AN INTELLIGENT SHOPPING COMPANION

providing insights in the methodologies, applications,

and problems posed by these tasks. It analyses lexicon-
Abstract—The proliferation of reviews in e-commerce has
based systems with their
transformed consumer decision-making; however, it also
generates occasions for deceitful practices like reviewing
falsely. Among these techniques is sentiment analysis, a core
area of work in natural language processing. SA rose to simplicity, ML models through their adaptability, and
become a vital tool for identifying the aforementioned neural networks through their capabilities in handling
reviews by analysing textual emotions and opinions. This complex
review embodies the overlap of sentiment analysis with fake
review detection, reflecting key methodologies: lexicon-based
approaches, machine learning models, and advanced linguistic structures. The hybrid approaches are then
architectures of neural networks. The text further explores discussed, such as ABSA and PU learning models,
hybrid techniques that incorporate traditional methods with which combine traditional and more modern
modern approaches, such as aspect-based sentiment analysis approaches.
and PU learning frameworks, to improve detection accuracy.
Challenges discussed include limited labelled datasets, The paper also speaks about the challenges related to
evolving deceptive strategies, and cross-domain
fake review detection-a deficit of labelled datasets,
generalization issues. Propose future research directions on
dynamic behaviour of deceptive reviewers, and
explainable AI, cross-lingual models, and real-time detection
systems. This paper will give the reader an understanding of
limitation in cross-domain generalization. This review,
the state of current developments and highlights how based on state-of-the-art advancements and identifying
importance sentiment analysis bears in winning through the gaps in current research, draws potential opportunities
wars of fake reviews. for innovation.

Keywords—machine learning , sentiment analysis , deep

learning, fake review detection , text summarization . A. SENTIMENT ANALYSIS
Sentiment analysis, also known as opinion mining, is a
I. INRTODUCTION critical area of research and application that focuses on
extracting and analyzing sentiments expressed in textual
Online reviews make up a significant position for data. It plays a significant role in various domains,
shaping consumer decisions, serving as a critical
including marketing, customer service, and social media
source of information in the digital marketplace. From
analysis. The primary objective of sentiment analysis is
electronics shopping to booking hotels and selecting a
to determine the emotional tone behind a series of
restaurant, nowadays users are always relying on
reviews and ratings to assess the quality and reliability words, which can be classified into categories such as
of products and services. With this, the reliance of positive, negative, or neutral. This classification can be
people on online feedback has increased more and further refined into more granular levels, such as very
more, which in turn has amplified deceptive reviews— positive or very negative sentiments, depending on the
insidious content specifically designed to sway context and requirements of the analysis.
consumer perception for individual or organizational
benefit. A trust killer, such deception in online
platforms has very strong economic and social
implications.

Sentiment analysis (SA), a subfield of natural language

processing (NLP), has emerged as a powerful tool to
address this challenge. By extracting and analysing
emotional and opinionated content from textual data,
SA facilitates the identification of patterns indicative
of fake reviews. Techniques such as lexicon-based
methods, machine learning (ML) approaches, and deep
neural networks (DNNs) have been employed to
uncover deceptive practices within review systems. The methodologies employed in sentiment analysis can
be broadly categorized into three main approaches: rule-
This review seeks to clarify the intersection of based, machine learning, and hybrid methods. The rule-
sentiment analysis and fake review detection, based approach relies on predefined lexicons and
handcrafted rules to classify sentiments. This method sadness. This involves using lexicons that map words to
involves identifying sentiment-bearing words and emotional categories combined with machine learning
phrases within the text and applying a scoring classifiers that can discern subtle emotional cues in
mechanism based on their frequency and intensity. language (Cambria et al., 2013). The challenge lies in
While this approach can be effective for straightforward accurately interpreting emotions given the complexity of
sentiment classification tasks, it often struggles with human language; for example, the word "insane" could
nuanced expressions of sentiment, such as sarcasm or imply a positive sentiment in certain contexts but may
idiomatic phrases (Pang & Lee, 2008; Liu, 2012). also evoke fear or panic if interpreted literally.

Recent advancements in deep learning have significantly

enhanced sentiment analysis capabilities by utilizing
neural networks that can process sequential data
effectively. Models like BERT (Bidirectional Encoder
Representations from Transformers) have revolutionized
how sentiment is analysed by providing contextual
embeddings that capture word meanings based on
surrounding text (Li et al., 2021). These models
facilitate more nuanced understanding and classification
of sentiments across various contexts and languages.

