Grenze International Journal of Engineering and Technology, Jan Issue

Fake Review Detection and Removal: A Comparative

Analysis using ML and DL Models
Garima1, Vaibhavi2, Yamini Singh3, Rupika Teotia4, Karuna Kadian5, Sunita Garhwal6, and Vimal Dwivedi7
1-5
CSE Department, Indira Gandhi Delhi Technical University for Women, New Delhi, India
Email: {garimakanwaria8, jhavai2201, yaminisingh.yrg, rupikateotia}@gmail.com, [email protected]
6
CSE Department, Thapar Institute of Engineering & Technology, Patiala
Email: [email protected]
7
School of Electronics, Electricals and Computer science, Queen’s University Belfast
Email: [email protected]

Abstract— Reviews are increasingly used for purchase decisions by the customers, they are
important for e-commerce and social networking sites. However, not every review is necessarily
authentic. Researchers have put out a variety of machine learning techniques in the past to
identify false product reviews. Finding the proper machine learning algorithm to spot fake
reviews for a particular type of data is crucial, though. Consequently, in this research, algorithms
such as SVC (Support Vector Classifier), Decision Tree Classifier, Logistic Regression, Random
Forests Classifier, Multinomial Naive Bayes and k-Nearest Neighbors are compared on different
kinds of datasets like Amazon dataset, Yelp Dataset and TripAdvisor Hotel Reviews Dataset.
Accuracy, Recall, Precision, and F1-Measure evaluation findings are the basis for the
comparison. The findings of this study indicate that, Support Vector classifier comes out to be
the best performing algorithm for detecting fake reviews when compared with the other five
techniques. While the k-Nearest Neighbors algorithm has the worst performance.

Index Terms— Algorithms; Reviews; Machine Learning; Fake; Comparative Analysis; Dataset
- Amazon, Yelp, TripAdvisor; Spam;

I. INTRODUCTION
Reviews are incredibly important when making decisions in today’s electronic commerce. The majority of
consumers check reviews of products or businesses before choosing what to purchase and where to purchase it,
and whether or not to buy it. The existence of these comments can be a source of information. Businesses might
use it, for instance, to choose how to produce their products or services. But regrettably, some people take
advantage of the significance of reviews by fabricating reviews with the intention of either boosting the popularity
of the product or undermining it.
It’s customary for people to research products before making a purchase. Customers may evaluate several brands
and decide on a certain product based on reviews. These internet reviews have the power to alter a consumer’s
perception of a product. If these reviews are accurate, it may be easier for people to choose a product that meets
their needs. On the other side, if the reviews are falsified or manipulated, the user may be misled.
The suggested system provided an e-commerce website with product reviews that aid users in making decisions
about what to buy. Every user has a distinctive profile on social networking and e-commerce websites in order to

Grenze ID: 01.GIJET.10.1.28

© Grenze Scientific Society, 2024
submit a distinctive order. Users may search for things there and read customer reviews on them. These reviews
will assist customers in determining the product’s quality and assisting them in making the best choice. However,
the reality is that this product review for each item cannot assist clients pick the best one.

Fig. 1. Process Diagram

Customer review websites have encountered problems as a result of the way users freely express and use their
comments. Platforms for social media like Twitter, Facebook, YouTube, and others let anybody express criticism
of any business at any time with no restrictions or commitments. Because there are no constraints, some companies
utilize social media to unjustly advertise their products, stores, or brands while disparaging those of their rivals
[11].
To address this issue, many online marketplaces and retailers have implemented various techniques to detect and
remove fake reviews. These techniques include the use of machine learning algorithms, natural language
processing, and manual moderation by human experts.
In this paper, we focus on the problem of detecting and removing fake product reviews using a combination of
machine learning algorithms and natural language processing techniques. We propose an approach that analyzes
various features of product reviews, such as sentiment, grammar, and vocabulary, to identify fake reviews. Our
approach can help online marketplaces and retailers improve the quality of their reviews, maintain consumer trust,
and ultimately, enhance the overall user experience.
The research on review spam identification [24], [29] is extremely valuable since it can ensure the validity of
reviews, decrease the cost of cleaning up fraudulent reviews on ecommerce sites and social networking sites, and
give consumers a better purchasing experience. Good product evaluation will increase consumer willingness to
acquire items while also protecting merchants’ and consumers’ rights and interests. Manufacturers can also utilize
false review detection technology to gather authentic review data, assess customers’ true feelings, keep updating
products, and raise the standard of service by adhering to the procedure described in figure 1.

