2024 IEEE 9th International Conference for Convergence in Technology (I2CT)

Pune, India. Apr 5-7, 2024

Exploring Progress in Aspect-based Sentiment

Analysis: An In-depth Survey
Vishal Singhi Charulata Chauhan PiyushKumar Soni
2024 IEEE 9th International Conference for Convergence in Technology (I2CT) | 979-8-3503-9447-4/24/$31.00 ©2024 IEEE | DOI: 10.1109/I2CT61223.2024.10543612

B Tech. in Information Technology B Tech. in Information Technology Assistant Professor, Information Technology
SVKM’s NMIMS, Shirpur SVKM’s NMIMS, Shirpur SVKM’s NMIMS, Shirpur
[email protected] [email protected] [email protected]

Abstract—Aspect-based sentiment analysis (ABSA), which fo- sentiment analysis at an aspect level allows for a more nuanced
cuses on extracting and assessing thoughts expressed toward understanding of customer opinions, preferences, and satisfaction re-
specific characteristics or entities inside of a given text, has garding different aspects of their domain experiences. By conducting
emerged as a result of Natural Language Processing (NLP). It a detailed sentiment analysis, proprietors of dining establishments
has become crucial to comprehend how individuals feel about can pinpoint strengths and areas needing refinement. This approach
multiple elements of goods, services, or activities for businesses enables focused enhancements, ultimately resulting in elevated overall
and scholars alike because to the ever-growing number of user- patron contentment.
generated content on the internet. Social media networks’ availability of user-generated material
This review paper provides an extensive overview of cutting- offers a wealth of data for sentiment analysis. Twitter, in particular,
edge techniques and methodologies employed in ABSA, with a is a popular platform for sharing concise opinions and experiences,
specific focus on restaurant reviews. We explore the key chal- making it a valuable resource for conducting sentiment analysis
lenges faced in this field, including aspect extraction, sentiment within the restaurant domain. Analyzing restaurant-related tweets at
classification, and domain adaptation, with a specific emphasis an aspect level can offer valuable insights that assist in decision-
on the unique aspects relevant to restaurant reviews. making processes, marketing strategies, and service enhancements.
These evaluations cover a wide range of topics, including This paper aims to explore sentiment analysis techniques applied
cost, ambience, meal quality, and service. The review provides a to different dataset at an aspect level. The results of this survey
thorough examination of cutting-edge methods with a focus on will help the hospitality industry better understand consumer attitudes
implicit aspect identification in restaurant reviews. and preferences, allowing restaurant owners and managers to make
As a result, this review aims to underscore the significance of educated choices for upgrading their food selections and increasing
detecting implicit aspects in ABSA, particularly in the context of customer happiness.
restaurant reviews. By presenting the advancements achieved in
ABSA for restaurant sentiment analysis and identifying research II. A SPECT BASED S ENTIMENT A NALYSIS
directions specific to this domain, we analyze different techniques This focuses on interpreting the emotions and attitudes depicted
of ABSA in this paper. in text, includes sentiment analysis as a crucial component. It is
Keywords- ASBS, Sentiment Analysis, Aspect Level, Encoder and important in many applications, including market research, customer
Decoder. feedback analysis, and social media tracking. Traditional sentiment
analysis, on the other hand, frequently offers a high-level assessment
I. I NTRODUCTION of the overall sentiment in a phrase or document but is unable to
In the era of social media, online platforms have become a treasure capture nuanced judgments on specific information or entities in the
trove of opinions and sentiments expressed by individuals across text.
various domains. Among these domains, the restaurant industry piyush kumar soni et al. [1], Aspect-level sentiment analysis, on
stands out as an area where customer experiences and opinions play the other hand, addresses this limitation by offering a more fine-
a vital role in shaping reputation and success. Analyzing customer grained approach. Instead of analyzing sentiments at the document
sentiments towards restaurants is crucial for understanding their pref- level, it aims to identify and extract sentiments related to particular
erences, identifying areas for improvement, and providing valuable features, attributes, or topics mentioned in the text. This approach
insights to restaurant owners and managers. Through an emphasis on allows for a deeper understanding of how people feel about specific
data about restaurants, this study intends to explore sentiment analysis components or aspects of a product, service, or any subject matter
in an aspect perspective. under consideration.
The computational study of obtaining irrational information, atti- Aspect-level sentiment analysis is particularly useful when dealing
tudes, and emotions from textual data is known as sentiment analysis with texts that contain multiple opinions on different aspects. It allows
or opinion mining. It entails the automatic classification and analysis a more detailed understanding of how people feel about various
of sentiments expressed in text using machine learning algorithms components or features of a product, service, or any subject matter,
and natural language processing (NLP) approaches. as shown in fig.1 . This information can be valuable for businesses
To automatically classify and analyze feelings in text, it makes to pinpoint specific strengths and weaknesses and make targeted
use of machine learning algorithms and NLP (natural language improvements based on customer feedback.
processing) techniques. While traditional sentiment analysis methods To perform aspect-level sentiment analysis, specialized NLP tech-
primarily focus on overall sentiment classification of documents or niques and machine learning models are used. These models often
sentences, aspect level analysis aims to delve deeper and analyze utilize more sophisticated algorithms to recognize and Identify the
sentiments associated with specific aspects or attributes of a given text’s characteristics, then classify the associated sentiment for each.
subject. In the context of restaurant tweets, these aspects can include Combining sentiment analysis techniques with aspect-level sen-
food quality, service, ambiance, price, and cleanliness. Conducting timent analysis becomes especially relevant. By incorporating both

979-8-3503-9447-4/24/$31.00 ©2024 IEEE 1

Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
 embeddings contribute significantly to downstream sentiment analysis
tasks, resulting in more precise and complex sentiment predictions.
Moreover, variants of BERT, including DistilBERT, RoBERTa,
and ALBERT, have been introduced, offering improved efficiency
and performance while retaining high-quality results. Sentiment
analysis has been transformed by these cutting-edge language
models, increasing its capacity to understand complex emotions
portrayed in text.