Despite these advancements, several challenges remain

in the field of sentiment analysis. One major issue is the
ambiguity inherent in natural language; words can carry
different meanings based on context, making it difficult
for models to achieve high accuracy consistently.
Additionally, the presence of fake reviews complicates
sentiment detection further by introducing noise into
In contrast, machine learning-based sentiment analysis
datasets used for training models (Yousif & Buckley,
leverages algorithms that learn from labelled datasets to
2022). As such, ongoing research is focused on
classify text. This approach typically involves several
developing more sophisticated algorithms that can adapt
stages: preprocessing the text to remove noise (e.g.,
to these complexities while maintaining high levels of
punctuation and stop words), feature extraction to
precision.
identify relevant keywords or phrases, and training
classifiers using algorithms such as Support Vector In conclusion, sentiment analysis has emerged as a vital
Machines (SVM), Naive Bayes, or deep learning tool for understanding public opinion and consumer
techniques like Long Short-Term Memory (LSTM) behaviour in an increasingly digital world. By
networks (Vaswani et al., 2017; Alharbi et al., 2021). employing a combination of rule-based methods,
Machine learning methods have shown considerable machine learning techniques, and advanced neural
promise in improving the accuracy of sentiment networks, researchers and practitioners can extract
classification by capturing complex patterns and valuable insights from vast amounts of textual data. As
contextual information within the data. technology continues to evolve, so too will the
methodologies employed in sentiment analysis, paving
the way for more accurate and context-aware systems
One of the most advanced forms of sentiment analysis is
capable of discerning human emotions from text with
aspect-based sentiment analysis, which focuses on
greater fidelity.
identifying sentiments related to specific attributes or
features of a product or service. For instance, a review
may express satisfaction with a hotel’s location while B. FAKE REVIEW DETECTION
criticizing its cleanliness. Aspect-based analysis allows
businesses to pinpoint strengths and weaknesses in their The detection of fake reviews has become an
offerings by analyzing sentiments associated with increasingly important area of research due to the
different aspects. This method not only enhances the significant impact that deceptive reviews can have on
granularity of sentiment detection but also provides consumer behavior and market dynamics. Fake reviews
actionable insights for product improvement (Sun et al., not only mislead potential buyers but also undermine the
2019; Calikli & Bener, 2017). credibility of legitimate reviews, making it crucial for e-
commerce platforms and businesses to develop effective
detection mechanisms. Various methodologies have
Another critical area within sentiment analysis is
emerged to tackle this problem, leveraging
emotion detection, which aims to identify specific
advancements in machine learning, natural language
emotional states expressed in text, such as joy, anger, or
processing (NLP), and network analysis.
One prominent approach in fake review detection is the The challenge of detecting fake reviews is compounded
use of text-based features. Traditional methods often by the evolving tactics employed by those who generate
rely on linguistic cues, such as sentiment polarity, them. As detection algorithms improve, so do the
review length, and the frequency of certain words or strategies used by fraudsters to evade these systems.
phrases. For instance, researchers have employed This cat-and-mouse dynamic necessitates continuous
supervised machine learning techniques, including innovation in detection methodologies. Researchers are
Support Vector Machines (SVM) and Naive Bayes exploring hybrid models that integrate various
classifiers, to distinguish between genuine and fake techniques—combining text analysis, behavioral
reviews based on these textual characteristics insights, and network features—to create more resilient
(Mukherjee et al., 2013; Jindal & Liu, 2008). These detection systems capable of adapting to new forms of
models are trained on labeled datasets where reviews are deception (Chen et al., 2022; Zhang et al., 2023).
classified as either authentic or fraudulent, allowing
them to learn patterns that are indicative of deception. In conclusion, the landscape of fake review detection is
rapidly evolving with advancements in technology and
However, as the sophistication of fake review strategies methodology. The integration of machine learning
increases, so too must the detection methods. Recent techniques with insights from network analysis and
studies have highlighted the effectiveness of deep behavioral science offers a promising path forward in
learning architectures in improving detection accuracy. combating fake reviews effectively. As researchThe
Models such as Convolutional Neural Networks (CNNs) detection of fake reviews has become an increasingly
and Long Short-Term Memory (LSTM) networks have important area of research due to the significant impact
been utilized to capture complex patterns in review text that deceptive reviews can have on consumer behavior
that may not be immediately apparent through and market dynamics. Fake reviews not only mislead
traditional feature extraction methods (Alharbi et al., potential buyers but also undermine the credibility of
2021; Medina & Torres, 2021). For example, a study legitimate reviews, making it crucial for e-commerce
demonstrated that combining CNNs for text platforms and businesses to develop effective detection
representation with LSTMs for sequential behaviour mechanisms. Various methodologies have emerged to
modelling significantly enhances the detection tackle this problem, leveraging advancements in
performance by capturing both the content and temporal machine learning, natural language processing (NLP),
aspects of reviewer behaviour (Hajek et al., 2020). and network analysis.