II. LITERATURE REVIEW

The most critical stage in any type of study is a review of the literature. Before constructing a model or presenting
a solution, it is necessary to examine past publications in the domain. Based on the study, the author can forecast
or produce the disadvantage and begin working with the mention of earlier papers.
Researchers have been utilizing supervised learning for detecting fake reviews for years, however the truth of the
underlying big data sets is still inaccessible, Instead of actual false reviews, the majority of current supervised
learning methods rely on pseudo-fake reviews. This give the first study on false review detection in Chinese
utilizing filtered reviews from Dianping’s fake review detection system in [16].
The difficulties in detecting bogus reviews are explored in [27]. The method put out in this paper functions with
both labelled and unlabeled data. The study shows that supervised machine learning is 90.19% accurate, and semi-
supervised machine learning is 83.70% accurate. This paper investigated and discussed all possible aspects and
combines the experiment and results using the provided dataset.
In the suggested research [6], a dataset of reviews in Urdu and Roman is created, and the n-gram approach [1] is
used to identify fake reviews in different languages. The research revealed that the most efficient means of
identifying fake reviews is text categorization with an SVM classifier.
Several online data mining strategies are discussed in [20]. These techniques are further divided into three
categories: internet structure, web content, and web usage mining. For instance, Naive Bayes, Neural Network,

201
and Support Vector Machine are all used in web content mining. These various kinds each employ a different
methodology, tool, strategy, and algorithm to extract information from the vast amounts of data on the internet.
A supervised method is applied to train on Yelp’s filtered reviews in [21]. All currently used supervised learning
methods rely on pseudo-false reviews in place of falsified reviews screened by a reputable website. Using actual
Yelp data, this study assessed and evaluated performance using established research approaches. Unexpectedly,
behavioral factors perform better than language features [31].
The “Amazon’s Yelp” dataset is utilized in [3], and two models, Random Forest and Naive Bayes, are created to
compare model performance. As a result, the Random Forests model outperformed the Naive Bayes method by a
wide margin. The problem of fake review detection is treated honestly, with the goal of selecting the best suitable
algorithm to complete the work of detection and eradication. The research study examined linguistic elements in
[30] such as unigram frequency, bigram frequency, unigram presence, and review length to identify fraudulent
reviews on the Yelp dataset. The key issue is data scarcity, and the model requires both language and behavioral
elements to function
[14]. Online reviews’ influence on businesses has significantly increased in 2018, and they are now critical in
assessing business performance in a wide range of industries, from hotels and restaurants to e-commerce.
Unfortunately, some consumers create bogus reviews of their businesses or competitors’ businesses in order to
improve their internet reputation. Previous research has concentrated on detecting false reviews in a range of
industries, such as restaurant and hotel product or company reviews. Consumer electronics companies have not
yet been thoroughly studied, despite their economic importance [4].
The majority of research on fake review identification is done using supervised, binary text classification [9]. Since
it involves extracting (metadata) features from review text, representing them in a machine-processable format
(feature representation), and teaching a prototype that can generalize patterns based on the features and apply the
patterns to previously unexplored data, it is comparable to product classification (algorithm). [28] classified this
metadata as “lexical” or “non-lexical.” Words, n-grams, punctuation, and latent themes are instances of lexical
features, which are textual properties. Non-lexical elements include metadata on reviews (such as ratings or stars)
or associated authors.
A detailed examination of opinion-rich product reviews [12], which are frequently used by manufacturers and
consumers alike, and considers how opinion spam differs from web spam and email spam, requiring various
detection techniques to manage this. It analyses such spam actions in millions of Amazons’ reviews and generates
models utilizing Logistic regression, feature identification such as review centric, product centric, and reviewer
centric, type 1 - type 2 spam detection, and so on.
The survey assesses and summarizes current feature extraction techniques [19] in order to identify gaps according
to two categories: learning algorithms and traditional statistics [7], [8]. Additionally, they conduct research on the
effectiveness of other converters and neural network models that have not yet been used to identify fraudulent
reviews. RoBERTa beats state-of-the-art approaches in a blended domain, with an accuracy of 91.2%, according
to the experimental results on two baseline methods, and can be used as a baseline for further study.

TABLE I. DESCRIPTION OF PUBLIC DATASET SOURCES

Dataset Dataset (Source) Description
Jindal & Liu (2008) [12] dataset link 58,38,041 reviews, 12,30,915 products 21,46,057
Blitzer et al. (2007) [5] dataset link reviewers. Product reviews for 4 categories, including
Ni et al. (2019) [22], kitchen, books, dvds, and electronics.
Amazon datasets 142.8million reviews about different products.
McAuley et al. (2015) [18], He & McAuley (2016)
[10]
dataset link
800 positive ratings,
Ott et al. (2011) [25] dataset link 400 of which are genuine and 400 are fraudulent.
TripAdvisor datasets
Ott et al. (2013) [24] dataset link 800 bad ratings,
400 of which are real and 400 of which are fraudulent
TripAdvisor Li & Ott et al. (2014) [17] dataset link authentic reviews: 1200 fake reviews: 1636 three domains:
and restaurant, hotel etc
Yelp datasets
Mukherjee et al. (2013) [21]
(YelpCHI dataset) dataset link 67,395 reviews from 200 hotels and restaurants by 38,063
Rayana and Akoglu (2015) [26] (YelpNYC dataset) reviewers 3,59,052 reviews from 932 restaurants by
dataset link 1,60,225 reviewers 6,08,598 reviews from 5,044 restaurants
Yelp datasets Rayana and Akoglu (2015) [26](YelpZIP dataset) by 2,60,277 reviewers.
dataset link 9456 authentic reviews and
Barbado et al. (2019) [4] 9456 fake reviews from four
(Write to [email protected] if you wish to acquire the US cities
dataset for analysis.)
202
The study proposes a method for identifying false reviews that takes into account the actions of reviewers as well
as characteristics obtained from the reviews’ content [13]. In other words, their goal is to create a white-box model
that will enable people to comprehend what is happening with respect to their personal data, as opposed to
Blackbox models, which are difficult to generally describe, like deep learning.