In Guangyao Pang et al. [2], Several methods have been developed

Fig. 1. An example of sentiment element of ABSA having the four key
in recent years for aspect-level sentiment analysis, with supervised
system machine learning algorithms providing the most promising outcomes.
However, these statistical-based methods often require meticulous
manual feature design on large datasets, leading to significant time
and resource consumption. Instead of depending on artificial feature
engineering, neural network models have the advantage of automat-
ically learning low-dimensional representations from reviews. This
quality attracts researchers’ attention and makes neural networks an
appealing alternative for aspect-level sentiment analysis tasks.
Despite these developments, current methods to extracting aspect
words’ situation and semantic information mostly depend on recurrent
neural networks (RNNs) or convolutional neural networks (CNNs),
ignoring the importance of the context of the aspect words in the text.
Studies have shown a significant relationship between aspect words’
emotional polarity and their word order, demonstrating that the con-
text of other aspect words nearby impacts an aspect word’s emotional
polarity. In addition, long-term connections between aspect words and
context are difficult for neural networks to capture, which results in
the loss of important data. Although the attention mechanism can
partially mitigate this difficulty by focusing on pertinent context, the
Fig. 2. Flow diagram of Different Techniques issue still exists and has a negative impact on their performance.
This work introduces an aspect-level sentiment model to solve these
issues.
methods, researchers can gain more comprehensive insights into Aspect-level sentiment analysis tests were carried out on three
the sentiments expressed in the text. For instance, when studying reference datasets to assess the ALM-BERT model. The outcomes
customer reviews of a product, aspect-level sentiment analysis can show that ALM-BERT beats the base MGAN models on the
provide detailed information about sentiments related to various Restaurant dataset with greater accuracy and macro-F1 score. In
product attributes like design, usability, performance, etc. Business comparison to other comparison approaches, ALM-BERT also shows
looking to enhance their goods and services based on consumer input higher accuracy in comment texts of various lengths. These results
need to have this detailed insight. demonstrate how well the ALM-BERT method extracts aspect-level
feelings from user reviews.
III. ABSA T ECHNIQUES
In our research paper, we have categorized our content into four Alexander Rietzler et al. [3], The study of Aspect-Based Senti-
distinct sections, each encompassing specific techniques, as illustrated ment Analysis (ABSA) in this work concentrates on Aspect-Target
in Figure 2. In the field of sentiment analysis, various techniques Sentiment Classification (ATSC). By detecting both the aspect (such
have been employed to analyze and interpret emotions in text. as ”food” or ”service”) and the sentiment associated with it in
Some prominent methods which are included in our research paper a given text, the major goal is to improve sentiment analysis.
are Encoder and decoder(BERT), Recurrent Neural Network(RNN), To improve ATSC performance, the authors employ deep transfer
SVM (Support Vector Machines), and a range of Miscellaneous learning techniques, including the BERT language model.
approaches as shown in fig.2. The two distinct versions of ABSA that the authors provide in
this study are aspect-category detection (ACD) followed by aspect-
A. Encoder and Decoder(BERT) category sentiment classification (ACSC) and aspect-target extraction
Sentiment analysis, an essential part of natural language process- (ATE) followed by ATSC. To address the drawbacks of traditional
ing, involves comprehending and categorizing the emotional tone models, they propose a two-step procedure for ATSC that entails
expressed in textual data. Machine learning models can handle text self-supervised domain-specific finetuning of the BERT language
better with the aid of encoding and decoding techniques. Encoding model, followed by supervised task-specific finetuning for ATSC.
turns raw text into numerical representations that computers can This approach enables BERT to gain domain-specific knowledge and
interpret, while decoding changes prediction numbers back into better understand the intricacies of sentiment analysis.
sentiment labels that people can comprehend. The BERT architecture, the fine-tuning procedure, and the con-
Recent years have seen a lot of interest in contemporary techniques version of the ATSC task into a BERT sequence-pair classification
like bidirectional encoder representations from transformers (BERT). job are all covered in the methods section. To assess the model
The potent pre-trained language model BERT, which is based on generalization abilities, several training and fine-tuning configurations
the transformer architecture, can encode words in a sentence while are looked at.
taking into consideration their left and right context. BERT’s ability to Furthermore, the paper introduces the concept of input reduction,
gather deeper semantic data and understand the intricacies of human an interpretation method for neural NLP models, which is applied
language is improved by its contextual awareness. Therefore, BERT to the ATSC task in a case study. The case study helps analyze

2
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
specific reasons for erroneous classifications, shedding light on areas
for model enhancement.
In summary, the paper highlights the significance of domain-
specific finetuning of the BERT language model for ATSC and
showcases its superior performance compared to baseline models.
The research provides valuable insights into the behavior of language
model finetuning and offers a robust approach for aspect-target
sentiment.

In Maria Mihaela Trusca et al. [4], The goal of the paper is to

improve sentiment prediction accuracy in opinionated literature by
presenting an improved version of the Hybrid approach for aspect-
based sentiment analysis (HAABSA) approach. Deep contextual
word embeddings like ELMo and BERT, which better capture word
semantics, would take the place of non-contextual word embeddings
in the additions that are being proposed. The modeling versatility Fig. 3. Classification of Transformer and BERT
is increased by introducing a hierarchical the HAABSA high-level
representations include an extra attention layer and an attention
mechanism.
In conclusion, leveraging customer reviews effectively through
Target polarities are first anticipated using a domain sentiment
NLP techniques and advanced models enhances e-commerce
ontology in the method’s first phase. A neural network dubbed Multi-
experiences. Approaches like aspect extraction and sentiment
Hop LCR-Rot serves as a backup model in the event of a tie. The
analysis provide valuable insights for customers and businesses
relationship between opinionated utterances and their surroundings is
alike, leading to informed decisions and improved trust. Continued
captured by the neural network with several hops that uses rotatory
research in this field promises further advancements and benefits in
attention. To evaluate the extensions, the approach is evaluated on
the realm of e-commerce.
the widely used datasets SemEval 2015 and SemEval 2016, and
contrasted with the Multi-Hop LCR-Rot model. Investigated are a
Mohammad Tubishat et al. [6], Customer reviews on product
variety of word embeddings, such as deep contextual (ELMo, BERT)
websites are becoming more common, making them essential sources
and context-free (GloVe, fastText, CBOW, SG) embeddings. The
of information for both consumers and businesses. To deal with the
implementation of hierarchical attention is also evaluated.
overwhelming number of reviews, sentiment analysis has emerged
The results of the experiment show that, when compared to the as an automatic technique to extract relevant information based on
original model, the suggested additions greatly improve sentiment users’ needs. It aids in deciphering consumer feedback and drawing
prediction accuracy. Deep contextual word embeddings, especially conclusions from reviews pertaining to various features of a good or
BERT, work more effectively than other methods. Additionally, service.
the inclusion of hierarchical attention further improves the model’s Recent research provide novel approaches for both explicit and
performance. implicit aspect extraction. To explicitly extract an aspect, methods
The report suggests additional research on the effects of including like the Whale Optimization Algorithm (WOA) are used to select
word embeddings in the architecture’s top levels and on diagnostic effective dependency relation patterns through web-based similarity,
classification to acquire a deeper understanding of the model’s overcoming limitations of hand-crafted patterns. A hybrid approach
operation. using corpus co-occurrence, dictionary-based analysis, and web-based
The techniques section covers the BERT architecture, the similarity is suggested in order to extract insights from implicit
fine-tuning approach, and the transformation of BERT sequence-pair characteristics and their corresponding opinion terms.
classification work from the ATSC task. Several training and Additionally, incorporating category-specific information can
fine-tuning setups are examined to judge the models’ generalization strengthen sentiment analysis, improving the performance of sen-
capabilities. timent classifiers. The use of gated RoBERTa models has shown
promising results, outperforming previous recurrent neural network
Haozheng Tian et al. [5], Customer reviews play a pivotal role in and transformer-based models, leading to enhanced customer engage-
the e-commerce domain, providing valuable insights and aiding pur- ment and potentially driving revenue growth.
chase decisions. However, the vast volume and diverse vocabulary of In conclusion, sentiment analysis and aspect extraction techniques
reviews make it challenging for customers to comprehend and assess are essential for extracting valuable insights from customer reviews.
them effectively. To address these challenges, several approaches have The proposed approaches offer innovative solutions to efficiently
been proposed. analyze and extract information, enhancing the browsing experience
Aspect extraction is one such approach, using NLP techniques for customers and driving revenue growth for companies.
to summarize product-relevant information and determine expressed
opinions in customer reviews. This enhances readability and interpre- In Mickel Hoang et al. [7], A well-known area of natural language
tation, ensuring a smooth shopping experience. Combining explicit processing (NLP) study called sentiment analysis, often known as
and implicit aspects allows for a comprehensive understanding of opinion mining, aims to automatically identify the sentiment con-
customer sentiment. tained in a document. Traditional techniques to sentiment analysis,
Additionally, NLP techniques like hierarchical clustering and on the other hand, usually miss the specific issue or component to
semantic relationship inference can organize and merge aspects, which the sentiment is directed, which limits their effectiveness in
grouping related ones together for a coherent representation of fully understanding user opinions. Analysis of sentiment based on
opinions. Leveraging synonyms and antonyms aids in clustering aspects (ABSA), a more difficult task that includes locating both
implicit aspects, facilitating nuanced customer feedback analysis. To sentiments and aspects inside a text, has developed as a solution to
evaluate sentiment in reviews, a gated RoBERTa sentiment classifier this restriction.
is proposed, outperforming previous models in cross-domain senti- In this paper, the authors suggest an ABSA model which improves
ment analysis with improved accuracy. aspect-based sentiment analysis, particularly in instances outside the