Another innovative approach involves graph-based One prominent approach in fake review detection is the
methods, which analyse the relationships between use of text-based features. Traditional methods often
reviewers and products within a network structure. This rely on linguistic cues, such as sentiment polarity,
method recognizes that fake reviewers often exhibit review length, and the frequency of certain words or
clustered behaviours—such as writing multiple reviews phrases. For instance, researchers have employed
for products from the same seller or engaging in supervised machine learning techniques, including
coordinated review campaigns (He et al., 2020; Support Vector Machines (SVM) and Naive Bayes
Hollenbeck et al., 2022). By constructing graphs that classifiers, to distinguish between genuine and fake
represent these interactions, researchers can apply reviews based on these textual characteristics
unsupervised learning techniques to identify anomalous (Mukherjee et al., 2013; Jindal & Liu, 2008). These
patterns that suggest fraudulent activity. This network- models are trained on labeled datasets where reviews are
centric approach has shown promise in detecting fake classified as either authentic or fraudulent, allowing
reviews without relying solely on textual analysis, them to learn patterns that are indicative of deception.
thereby providing a more robust defense against However, as the sophistication of fake review strategies
manipulation (Hajek et al., 2023). increases, so too must the detection methods. Recent
studies have highlighted the effectiveness of deep
Moreover, behavioural features derived from reviewer learning architectures in improving detection accuracy.
activity—such as the frequency of reviews posted by a Models such as Convolutional Neural Networks (CNNs)
user or the time taken to complete a review—have been and Long Short-Term Memory (LSTM) networks have
integrated into detection models to enhance their been utilized to capture complex patterns in review text
effectiveness. Studies have indicated that these that may not be immediately apparent through
behavioural indicators can significantly improve traditional feature extraction methods (Alharbi et al.,
classification accuracy when combined with text-based 2021; Medina & Torres, 2021). For example, a study
features (Feldman & Rosenfeld, 2022; Xu et al., 2019). demonstrated that combining CNNs for text
By analysing how users interact with review platforms representation with LSTMs for sequential behavior
over time, models can better discern genuine user modeling significantly enhances the detection
behaviour from that of spammers.
performance by capturing both the content and temporal C. CHALLENGES IN SENTIMENT ANALYSIS
aspects of reviewer behavior (Hajek et al., 2020). AND FAKE REVIEW DETECTION

