutional neural networks. In: Proceedings of COLING
2016, the 26th International Conference on Computational Linguistics: Techni-
cal Papers, pp 1601–1612
82. Poria S, Cambria E, Howard N, Huang GB, Hussain A (2016) Fusing audio,
visual and textual clues for sentiment analysis from multimodal content. Neu-
rocomputing 174:50–59
83. Poria S, Cambria E, Bajpai R, Hussain A (2017) A review of affective comput-
ing: From unimodal analysis to multimodal fusion. Information Fusion 37:98–
125
84. Prasad DK, Liu S, Chen SHA, Quek C (2018) Sentiment analysis using eeg
activities for suicidology. Expert Systems with Applications 103:206–217
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 27

85. Qazi A, Tamjidyamcholo A, Raj RG, Hardaker G, Standing C (2017) Assessing

consumers’ satisfaction and expectations through online opinions: Expectation
and disconfirmation approach. Computers in Human Behavior 75:450–460
86. Radford A, Narasimhan K, Salimans T, Sutskever I (2018) Improving language
understanding with unsupervised learning. Technical report, OpenAI
87. Ren Y, Wang R, Ji D (2016) A topic-enhanced word embedding for twitter
sentiment classification. Information Sciences 369:188–198
88. Rodriguez A, Argueta C, Chen YL (2019) Automatic detection of hate speech
on facebook using sentiment and emotion analysis. In: 2019 international con-
ference on artificial intelligence in information and communication (ICAIIC),
IEEE, pp 169–174
89. Sack W (1994) On the computation of point of view. In: AAAI, p 1488
90. Saif H, He Y, Fernandez M, Alani H (2016) Contextual semantics for sentiment
analysis of twitter. Information Processing & Management 52(1):5–19
91. Sánchez-Franco MJ, Arenas-Márquez FJ, Alonso-Dos-Santos M (2021) Using
structural topic modelling to predict users’ sentiment towards intelligent per-
sonal agents. an application for amazon’s echo and google home. Journal of
Retailing and Consumer Services 63:102658
92. Sariyanidi E, Gunes H, Cavallaro A (2014) Automatic analysis of facial affect:
A survey of registration, representation, and recognition. IEEE transactions on
pattern analysis and machine intelligence 37(6):1113–1133
93. Scherer KR (2005) What are emotions? and how can they be measured? Social
science information 44(4):695–729
94. Schmitt M, Steinheber S, Schreiber K, Roth B (2018) Joint aspect and polarity
classification for aspect-based sentiment analysis with end-to-end neural net-
works. In: Proceedings of the 2018 Conference on Empirical Methods in Natu-
ral Language Processing, pp 1109–1114
95. Schuller B, Batliner A, Steidl S, Seppi D (2011) Recognising realistic emotions
and affect in speech: State of the art and lessons learnt from the first challenge.
Speech Communication 53(9-10):1062–1087
96. Shanmugam S, Padmanaban I (2020) Twitter emotion analysis for brand com-
parison using naive bayes classifier. In: International Conference on Soft Com-
puting and its Engineering Applications, Springer, pp 199–211
97. Socher R, Perelygin A, Wu J, Chuang J, Manning CD, Ng AY, Potts C (2013)
Recursive deep models for semantic compositionality over a sentiment tree-
bank. In: Proceedings of the 2013 conference on empirical methods in natural
language processing, pp 1631–1642
98. Sreesurya I, Rathi H, Jain P, Jain TK (2020) Hypex: A tool for extracting
business intelligence from sentiment analysis using enhanced lstm. Multime-
dia Tools and Applications 79(47):35641–35663
99. Stappen L, Baird A, Schumann L, Bjorn S (2021) The multimodal sentiment
analysis in car reviews (muse-car) dataset: Collection, insights and improve-
ments. IEEE Transactions on Affective Computing
100. STONE J (1997) Thematic text analysis-new agendas for analyzing text con-
tent. Test analysis for the social sciences-Methods for drawing statistical infer-
ences from texts and transcripts pp 35–54
28 Sudhanshu Kumar∗1 et al.