3
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
authors’ area of expertise, through using BERT’s abilities as a pre- datasets. Graph-based neural networks and cutting-edge embedding
trained language model. BERT, a language model trained on vast methods like BERT and ELMO may be used to boost the accuracy
amounts of unannotated text, has shown promise in achieving better of the pipeline.
performance, requiring less labeled data and shorter training times. In conclusion, this research presents a comprehensive NLP
However, its application to ABSA tasks has been limited in previous pipeline for The problem of opinions being in line with elected
evaluations. figures is effectively solved by sentiment evaluations of mainstream
To address this, the authors adopt a fine-tuning approach, trans- media messages. The pipeline’s components, including PoS tagging,
forming BERT into a sentence pair classification model customized NER, anaphora resolution, and sentiment analysis, achieved
for ABSA. The aspect classifier is trained using data labeled as favorable performance metrics. By visualizing politician popularity
”related” or ”unrelated” to the aspect. Moreover, the authors explore through time series graphs, the study provides valuable insights
fine-tuning the model for sentiment classification, enabling it to learn for politicians, political affiliations, the general public, and media
the sentiment context associated with specific aspects. houses. Future work can involve incorporating advanced neural
The paper also introduces a combined model that uses a single network models and embedding techniques to further enhance the
sentence pair classification model to simultaneously classify aspects pipeline’s accuracy and performance.
and sentiments. The efficiency of state-of-the-art methods in aspect-
based sentiment categorization, is demonstrated by experimental Li Yu et al. [9], This study focuses on aspect-based sentiment
findings. analysis (ABSA) of product reviews in the context of e-commerce.
The research concludes by presenting an innovative ABSA model It goes into great detail about the method for gathering implicit
based on BERT, which addresses the limitations of traditional aspects—aspects that aren’t directly specified in review phrases. The
sentiment analysis by considering both aspects and sentiments. The authors present a BERT-based fine-tuning technique, drawing on
model is fine-tuned using a sentence pair classification approach and BERT’s success in earlier natural language processing tasks, to extract
augmented with generated data. The author discusses the fine-tuning implicit characteristics from Chinese online reviews.
of sentiment classification and suggest a powerful combination The relevance of product reviews as a useful tool for understanding
model for aspect and sentiment classification. The results of the consumer demands, spotting problems with products, and enhancing
experiments show how useful the recommended models are for product quality and profitability is emphasized in the introduction.
aspect-based sentiment analysis. Since 30 percent of Chinese reviews contain implicit aspects, which
conventional dictionary-based algorithms find difficult to handle
Birat Bade Shrestha et al. [8], In order to correlate expressed ideas effectively, implicit aspect extraction becomes an important ABSA
with particular political leaders,The sentiment analysis of news media subtask.
texts is the main emphasis of this study. The resolution of anaphoras The paper outlines three main approaches to implicit aspect
and named-entity identification are required since this is a serious extraction: unsupervised, semi-supervised, and supervised methods.
issue. The main goal is to create an NLP pipeline for natural language Unsupervised methods use co-occurrence matrices or count-based
processing that uses reliable methods for these tasks and accurately approaches, while semi-supervised methods combine labeled and
associates political figures with the right category of viewpoints in unlabeled data, and supervised methods rely on labeled data using
the texts. It is possible to learn a lot by analyzing the popularity of techniques like support vector machines (SVM) or rule-based ap-
politicians using time series graphs of positive and negative sentiment. proaches.
As traditional news sources decline in usage, online news portals In light of BERT’s success in NLP tasks, the authors propose a
and social media have become prominent sources for information. modified BERT approach for implicit aspect extraction in Chinese
Because of this change, texts from the media are a valuable source online reviews. The BERT model is pre-trained on a Chinese general
for sentiment analysis in the socio-political sphere. It is important to corpus, and the hidden layer parameters are adjusted using labeled
assess a politician’s popularity since it represents public opinion and data. The resulting improved BERT model is then utilized to extract
affects decisions about public policy and election results. However, implicit properties. Experimental results on reviews of clothing show
analyzing sentiments in news articles poses challenges, particularly in that the proposed method outperforms the pre-training BERT model
accurately identifying and resolving named-entities, especially when and the word2vec+LSTM model in implicit aspect recognition.
referring to multiple politicians. This research focuses on calculating The research concludes by presenting a refined BERT technique
popularity ratings of Nepali politicians by analyzing news media to extract implicit characteristics from Chinese online reviews, with
texts. an emphasis on clothing reviews in particular.By adjusting the
To tackle these challenges, the research introduces a compre- parameters of the pre-trained BERT model, the proposed approach
hensive NLP pipeline comprising several sub-tasks. Contributions achieves superior performance in implicit aspect extraction. The
include a Part-of-Speech (PoS) tagger, a Named-Entity Recognition study makes a significant contribution as the first to apply fine-tuning
(NER) classifier, a rule-based Anaphora Resolution module, and a BERT for extracting implicit aspects. The authors suggest future
sentiment classifier using a manually labeled Sentiment Corpus. The work to expand the evaluation on a larger set of clothing reviews and
pipeline first identifies and resolves named-entities and then classifies improve the model’s capability to handle sentences with multiple
sentences based on sentiments expressed towards politicians. For implicit aspects.
sentiment classification, Support Vector Machine (SVM) achieved
the highest performance using Word2Vec with skipgram for feature Pulung Nurtantio Andono et al. [10], Hotel reviews have become
extraction. The popularity ratings are then visualized through time- crucial for travelers to make informed decisions about hotel selec-
series graphs. tion. However, processing large-scale review data poses challenges
While each component of the pipeline demonstrated relatively high in accurate term selection and aspect categorization for sentiment
performance, there are some limitations to consider. Probabilistic and analysis. To tackle this, the study proposes a two-step process to
rule-based models used in this research underperformed compared to improve the accuracy of sentiment analysis and aspect categorization
neural network models, but data limitations prevented their use. The in hotel evaluations.
anaphora resolution component ignored other expressions relating to Pre-processing review data with TF-ICF in order to extract phrases
entities and solely dealt with pronouns. The target or the opinion The first level is based on aspect keyword variables for each hotel
holder were not specifically addressed in the work in terms of aspect category. The accuracy of each phrase is then increased
opinions. Future improvements might use more sophisticated neural by using Latent Dirichlet Allocation (LDA), which reveals hidden
network models designed specifically for NLP tasks and greater themes related to each term. In order to find substantial discrepancies