Another innovative approach involves graph-based Sentiment analysis and fake review detection face
methods, which analyze the relationships between numerous challenges that stem from the complexities of
reviewers and products within a network structure. This human language, the nuances of social interactions, and
method recognizes that fake reviewers often exhibit the evolving tactics employed by those attempting to
clustered behaviors—such as writing multiple reviews manipulate online reviews. Understanding these
for products from the same seller or engaging in challenges is crucial for developing more effective
coordinated review campaigns (He et al., 2020; algorithms and systems.
Hollenbeck et al., 2022). By constructing graphs that Contextual Understanding and Ambiguity
represent these interactions, researchers can apply
unsupervised learning techniques to identify anomalous One of the foremost challenges in sentiment analysis is
patterns that suggest fraudulent activity. This network- achieving contextual understanding. Words can carry
centric approach has shown promise in detecting fake different meanings based on their context, making it
reviews without relying solely on textual analysis, difficult for models to accurately interpret sentiments.
thereby providing a more robust defense against For instance, phrases like "killing it" can have positive
manipulation (Hajek et al., 2023). connotations in certain contexts but imply negativity in
others (Pang & Lee, 2008). Additionally, the presence of
Moreover, behavioral features derived from reviewer idioms, slang, and cultural references can vary widely
activity—such as the frequency of reviews posted by a across different demographics and regions, complicating
user or the time taken to complete a review—have been sentiment classification (Liu, 2012). The interpretation
integrated into detection models to enhance their of emojis and abbreviations further adds to this
effectiveness. Studies have indicated that these complexity; a smiley face might indicate happiness in
behavioral indicators can significantly improve one context but could be used sarcastically in another,
classification accuracy when combined with text-based leading to potential misinterpretations by sentiment
features (Feldman & Rosenfeld, 2022; Xu et al., 2019). analysis algorithms (Feldman, 2013).
By analyzing how users interact with review platforms Sarcasm and Irony Detection.
over time, models can better discern genuine user
behavior from that of spammers.
Detecting sarcasm and irony presents another significant
hurdle. Sarcasm often involves expressing a sentiment
The challenge of detecting fake reviews is compounded that is opposite to what is literally stated, which can
by the evolving tactics employed by those who generate easily confuse sentiment analysis models that rely on
them. As detection algorithms improve, so do the straightforward word associations. For example, a
strategies used by fraudsters to evade these systems. comment like "Great job on the late delivery!" may
This cat-and-mouse dynamic necessitates continuous appear positive but is actually sarcastic. Current models
innovation in detection methodologies. Researchers are struggle with such nuances because they typically lack
exploring hybrid models that integrate various the deep understanding of context required to discern
techniques—combining text analysis, behavioral intent (Turney, 2002). This limitation underscores the
insights, and network features—to create more resilient need for more sophisticated approaches that can
detection systems capable of adapting to new forms of recognize these subtleties in language.
deception (Chen et al., 2022; Zhang et al., 2023).
Negation Handling
In conclusion, the landscape of fake review detection is
rapidly evolving with advancements in technology and
The presence of negation words—such as "not" or
methodology. The integration of machine learning "never"—can flip the sentiment polarity of a statement
techniques with insights from network analysis and entirely, complicating sentiment classification. For
behavioral science offers a promising path forward in instance, "not bad" conveys a positive sentiment despite
combating fake reviews effectively. As researchers containing the word "not." This complexity necessitates
continue to refine these approaches and develop new that models not only recognize negation but also
ones, it is essential to remain vigilant against emerging understand its impact on the overall sentiment expressed
threats in this critical area of online consumer in a sentence (Hajek et al., 2020). Without effective
protection.ers continue to refine these approaches and handling of negation, sentiment analysis systems may
develop new ones, it is essential to remain vigilant produce misleading results.
against emerging threats in this critical area of online
consumer protection.
Multipolarity in Sentiment
Another challenge is multipolarity, where a single text In the rapidly evolving landscape of digital
contains mixed sentiments towards different subjects or communication, sentiment analysis and fake review
aspects. For example, in a review stating, "I love the detection have emerged as critical fields of study and
product but hate the customer service," traditional application. As consumers increasingly rely on online
models may struggle to provide an accurate overall reviews to inform their purchasing decisions, the
sentiment score since they often reduce sentiments to integrity of these reviews becomes paramount. This
binary classifications (positive or negative) (Singh & review has highlighted the significant advancements
Chatterjee, 2022). This limitation highlights the need for made in sentiment analysis methodologies, including
more sophisticated models capable of dissecting texts the use of machine learning, deep learning, and hybrid
approaches that integrate various data modalities.
into their constituent sentiments and providing nuanced
These techniques have improved the ability to classify
analyses.
sentiments with greater accuracy and nuance, enabling
businesses and researchers to gain deeper insights into
Data Quality and Bias consumer opinions.