101. Tadle RC (2022) Fomc minutes sentiments and their impact on financial mar-
kets. Journal of economics and business 118:106021
102. Tan S, Wu Q (2011) A random walk algorithm for automatic construc-
tion of domain-oriented sentiment lexicon. Expert Systems with Applications
38(10):12094–12100
103. Tang D, Wei F, Yang N, Zhou M, Liu T, Qin B (2014) Learning sentiment-
specific word embedding for twitter sentiment classification. In: Proceedings
of the 52nd Annual Meeting of the Association for Computational Linguistics
(Volume 1: Long Papers), pp 1555–1565
104. Thelwall M, Buckley K, Paltoglou G, Cai D, Kappas A (2010) Sentiment
strength detection in short informal text. Journal of the American society for
information science and technology 61(12):2544–2558
105. Tian Y, Chen G, Song Y (2021) Enhancing aspect-level sentiment analysis with
word dependencies. In: Proceedings of the 16th Conference of the European
Chapter of the Association for Computational Linguistics: Main Volume, pp
3726–3739
106. Tong RM (2001) An operational system for detecting and tracking opinions in
on-line discussion. In: Working Notes of the ACM SIGIR 2001 Workshop on
Operational Text Classification, vol 1
107. Turney PD, Littman ML (2003) Measuring praise and criticism: Inference of
semantic orientation from association. ACM Transactions on Information Sys-
tems (TOIS) 21(4):315–346
108. Wang L, Niu J, Song H, Atiquzzaman M (2018) Sentirelated: A cross-domain
sentiment classification algorithm for short texts through sentiment related in-
dex. Journal of Network and Computer Applications 101:111–119
109. Wang X, Zhang C, Ji Y, Sun L, Wu L, Bao Z (2013) A depression detection
model based on sentiment analysis in micro-blog social network. In: Pacific-
Asia Conference on Knowledge Discovery and Data Mining, Springer, pp 201–
213
110. Wang Z, Yu Z, Chen L, Guo B (2014) Sentiment detection and visualization
of chinese micro-blog. In: 2014 International Conference on Data Science and
Advanced Analytics (DSAA), IEEE, pp 251–257
111. Wiebe J, Bruce R, O’Hara TP (1999) Development and use of a gold-standard
data set for subjectivity classifications. In: Proceedings of the 37th annual meet-
ing of the Association for Computational Linguistics, pp 246–253
112. Wiebe J, et al (2000) Learning subjective adjectives from corpora. Aaai/iaai
20(0):0
113. Williams J, Comanescu R, Radu O, Tian L (2018) Dnn multimodal fusion tech-
niques for predicting video sentiment. In: Proceedings of grand challenge and
workshop on human multimodal language (Challenge-HML), pp 64–72
114. Xia R, Xu F, Zong C, Li Q, Qi Y, Li T (2015) Dual sentiment analysis: Con-
sidering two sides of one review. IEEE transactions on knowledge and data
engineering 27(8):2120–2133
115. Xu C, Cetintas S, Lee KC, Li LJ (2014) Visual sentiment prediction with deep
convolutional neural networks. arXiv pp arXiv–1411
A Comprehensive Review on Sentiment Analysis: Tasks, Approaches and Applications 29

116. Xu G, Yu Z, Yao H, Li F, Meng Y, Wu X (2019) Chinese text sentiment analysis

based on extended sentiment dictionary. IEEE Access 7:43749–43762
117. Yan C, Tu Y, Wang X, Zhang Y, Hao X, Zhang Y, Dai Q (2019) Stat: spatial-
temporal attention mechanism for video captioning. IEEE transactions on mul-
timedia
118. Yang C, Zhang H, Jiang B, Li K (2019) Aspect-based sentiment analysis
with alternating coattention networks. Information Processing & Management
56(3):463–478
119. Yang J, She D, Sun M, Cheng MM, Rosin PL, Wang L (2018) Visual sentiment
prediction based on automatic discovery of affective regions. IEEE Transactions
on Multimedia 20(9):2513–2525
120. Yang L, Li Y, Wang J, Sherratt RS (2020) Sentiment analysis for e-commerce
product reviews in chinese based on sentiment lexicon and deep learning. IEEE
access 8:23522–23530
121. Yang Z, Yang D, Dyer C, He X, Smola A, Hovy E (2016) Hierarchical attention
networks for document classification. In: Proceedings of the 2016 conference
of the North American chapter of the association for computational linguistics:
human language technologies, pp 1480–1489
122. Zadeh A, Liang PP, Poria S, Vij P, Cambria E, Morency LP (2018) Multi-
attention recurrent network for human communication comprehension. In:
Thirty-Second AAAI Conference on Artificial Intelligence
123. Zeng Z, Pantic M, Roisman GI, Huang TS (2008) A survey of affect recogni-
tion methods: Audio, visual, and spontaneous expressions. IEEE transactions
on pattern analysis and machine intelligence 31(1):39–58
124. Zhang Z, Zou Y, Gan C (2018) Textual sentiment analysis via three different at-
tention convolutional neural networks and cross-modality consistent regression.
Neurocomputing 275:1407–1415
125. Zhao J, Zeng D, Xiao Y, Che L, Wang M (2020) User personality prediction
based on topic preference and sentiment analysis using lstm model. Pattern
Recognition Letters 138:397–402