4
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
in similarity results among aspect categories, aspect categorization
uses BERT embeddings and semantic similarity, which leads in more
accurate aspect category classification in reviews. High precision,
recall, and F1-measure scores of 0.86, 0.92, and 0.89 were obtained
in the examination of aspect extraction.
For aspect-based sentiment analysis, the study used the BERT
sentiment analysis method. When the performance of sentiment
analysis was examined, excellent accuracy, recall, and F1-measure
scores of 0.96, 0.98, and 0.97 were observed, surpassing previous
studies on hotel review ratings. Notably, the proposed approach
efficiently processes large-scale review data containing multiple sen-
tences, addressing the limitations of prior research that focused on
simple sentences. Fig. 4. Accuracy of different word embedding like, LSTM, TD-LSTM and
While the approach shows promising results, there are areas for TC-LSTM
future improvement. Implicit aspect and opinion extraction, as well
as the extraction of word phrases representing multi-category aspects
within reviews, require further investigation. Exploring alternative By considering how closely related a target word is to the other words
aspect categorization methods could also enhance aspect and opinion in its context, the target-dependent LSTM approach enables the model
extraction accuracy, leading to more comprehensive aspect-based to infer sentiment polarity toward the target. For supervised sentiment
sentiment analysis. categorization, the model is trained using traditional backpropagation
The study’s effective method for categorizing aspects and per- with cross-entropy error.
forming aspect-based sentiment analysis in hotel reviews, utilizing The proposed method is evaluated and compared with alternative
methods including LDA, semantic similarity, and BERT, is presented strategies using a benchmark dataset. Findings show that the proposed
as a conclusion. The suggested strategy outperforms earlier studies technique, without the use of an external sentiment lexicon or
and produces results with high accuracy based on data from large- syntactic parser, achieves state-of-the-art classification accuracy. The
scale reviews. Future studies should focus on addressing implicit inclusion of target information in the LSTM significantly improves
aspects and opinions as well as word phrase extraction for multi- classification accuracy.
category aspects in order to further increase the accuracy and efficacy
Additionally, as shown in fig.3, the research investigates how word
of aspect-based sentiment analysis.
embeddings affect LSTM, TD-LSTM, and TC-LSTM. Two types of
B. Recurrent neural network (RNN) word embeddings are studied: sentiment-specific word embeddings
(SSWE) and Glove vectors. SSWEu, which considers the sentiment
Due to their proficiency with sequential data, recurrent neural of sentences and contexts of words, proves to be beneficial.
networks (RNNs) have become a well-liked neural network design for
Overall, the proposed target-dependent LSTM approach shows
sentiment analysis. Sentiment analysis includes classifying the text’s
promising results in sentiment classification by effectively
emotional tone, and RNNs are excellent for jobs requiring sequence
incorporating target information and outperforming other methods
modeling in natural language processing.
on the benchmark dataset. The study of different word embeddings
RNNs process input sequences by maintaining a hidden state that
emphasizes the significance of creating effective text representations
evolves as it processes each element in the sequence, allowing them
by employing high-quality word choices.
to capture dependencies and contextual information between words
in a sentence. On the other hand, the vanishing gradient problem,
which affects traditional RNNs, makes it difficult for them to detect Bing Qin et al. [12], The purpose of this work is to solve the
long-distance links in text. shortcomings of traditional neural models, such as the LSTM, in
Advanced RNNs, such Gated recurrent units (GRU) and Long aspect-level sentiment categorization. They contend that conventional
Short-Term Memory (LSTM), have been developed to overcome LSTM implicitly absorbs context information and neglects to explic-
this restriction. GRU offers comparable outcomes with fewer gating itly draw attention to crucial context cues for an aspect. They contend
units than LSTM while introducing gating mechanisms that regulate that just a portion of the context vocabulary is required to infer the
information flow. As a result of their ability to capture long-term attitude toward a certain feature, and the ideal solution would be able
relationships and deal with the vanishing gradient problem, LSTM to directly convey the significance of these context terms.
and GRU have demonstrated promising results in sentiment analysis. To achieve this goal, the researchers propose a deep memory
To prepare text data for RNNs, words are typically transformed network for aspect-level sentiment classification, taking cues from
into dense vector representations called word embeddings, which effective computational models that make use of explicit memory
capture semantic relationships and enhance the RNN’s understanding and attention processes. Their method is data-driven, computationally
of word meaning and context. In sentiment analysis, RNNs are trained effective, and independent of emotion lexicons or syntactic parsers.
on labeled datasets containing text samples with corresponding senti- The suggested method uses numerous computational levels with
ment labels. The RNN can be used to identify the sentiment of fresh, common input and output. Each layer functions as a content- and
unexplored text data by learning to predict sentiment labels based on location-based attention model, which decides how to display con-
input phrases during training. tinuous text by learning the importance or weight of each context
In general, RNNs have shown high performance in sentiment word. The text representation in the top layer is a feature for
analysis tasks, particularly when context and word dependencies sentiment categorization. The complete model may be effectively
are important considerations. RNNs continue to be useful tools for trained end-to-end using gradient descent and the cross-entropy error
sentiment analysis and other sequential NLP applications, despite of sentiment classification as the loss function because all components
the emphasis given to more recent models like Transformer-based are differentiable.
architectures. On the basis of laptop and restaurant datasets from SemEval
2014, the researchers assess their methodology. Experimental results
Duyu Tang et al. [11], The research presents neural network mod- demonstrate that their method outperforms a top system that uses
els for target-dependent sentiment categorization. By adding target feature-based SVM. In terms of classification precision and running
knowledge, the method enhances long-short-term memory (LSTM). speed, it performs better than both attention-based LSTM models and