Data quality and bias pose significant challenges in both However, despite these advancements, numerous
sentiment analysis and fake review detection. Many challenges persist that hinder the effectiveness of
models are trained on datasets that may be imbalanced sentiment analysis and fake review detection systems.
or contain biased representations of sentiments. If The complexity of human language—characterized by
training data predominantly features positive reviews, ambiguity, sarcasm, irony, and contextual variability—
poses formidable obstacles to accurate sentiment
for instance, the model may become biased towards
interpretation. Furthermore, the presence of negation
predicting positive sentiments even when presented with
and multipolarity within texts complicates sentiment
negative content (Mukherjee et al., 2013). Such skew
classification efforts, necessitating more sophisticated
can lead to inaccurate predictions and undermine the models capable of dissecting sentiments into their
reliability of sentiment analysis outcomes (Abbasi et al., constituent parts.
2008).
In the realm of fake review detection, the evolving
Evolving Tactics in Fake Review Detection tactics employed by fraudsters present an ongoing
challenge. As deceptive practices become more
In fake review detection specifically, identifying sophisticated, detection systems must continuously
adapt to recognize subtle patterns indicative of
deceptive patterns amidst vast amounts of data is a
inauthentic behavior. The integration of behavioral
formidable challenge. Fraudulent reviews often mimic
features alongside textual analysis represents a
legitimate ones in structure and language but may
promising avenue for enhancing detection accuracy;
exhibit subtle differences that are difficult to quantify. however, this complexity adds layers of difficulty to
Advanced techniques such as machine learning and model development.
natural language processing have been employed to
detect these anomalies; however, as deceptive tactics Moreover, issues related to data quality and bias
evolve, so too must detection strategies (Yousif & significantly impact both fields. Many existing models
Buckley, 2022). Moreover, integrating behavioral are trained on imbalanced datasets that may not
features—such as user activity patterns—into detection accurately reflect the diversity of sentiments present in
algorithms can enhance accuracy but also adds real-world scenarios. This bias can lead to skewed
complexity to model development (Xu et al., 2019). predictions and undermine the reliability of sentiment
analysis outcomes. Additionally, the multilingual
Multilingualism nature of online content presents further complications,
as many sentiment analysis systems are primarily
designed for English and may struggle with idiomatic
Finally, multilingualism presents another layer of expressions or cultural nuances in other languages.
difficulty for sentiment analysis systems. Many existing
models are primarily trained on English data and may To address these challenges effectively, ongoing
struggle with idiomatic expressions or cultural nuances research is essential. Future studies should focus on
found in other languages (Cambria et al., 2017). This developing more robust algorithms that can better
limitation can lead to significant inaccuracies when understand context and intent while also improving
analyzing texts written in different languages or dialects. their ability to detect sarcasm and irony. Enhancing
data collection methods to ensure diverse and
representative training datasets will also be crucial in
mitigating bias. Furthermore, interdisciplinary
II. CONCLUSION
approaches that incorporate insights from linguistics,
Conclusion psychology, and sociology may provide valuable
perspectives on human expression that can enrich [7] Turney, P. D. (2002). Thumbs Up or Thumbs
sentiment analysis models. Down? Semantic Orientation Applied to
Unsupervised Classification of Reviews.
In conclusion, while significant progress has been Proceedings of the 40th Annual Meeting of
made in sentiment analysis and fake review detection, the Association for Computational Linguistics
the journey is far from complete. As technology (ACL), 417-424.
continues to evolve and consumer behavior shifts in https://doi.org/10.3115/1073083.1073153
response to new digital landscapes, researchers and
practitioners must remain vigilant in adapting their [8] Singh, A., & Chatterjee, K. (2022). A