5
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
LSTM models. Performance is also enhanced by utilizing numerous Binxuan Huang et al. [15], This study focuses on target-dependent
processing layers over external memory. sentiment classification, where the goal is to ascertain the statement’s
In summary, the researchers introduce a deep memory network sentiment polarity toward a certain target. The authors talk about how
for aspect-level sentiment categorization that is motivated by challenging it is to simulate the semantic linkages between the target
explicit memory and attention processes. They do not rely on words and their context in a phrase.
sentiment lexicons or syntactic parsers in their data-driven method, To tackle this challenge, the authors propose two target-dependent
which is computationally effective. Experimental results indicate LSTM models. The first model is a basic LSTM that represents
its superiority over traditional LSTM models and attention-based the sentence’s semantic composition without considering the target
LSTM models both in terms of categorization speed as well as word. The second model, called target-dependent LSTM (TD-LSTM),
precision. The approach shows promising potential for improving incorporates the target word to capture the relatedness between the
sentiment analysis in various domains. target and its context words, leading to more accurate sentiment
classification.
Youwei Song et al. [13], The paper introduces the Attentional The paper also introduces an extended version of TD-LSTM that
Encoder Network (AEN), a unique model for focused sentiment incorporates target connection to further enhance the understanding
categorization. Identifying the polarity of a sentence’s attitude toward of the sentiment expressed towards the target. Target information
certain targets is a requirement of this employment. Prior systems is incorporated into LSTM models, which evaluation on a Twitter
have employed attention-based recurrent neural networks (RNNs), benchmark dataset shows dramatically raises classification accuracy.
but the AEN adopts a different strategy by relying on attention- Without utilizing external sentiment lexicons or syntactic parsers,
based encoders to forge meaningful connections between the context the proposed target-dependent LSTM models achieve cutting-edge
and the target words. With this move away from RNNs, problems performance.
with parallelization and long-term pattern recognition are addressed, In summary, the study emphasizes the importance of considering
leading to a sentiment analysis model that is more effective and target information when categorizing attitudes and provides effective
efficient. LSTM-based models that successfully capture the connections
The label unreliability issue, particularly with regard to neutral between target words and their context, leading to increased
emotion states that might be ambiguous, is another crucial topic performance when compared to previous methods.
covered in the study. Label smoothing regularization, a method
included when confronted with hazy sentiment labels, into the loss LI XIAOYAN et al. [16], Sentiment classification in the context
function to make the model less confident, is introduced in the study of Chinese language has been a prominent area in natural language
as a solution to this issue. This regularization increases the targeted processing research. To address limitations in traditional methods,
sentiment classification’s overall robustness and flexibility. this study introduces the BiLSTM-Attention model, a sentiment
classification approach based on deep learning. For the objective
The research also investigates the use of BERT (Bidirectional
of handling both the long and short parts of Chinese mixed
Encoder Representations from Transformers) for targeted sentiment
texts, the model includes Bi-directional Long Short-Term Memory
categorization. The addition of BERT improves the model’s
(BiLSTM) and an ann model.The suggested model excels in
performance even more, producing cutting-edge findings and
sentiment classification for mixed and lengthy texts, outperforming
showcasing the AEN’s ability to improve sentiment analysis
approaches like CNN-LSTM, BiLSTM, TextCNN, CNN, and LSTM.
based on learned language representations. The attentional encoder
The practical implications of this model are significant, enabling
network provides a promising and portable approach for fine-grained
businesses to gain insights from customer opinions and informing
targeted sentiment classification, with implications for a range of
policy decisions based on public sentiment. The study acknowledges
natural language processing sentiment analysis and opinion mining
limitations related to small experimental datasets and a focus on
applications.
ordinary Chinese characters, suggesting future research directions
to expand datasets and explore sentiment analysis in texts with
Hu Xu et al. [14], The authors of the research paper propose a traditional Chinese characters. Overall, the BiLSTM-Attention model
deep memory network for aspect-level sentiment categorization that offers an effective solution for sentiment classification in Chinese
is motivated by computational models’ explicit memory and attention texts, contributing to the advancement of sentiment analysis.
processes. Capturing clear context cues and improving sentiment
categorization for a particular element are the major objectives. Yu Zhang et al. [17], The e-commerce industry, it may be difficult
Their method uses many computational layers with common to glean insightful perspectives from a sea of unhelpful consumer
parameters. A content- and location-based attention model is used evaluations. While most existing research focuses on explicit features
by each layer to determine the significance of context words. and in reviews, this study aims to extract implicit features related to
calculate continuous text representations, focusing on relevant words online transactions, including sellers, services, and logistics. The
for the target aspect. A top system that employs feature-based proposed method is a co-occurrence association-based approach that
SVM outperforms the suggested strategy, according to experiments goes beyond product features, providing a deeper comprehension of
Using laptop and dining establishment SemEval 2014 datasets. It client preferences and perspectives.
outperforms standard LSTM and attention-based LSTM models in The article makes two major contributions. First of all, it broadens
terms of accuracy and running speed. the definition of feature extraction to include other facets of on-
Domain-specific embeddings are found to be crucial for datasets line transactions in addition to product characteristics. Secondly,It
with domain-specific aspects, while general embeddings perform investigates relationships between feature words and other notional
better on datasets with more general aspects. The contribution of the terms inside sentences as well as correlations between feature words
CRF layer is limited, especially when the initial accuracy is already and opinion words, enabling a more in-depth examination of implicit
high. It is more effective for capturing label dependences, which are characteristics in customer evaluations.
not the main types of errors in this task. Through real-world data trials on Chinese web reviews, the
In conclusion, the deep memory network with attention efficiency of the suggested methodology is shown, with possible
mechanisms and explicit memory is a promising approach for applicability to reviews in other languages. The authors think that
emotion categorization at the aspect level, showing improved by using their method, new avenues for study into the extraction of
performance compared to traditional LSTM models. implicit characteristics from online consumer evaluations are opened
up.

6
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
 For the purpose of extracting implicit characteristics from strategy, which performs better than many supervised and unsuper-
customer reviews in the e-commerce arena, this research offers a vised baseline methods, showing promising results without the need
unique co-occurrence association-based strategy. By taking into for feature engineering.
account a variety of online transactional factors, the proposed The interpolation coefficient and the number of clusters in the
method offers a more comprehensive analysis of customer opinions. k-means algorithm have a considerable influence on classification
The technique’s effectiveness is validated through experiments, and performance, according to the approach’s hyperparameter tweaking
its contributions lie in its broader scope of feature extraction and tests.
utilization of associations between feature words and other notional Using soft cosine similarity and k-means clustering, the proposed
words within clauses. unsupervised technique for aspect category classification adds
significantly to aspect-based sentiment analysis. It performs well,
Hanane Elfaik et al. [18], In this trend, the popularity of social especially in sectors with little labeled data or in languages with
networks like Twitter, affect analysis—the technique of identifying limited resources.
emotions from text—has received more attention in natural language
processing. This study focuses on affect analysis of Arabic tweets, Rayees Ahmad et al. [20], This research paper proposes an
which presents unique challenges due to the informal, noisy, and effective machine learning technique, SVMS (SVM using Spacy),
multilingual nature of Twitter language, as well as the complexities for customer feedback evaluations’ aspect-based sentiment analysis
of the Arabic language itself. (ABSA). The study compares the performance of the SVMS model
The authors suggest a method that combines AraBERT, a against four fundamental classifiers (LR, NB, RF, and SVM) using
transformer-based model for comprehending Arabic, with an atten- a dataset of 7,570 tweets that were obtained from Twitter. using
tional LSTM-BiLSTM model for classifying multi-label emotions. various evaluation metrics. The SVMS model outperforms the other
Their strategy is tested against baseline techniques and existing algorithms with good F1 score, recall, accuracy, and precision.
methodologies using the SemEval-2018 Task1 dataset, proving its
superiority. The suggested model outperforms both the top-placed The study’s innovation consists in comparing sentiment analysis
model from the SemEval2018-Task1 competition and the best model classifiers and focusing on certain review elements, which advances
that has been previously published in the correctness in literary works. sentiment analysis methodologies. The implementation of the SVMS
The implications of this work extend beyond academia. The model utilizes belief maximization and the Spacy pipeline for aspect
proposed approach can benefitGovernments, health organizations, recognition and overall sentiment prediction. The study showcases
and decision-makers can track emotions by looking at social media the effectiveness of the SVM model integrated with Spacy, giving
posts. It can also aid businesses in analyzing customer emotions for insightful information for next field researchers.
improving strategies and assist in identifying potential criminals by The results highlight the value of aspect-based sentiment analysis
analyzing emotions after attacks or crimes. Considering The authors and the useful applications of the suggested model in several fields.
recommend further research on a web-based emotion recognizer to The model will need to be improved in the future with more data,
examine real-time feelings during upcoming epidemics in light of the feature engineering, deep learning techniques, and an expansion to
COVID-19 pandemic. neutral tweets and different languages.
In conclusion, this study presents an effective approach for affect
analysis in Arabic tweets using deep contextualized language models. Salim Sazzed et al. [21], This study suggests SSentiA (Self-
The practical applications include government monitoring, business supervised Sentiment Analyzer), a hybrid technique, to enhance
strategy improvement, and post-attack or crime analysis. Future work sentiment categorization utilizing unlabeled data. The LRSentiA
involves developing a web-based emotion recognizer for real-time (Lexical Rule-based Sentiment Analyzer) framework is presented by
emotion detection during epidemics. the authors as the cornerstone for producing precise pseudo-labels.
To improve sentiment classification performance, especially for less
C. SVM (Support Vector Machines) polarized and complicated evaluations, SSentiA combines LRSentiA
Sentiment analysis, the method of determining a text’s emotional with a supervised ML algorithm. The results show significant
content, is often used in natural language processing. A popular improvements in sentiment classification accuracy compared to
technique for sentiment analysis is the Support Vector Machine existing unsupervised and lexicon-based methods. SSentiA achieves
(SVM) algorithm. Feature extraction from the text using the bag-of- similar performance using only 30 percent - 60 percent labeled
words model, labeling, training, classification, and assessment are all training data, making it effective in domains with limited annotated
steps in the SVM process. SVM is effective for sentiment analysis in data. The study demonstrates the value of sentiment analysis as well
smaller datasets with feature engineering but may be outperformed as the potency of the hybrid technique that has been suggested for
by deep learning models for larger datasets and complex tasks. managing unlabeled data in sentiment classification.

Erfan Ghadery et al. [19], Aspect-based sentiment analysis, the Shini George et al. [22], This study examines the difficulty of class
study presents an unsupervised method for aspect category recog- imbalance in sentiment analysis and suggests a two-level strategy
nition. This technique takes use of clusters of unlabeled reviews to overcome this issue. The article proposes the ensemble bagging
and makes use of the soft cosine similarity metric, in contrast to SVM (EBSVM) model for opinion mining and introduces the
supervised alternatives that need for labeled training data and feature Synthetic Minority Oversampling Technique (SMOTE) to balance
engineering. the dataset. The proposed EBSVM model outperforms individual
The approach involves clustering unlabeled review sentences using classifiers and achieves better sentiment classification performance
the k-means algorithm based on average word embeddings. Sentences on both unequal and equal datasets. The study evaluates opinion
within the same cluster are assumed to share similar information mining classifiers considering specific aspects of customer reviews
about the categories they belong to.By accounting for similarity for restaurants and highlights the importance of addressing class
between the phrase and manually chosen seed words linked with imbalance in sentiment analysis. The novel contributions include
the category, the soft cosine similarity metric is used to determine comparing the suggested model to current classification techniques
how similar a sentence is to a pre-defined category. and using SMOTE to balance the dataset. The outcomes show how
During testing, the scores obtained for the sentence and the well the suggested method works, especially in cases when the data
nearest cluster are interpolated to detect the mentioned categories. is unbalanced, to improve sentiment analysis findings.
The SemEval-2014 restaurant dataset is used to test the proposed