methodologies. By addressing the existing challenges Comparative Approach for Opinion Spam
head-on and fostering innovation in these fields, we Detection Using Sentiment Analysis. Springer
can enhance the accuracy and reliability of sentiment Lecture Notes in Networks and Systems.
analysis systems and protect consumers from the https://doi.org/10.1007/978-981-16-6246-
detrimental effects of fake reviews. Ultimately, this 1_43
will contribute to a more trustworthy online ecosystem
that empowers consumers with genuine insights into [9] Mukherjee, A., Venkataraman, V., Liu, B., &
products and services. Glance, N. (2013). Fake Review Detection:
Classification and Analysis of Real and
III. REFERENCES Pseudo Reviews. Proceedings of the 19th
ACM SIGKDD International Conference on
[1] Pang, B., & Lee, L. (2008). Opinion Mining Knowledge Discovery and Data Mining
and Sentiment Analysis. Foundations and (KDD), 443-451.
Trends in Information Retrieval, 2(1-2), 1- https://doi.org/10.1145/2487575.2487583
135. https://doi.org/10.1561/1500000011
[10] Chen, W., Zhang, X., & Dong, L. (2022). A
[2] Liu, B. (2012). Sentiment Analysis and Multi-Task Model for Fake Review Detection
Opinion Mining. Synthesis Lectures on Using Sentiment Analysis. Expert Systems
Human Language Technologies, 5(1), 1-167. with Applications, 201, 117863.
https://doi.org/10.2200/S00416ED1V01Y201 https://doi.org/10.1016/j.eswa.2022.117863
204HLT016
[11] Abbasi, A., Chen, H., & Salem, A. (2008).
[3] Yousif, A. M., & Buckley, J. J. (2022). Impact Sentiment Analysis in Multiple Languages:
of Sentiment Analysis in Fake Review Feature Selection for Opinion Classification
Detection. arXiv preprint arXiv:2212.08995. in Web Forums. ACM Transactions on
https://discovery.researcher.life/article/impact- Information Systems (TOIS), 26(3), 1-34.
of-sentiment-analysis-in-fake-review- https://doi.org/10.1145/1361684.1361685
detection/
dee1853c976131ec8412e75d3b36a398 [12] Feldman, R. (2013). Techniques and
Applications for Sentiment Analysis.
[4] Zhang, S., Zhu, G., & Wei, Z. (2023). Communications of the ACM, 56(4), 82-89.
Building Fake Review Detection Model https://doi.org/10.1145/2436256.2436274
Based on Sentiment Intensity and PU
Learning. IEEE Transactions on Neural [13] Hajek, P., Hikkerova, L., & Sahut, J. M.
Networks and Learning Systems. (2023). Fake Review Detection in E-
https://doi.org/10.1109/TNNLS.2023.3234427 Commerce Platforms Using Aspect-Based
Sentiment Analysis. Journal of Business
[5] Hajek, P., Barushka, A., & Munk, M. (2020). Research, 157, 114143.
Fake Consumer Review Detection Using https://doi.org/10.1016/j.jbusres.2023.114143
Deep Neural Networks Integrating Word
Embeddings and Emotion Mining. Neural [14] Cambria, E., Schuller, B., Xia, Y., & Havasi,
Computing and Applications, 32(13), 9709- C. (2013). New Avenues in Opinion Mining
9725. https://doi.org/10.1007/s00521-020- and Sentiment Analysis. IEEE Intelligent
04757-2 Systems, 28(2), 15-21.
https://doi.org/10.1109/MIS.2013.30
[6] Liu, B., Hu, M., & Cheng, J. (2005). Opinion
Observer: Analyzing and Comparing [15] Tang, D., Qin, B., & Liu, T. (2015).
Opinions on the Web. Proceedings of the 14th Document Modeling with Gated Recurrent
International Conference on World Wide Web Neural Networks for Sentiment Classification.
(WWW 2005), 342-351. Proceedings of the 2015 Conference on
https://doi.org/10.1145/1060745.1060797 Empirical Methods in Natural Language
 Processing (EMNLP), 1422-1432.
https://doi.org/10.18653/v1/D15-1167 [25] Li, Y., Jiang, X., & Li, M. (2021). Exploring
[16] Alharbi, A., De Roure, D., & Zhiwei, Z. Transformer-Based Architectures for Aspect-
(2021). Sentiment Analysis and Fake Review Based Sentiment Analysis. Proceedings of the
Detection Using Deep Learning Techniques. 2021 Conference on Empirical Methods in
Journal of Artificial Intelligence Research, 71, Natural Language Processing (EMNLP),
1347-1364. https://doi.org/10.1613/jair.7285 1248-1258.
https://doi.org/10.18653/v1/2021.emnlp-
[17] Ott, M., Choi, Y., Cardie, C., & Hancock, J. T. main.97
(2011). Finding Deceptive Opinion Spam by
Any Stretch of the Imagination. Proceedings [26] Calikli, G., & Bener, A. (2017). Detection of
of the 49th Annual Meeting of the Association Fake Reviews through Semantic and
for Computational Linguistics (ACL), 309- Sentiment Analysis. Journal of Information
319. Science, 43(5), 695-708.
https://doi.org/10.3115/2002472.2002512 https://doi.org/10.1177/0165551516658006