7
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
 Monisha V.N et al. [23], This research paper explores the ap- In this work, we compare ABAE with BTM experimentally on a
plication of sentiment analysis to predicting emotions expressed in variety of tasks.
tweets from Twitter. To create a prediction model, the study uses
the supervised learning techniques Naive Bayes and Support Vector Ruidan He et al. [27], The goal of the research paper’s Interactive
Machine (SVM). The goal is to properly classify the tweets into Multi-Task Learning Network (IMN) for aspect-based sentiment
positive, negative, or neutral attitudes using the training dataset analysis (ABSA) is to efficiently and effectively extract aspect
of movie reviews. Twitter and other social media platforms have words and estimate sentiment polarity. ABSA tasks are often carried
emerged as important sources of user-generated content, making out individually in a pipeline, which does not make use of joint
sentiment analysis valuable for various applications, including social knowledge or extra training data.
media marketing and brand monitoring. The study assesses the effi- IMN addresses this limitation by jointly learning multiple tasks
cacy of SVM and Naive Bayes in sentiment prediction., with Naive at both token and document levels. It uses a message passing
Bayes outperforming SVM in this specific scenario. However, the architecture to exchange information between tasks through shared
accuracy of the models is influenced by the dataset used, emphasizing latent variables, allowing interactions and leveraging document-level
the importance of different datasets and high-quality in dataset for labeled sentiment corpora.
sentiment analysis. Comparing with baseline models, IMN demonstrates superior
performance on three benchmark datasets, excelling in aspect term
D. Miscellaneous extraction and aspect-level sentiment classification. It effectively
Piyush Kumar Soni et al. [24], Sentiment analysis has emerged as a captures correlations between tasks and benefits from knowledge in
pivotal research area, driven by the surge in online opinions shared via larger corpora. The paper discusses multi-task learning and message
platforms like e-commerce sites and social networks. For decision- passing in related fields like NLP and computer vision, emphasizing
making in a variety of fields, these viewpoints provide insightful IMN’s unique contribution in modeling task interactions and infor-
information. Scalable and impartial sentiment analysis techniques mation propagation.
present sentiment summaries from in-depth reviews. Manufacturers Overall, IMN offers an advanced and integrated approach to
and customers alike stand to gain from the use of this technique in ABSA, improving performance by effectively using joint information
both public and private sectors. The sentiment analysis landscape uses and multiple data sources. The architecture shows potential for
natural language processing and data mining to categorize sentiments applications in other NLP tasks.
as neutral, positive, or negative. Sentiment evaluation for various
features of an entity is done at various granularities, from document- Soujanya Poria et al. [28], The work provides a distinctive rule-
level to aspect-level. Both explicit and implicit aspects are possible; based approach for aspect extraction in aspect-based opinion mining,
the latter is inferred from the former. Researchers have been inspired with a focus on extracting both explicit and implicit aspects from
to use the well-known Recurrent Neural Networks (LSTM) for text opinionated text. While implicit aspects are communicated using im-
classification. plicit aspect cues (IACs), explicit aspects are stated directly. Although
Wenxuan Zhang et al. [25], The objective of this paper is to implicit aspect extraction has received little attention, explicit aspect
provide a comprehensive overview of the aspect-based sentiment extraction has received substantial research.
analysis problem, covering all of its many tasks, methodologies, The proposed framework leverages common-sense knowledge and
current issues, and likely future directions. First, we established sentence dependency trees to identify aspects. Experimental evalua-
the context for ABSA research by defining it, outlining its four tion on a dataset showed showing the technique works better than
sentiment components, popular modeling approaches, and listing previous techniques in terms of recall and precision, proving its
available resources. Then, we carefully outline each ABSA difficulty usefulness.
and its associated solutions, with an emphasis on the most recent The evaluation of implicit aspect extraction remains challenging
advancements of the compound ABSA challenges. To give a due to the absence of a benchmark. Nevertheless, the proposed
clear picture of where we are right now, we categorize known framework opens up new avenues for further research in this area,
studies based on the sentiment components involved and highlight encouraging the development of additional rules and the exploration
sample approaches of various modeling paradigms for each task. of combining existing rules for more complex extraction tasks.
Pretrained language models have greatly enhanced a variety of Future work involves refining and expanding the aspect knowledge
ABSA occupations thanks to their application in the ABSA problem, base, incorporating resources like WordNet and SenticNet to enhance
is also discussed. We look into the benefits they offer as well as scalability and reduce noise. The unsupervised nature of the frame-
the restrictions they have. Additionally, we examine developments work, based on common-sense knowledge and sentence structure,
in cross-domain and cross-lingual ABSA, which may result in makes it a promising approach for aspect-based opinion mining.
ABSA systems that are more useful. Finally, we talk about some Overall, the proposed rule-based framework offers a valuable
of the existing issues and anticipated future developments in this area. contribution to sentiment analysis by pulling from an opinionated
text both explicit and tacit features. This method has the potential to
Wee Sun Lee et al. [26], The retrieved aspect phrases are typically improve sentiment analysis by collecting many characteristics and
too complex for rule-based algorithms to categorize. Yet, a sizable views stated in textual data as it is improved and advanced.
amount of labeled data must be accessible for supervised learning
to be successful. Subsequently, techniques without a need on labeled Farheen Pathan et al. [29], Sentiment analysis, often known as
data, particularly topic models, have been introduced. For each aspect, opinion mining or text mining, aims to extract the arbitrary views
these models frequently provide word distributions or rankings. and sentiments expressed in text texts. Coupled with the expansion
Without doing extraction and categorization independently, aspects of social media and the availability of vast amounts of controversial
are gained spontaneously. content in digital formats, sentiment analysis has experienced fast
Most recently proposed a restricted Boltzmann machine (RBM)- expansion. It entails figuring out if a text is positive, negative, or
based model that simultaneously extracts aspects and applicable neutral in terms of semantic orientation. Negations, intensifiers, and
sentiments from a given review phrase by treating aspects and diminishers are examples of valence shifters that can change the way
sentiments as independent hidden variables in RBM. The RBM- a text expresses emotion.
based model, on the other hand, is strongly dependent on historical The bag-of-words (BOW) model, which is commonly used for
data, such as part-of-speech (POS) tagging and sentiment lexicons. text representation, is a popular technique in sentiment analysis.
A biterm topic model (BTM) that generates co-occurring word pairs. Nevertheless, because it ignores word order, grammatical structures,