[18] Rayana, S., & Akoglu, L. (2015). Collective [27] Xu, R., Liu, B., Shu, K., & Yu, H. (2019).
Opinion Spam Detection: Bridging Review Enhancing Fake Review Detection with
Networks and Metadata. Proceedings of the Domain Adaptation. Proceedings of the 2019
21th ACM SIGKDD International Conference World Wide Web Conference (WWW), 1054-
on Knowledge Discovery and Data Mining 1062.
(KDD), 985-994. https://doi.org/10.1145/3308558.3313633
https://doi.org/10.1145/2783258.2783370
[28] Cambria, E., Poria, S., Gelbukh, A., &
[19] Poria, S., Cambria, E., Howard, N., Huang, G. Thelwall, M. (2017). Sentiment Analysis Is a
B., & Hussain, A. (2016). Fusing Audio, Big Suitcase. IEEE Intelligent Systems, 32(6),
Visual and Textual Clues for Sentiment 74-80.
Analysis. Proceedings of the 2016 Conference https://doi.org/10.1109/MIS.2017.4531236
of the North American Chapter of the
Association for Computational Linguistics [29] Akoglu, L., Chandy, R., & Faloutsos, C.
(NAACL), 1449-1458. (2013). Opinion Fraud Detection in Online
https://doi.org/10.18653/v1/N16-1170 Reviews by Network Effects. Proceedings of
the International AAAI Conference on Web
[20] De Smedt, T., & Daelemans, W. (2012). and Social Media (ICWSM), 2-11.
Pattern for Python. Journal of Machine https://ojs.aaai.org/index.php/ICWSM/article/
Learning Research, 13, 2063-2067. view/14403
http://jmlr.org/papers/v13/desmedt12a.html
[30] Mukherjee, S., Malu, G., & Liu, B. (2017).
[21] Heydari, A., Tavakoli, M., Salim, N., & Incorporating Context and Group Influence
Heydari, Z. (2015). Detection of Review for Spam Detection. Knowledge and
Spam: A Survey. Expert Systems with Information Systems, 50(1), 27-51.
Applications, 42(7), 3634-3642. https://doi.org/10.1007/s10115-016-0979-y
https://doi.org/10.1016/j.eswa.2014.12.029
[31] Poria, S., Cambria, E., Bajpai, R., & Hussain,
[22] Sun, L., Wang, S., & He, Q. (2019). A Survey A. (2017). A Review of Affective Computing:
on Aspect-Level Sentiment Analysis. IEEE From Perception to Emotion. Artificial
Transactions on Knowledge and Data Intelligence Review, 48(4), 457-499.
Engineering, 31(12), 2399-2412. https://doi.org/10.1007/s10462-016-9481-x
https://doi.org/10.1109/TKDE.2018.2878259
[32] Severyn, A., & Moschitti, A. (2015). Twitter
[23] Ghosh, S., & Shah, C. (2016). Comparative Sentiment Analysis with Deep Convolutional
Analysis of Feature Selection Methods for Neural Networks. Proceedings of the 38th
Sentiment Analysis. International Journal of International ACM SIGIR Conference on
Artificial Intelligence & Applications, 7(1), Research and Development in Information
33-46. https://doi.org/10.5121/ijaia.2016.7104 Retrieval (SIGIR), 959-962.
https://doi.org/10.1145/2766462.2767830
[24] Jindal, N., & Liu, B. (2008). Opinion Spam
and Analysis. Proceedings of the International [33] Medina, L., & Torres, C. (2021). Sentiment
Conference on Web Search and Data Mining Analysis Using BiLSTM and Attention
(WSDM), 219-230. Mechanisms for Fake Review Detection.
https://doi.org/10.1145/1341531.1341560 Journal of Computational Science, 56,
 101453. You Need. Advances in Neural Information
https://doi.org/10.1016/j.jocs.2021.101453 Processing Systems (NeurIPS), 30, 5998-
6008. https://arxiv.org/abs/1706.03762
[34] Shen, D., Wang, G., & Liu, B. (2018). Cross-
Domain Sentiment Analysis Using Sentiment [37] Alsmadi, I., & O’Brien, K. (2019). Sentiment
Ontology. Knowledge-Based Systems, 150, Analysis Using Machine Learning for Social
168-179. Media. Future Internet, 11(5), 114.
https://doi.org/10.1016/j.knosys.2018.02.034 [https://doi.org/10.3390/fi11050114](https://d
oi.org/10.339
[35] Feldman, R., & Rosenfeld, B. (2022). Fake
Review Detection via Joint Sentiment
Analysis and Reviewer Behavioral Analysis.
Proceedings of the AAAI Conference on
Artificial Intelligence, 3114-3121.
https://doi.org/10.1609/aaai.v36i3.20293

[36] Vaswani, A., Shazeer, N., Parmar, N.,

Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser,
Ł., & Polosukhin, I. (2017). Attention Is All