8
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
and significant semantic information, it has shortcomings when R EFERENCES
it comes to sentiment categorization. Polarity shift, where certain
linguistic terms can cause a text’s sentiment polarity to change, is [1] Piyush Kumar Soni and Radhakrishna Rambola, ”A Survey on Implicit
Aspect Detection for Sentiment Analysis: Terminology, Issues, and
a key difficulty in sentiment analysis. For example, the addition of Scope”.
a negation word like ”don’t” can change a positive sentiment to a [2] Guangyao Pang, Keda Lu, Xiaoying Zhu, Jie He, Zhiyi Mo, Zizhen
negative one. The BOW representation fails to capture such polarity Peng and Baoxing Pu1, ”Aspect-Level Sentiment Analysis Approach
shifts, leading to misclassification. via BERT and Aspect Feature Location Model”.
To address the shortcomings of the BOW model and tackle the [3] Alexander Rietzler, Sebastian Stabinger, Paul Opitz and Stefan Engl,
polarity shift problem, more sophisticated approaches have been ”Adapt or Get Left Behind: Domain Adaptation through BERT Lan-
proposed. One such approach is dual sentiment analysis (DSA), guage Model Finetuning for Aspect-Target Sentiment Classification”.
which extends sentiment classification to consider positive, negative, [4] Maria Mihaela Trusca, Daan Wassenberg, Flavius Frasincar and Rom-
mert Dekker, ”A Hybrid Approach for Aspect-Based Sentiment Analysis
and neutral sentiments. DSA utilizes a dual training algorithm that Using Deep Contextual Word Embeddings and Hierarchical Attention”.
incorporates original and reversed training reviews, along with a dual [5] Haozheng Tian and Haozheng Tian, ”A Pipeline of Aspect Detection
prediction algorithm that takes into account both sides of a review. and Sentiment Analysis for E-Commerce Customer Reviews”.
This approach allows for a better understanding of polarity shifts and [6] Mohammad Tubishat and Norisma Idris, ”Explicit and Implicit Aspect
improves sentiment classification accuracy. Extraction using Whale Optimization Algorithm and Hybrid Approach”.
Sentiment analysis has difficulties managing the enormous volume [7] Mickel Hoang, Oskar Alija Bihorac and Jacobo Rouces, ”Aspect-Based
of opinionated data present on numerous platforms, such as reviews, Sentiment Analysis Using BERT”.
forum debates, blogs, and social media, in addition to polarity [8] Birat Bade Shrestha and Bal Krishna Bal, ”Named-Entity Based Senti-
ment Analysis of Nepali News Media Texts”.
fluctuations. Market research, consumer feedback analysis, and brand [9] Li Yu and Xuefei Bai, ”Implicit Aspect Extraction from Online Clothing
management are a few areas where sentiment analysis is useful. Reviews with Fine-tuning BERT Algorithm”.
It helps companies to evaluate user sentiments, assess consumer [10] Pulung Nurtantio Andono, Sunardi, Raden Arief Nugroho and Budi
happiness, and make defensible decisions based on the sentiment Harjo,”Aspect-Based Sentiment Analysis for Hotel Review Using LDA,
revealed in textual data. Semantic Similarity, and BERT”.
Overall, The dynamic discipline of sentiment analysis works to [11] Duyu Tang, Bing Qin, Xiaocheng Feng and Ting Liu,”Effective LSTMs
identify views and feelings in text. Valence shifters and polarity for Target-Dependent Sentiment Classification”.
shifts pose challenges that necessitate more advanced approaches [12] Duyu Tang, Bing Qin and Ting Liu, ”Aspect Level Sentiment Classifi-
cation with Deep Memory Network”.
beyond the traditional BOW model. By developing more sophisticated
[13] Youwei Song, Jiahai Wang, Tao Jiang, Zhiyue Liu and Yanghui Rao,
algorithms and techniques, sentiment analysis can provide important ”Attentional Encoder Network for Targeted Sentiment Classification”.
insights into user attitudes, consumer sentiments, and public opinion [14] Hu Xu, Bing Liu, Lei Shu and Philip S. Yu,”Double Embeddings and
that help firms and organizations better their strategy through data- CNN-based Sequence Labeling for Aspect Extraction”.
driven decision-making. [15] Binxuan Huang, Yanglan Ou and Kathleen M. Carley, ”Aspect Level
Sentiment Classification with Attention-over-Attention Neural Net-
works”.
IV. C ONCLUSION [16] LI XIAOYAN AND RODOLFO C. RAGA JR, ”BiLSTM Model With
Attention Mechanism for Sentiment Classification on Chinese Mixed
Text Comments”.
Aspect-based sentiment analysis (ABSA) makes it possible to fully [17] Yu Zhang and Weixiang Zhu, ”Extracting Implicit Features in Online
understand the emotions expressed toward specific characteristics or Customer Reviews for Opinion Mining”.
entities in textual data (NLP). This study examines the most recent [18] Hanane Elfaik and El Habib Nfaoui, ”Combining Context-Aware Em-
state-of-the-art practices and methods used in ABSA, with a focus on beddings and an Attentional Deep Learning Model for Arabic Affect
implicit aspect recognition in restaurant reviews. Identifying slightly Analysis on Twitter”.
indicated features through other words and expressions is necessary [19] Erfan Ghadery, Sajad Movahedi, Heshaam Faili and Azadeh Shakery,
in order to detect implicit aspects, which is still a difficult task. A ”An Unsupervised Approach for Aspect Category Detection Using Soft
more extensive sentiment analysis is made possible by both explicit Cosine Similarity Measure”.
[20] Rayees Ahmad, Yasmin Shaikh and Sanjay Tanwani, ”Aspect Based
and implicit aspect identification, which gives important information
Sentiment Analysis and Opinion Mining on Twitter Data Set Using
about the preferences and opinions of customers. Linguistic Rules”.
Throughout this survey, various issues in detecting implicit as- [21] Salim Sazzed and Sampath Jayarathna, ”A Self-supervised Sentiment
pects have been discussed, highlighting the significance of incor- Analyzer for classification from unlabeled data”.
porating external knowledge, such as knowledge bases, to enhance [22] Shini George and V Srividhya, ”Performance Evaluation of Sentiment
performance.Additionally, a clear transition from syntax-based to Analysis on Balanced and Imbalanced Dataset Using Ensemble Ap-
semantics-based approaches has occurred, highlighting the signifi- proach”.
cance of collecting context and meaning for effective sentiment analy- [23] Monisha V.N, Madhumitha N, Nimishamba B, Poornima N and Dr.
Shashikumar D. R, ”Sentimental Analysis of Movie Reviews using
sis. To ensure broader applicability, the need for standardized datasets
Twitter”.
that encompass diverse domains and languages is emphasized. By [24] Piyush Kumar Soni and Radhakrishna Rambola, ”Deep Learning, Word-
focusing on domain-independent methods, ABSA solutions can be Net, and spaCy based Hybrid Method for Detection of Implicit Aspects
made more adaptable and applicable across various contexts. for Sentiment Analysis”.
In conclusion, this review advances knowledge of sentiment anal- [25] Wenxuan Zhang, Xin Li, Yang Deng, Lidong Bing and Wai Lam,
ysis and aspect detection by shedding light on ABSA’s difficulties ”A Survey on Aspect-Based Sentiment Analysis: Tasks, Methods, and
and developments, notably in the context of restaurant reviews. The Challenges”.
integration of explicit and implicit aspect detection, combined with [26] Ruidan He, Wee Sun Lee, Hwee Tou Ng, and Daniel Dahlmeier, ”An
Unsupervised Neural Attention Model for Aspect Extraction”.
knowledge-based and semantics-driven methods, holds the poten-
[27] Ruidan He, Wee Sun Lee, Hwee Tou Ng, and Daniel Dahlmeier, ”An
tial for more accurate and nuanced sentiment analysis in diverse Interactive Multi-Task Learning Network for End-to-End Aspect-Based
applications and domains. Utilizing cutting-edge approaches will Sentiment Analysis”.
advance aspect-based sentiment analysis and improve comprehension [28] Soujanya Poria, Erik Cambria, Lun-Wei Ku, Chen Gui and Alexander
of sentiments expressed in natural language texts as the discipline of Gelbukh, ”A Rule-Based Approach to Aspect Extraction from Product
NLP continues to develop. Reviews”.

9
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.
[29] Farheen Pathan, Anjali Phaltane and Prof.Sarika joshi., ”DUAL SEN-
TIMENT ANALYSIS”.

10
Authorized licensed use limited to: Zhejiang University. Downloaded on July 13,2024 at 16:17:50 UTC from IEEE Xplore. Restrictions apply.