III. DATA AND METHODS

A. Datasets
The datasets for this study were obtained from Kaggle. The first dataset includes the category of the product, rating
of that particular product, label (CG or OR) and review text. The dataset has 40432 occurrences and 4 ascribed
characteristics. As there are 20216 fake reviews and the same number of true reviews, the dataset is balanced. For
the second dataset this study considers Yelp product reviews consisting of fields like Review, Product-id, Rating,
User-id, Date and Label. The third dataset include TripAdvisor hotel reviews which consists of fields like
Deceptive, Hotel, Polarity, Source and Review text. To assist future researchers, relevant information has been
identified and summarized for convenience and summarized in Table I.
B. Classification Algorithms
To find the false reviews, a variety of supervised learning classification algorithms are used, such as Random
Forest Classifier, Decision Tree Classifier, k-Nearest Neighbors, Logistic Regression, Multinomial Naive Bayes
and Support Vector Classifier.
• Logistic Regression: Machine learning categorization issues are solved using logistic regression. They are
utilized to predict categorical variables and are comparable to linear regression. It can predict the outcome as 0
or 1, or Yes or No. Instead of the exact integers, it provides probabilistic values between 0 and 1.
• k-Nearest Neighbors: This approach can also solve problems related to classification and regression. This is a
simple algorithm that sorts new instances by having at least k neighbors agree, then keeps all existing instances.
The class with the most traits that the trade has in common wins the trade. This calculation is done using the
distance function.
• Support Vector Classifier: Unprocessed data can be represented as points in an n-dimensional space, where n is
the number of attributes that an individual has. SVM techniques allow users to classify their data. Since each
feature’s value is now associated with a specific coordinate, the data may then be simply categorized. Classifier
lines can be used to categorize data into categories and plot them.
• Decision Tree Classifier: a supervised learning method for classifying situations. It may be used to efficiently
categorize dependent variables that are categorical or continuous. This method divides a population into two or
more homogeneous sets based on the most important trait or independent variable.
• Random Forest Classifier: It refers to a set of decision trees. Every tree has a class given to it, and each tree
“votes” on that class to categorize new elements according to their properties. The forest’s most popular category
is selected (out of all trees in the forest).
• Multinomial Naive Bayes: The foundation of a Naive Bayes classifier is the idea that the presence of one feature
in a class does not preclude the presence of any further characteristics. Even if they are connected, a Naive
Bayes classifier takes into account each of these characteristics separately when calculating the likelihood of a
certain result. Simple to create and effective for huge data sets are naive Bayes models. Despite being
straightforward, it is known to outperform even the most sophisticated categorization methods.

IV. PROPOSED METHODOLOGY

The model consists of various phases of execution present in figure 2 which are explained as follows:
Data acquisition: Data must be acquired from various sources through web scraping then the data which is acquired
needs to be processed first in order to create a rich and productive data set [15]. This dataset should contain both
genuine and fake reviews. Genuine reviews are those that are written by actual customers who have purchased and
used the product. Fake reviews are those that are written by individuals who have not used the product or have
been paid to write a review.
1. Data pre-processing: Unstructured data is transformed into a useful and efficient format through data
preparation. The data pre-processing phase includes various steps such as data cleaning. Noise removal and
filtering. Noise is nothing more than a large amount of unwanted and irrelevant data that does not meet the model’s
data criteria, that must be removed for better performance and accurate results. Data cleaning removes data noise.

203
A. Data Cleaning: The data must be critical and data should be error-free and clear of unnecessary data. As a result,
the material will be cleaned prior to proceeding to the next phase. Data cleansing includes checking and eliminating
missing values, duplicate records, and malformations.
B. Data Transformation: A statistical transformation of a set of data is referred to as a data transformation. The
data is initially transformed into a suitable form for such a data mining procedure. This enables one to organize
hundreds of entries in order to better comprehend their data. Normalization, standardization, and attribute selection
are all examples of transformations.
2. Feature extraction: A dimensionality reduction technique called feature extraction breaks large amounts of raw
data into smaller, easier-to-process groups. The common feature of these huge data sets is that they have many
variables that require a lot of processing power to process. The process of selecting and/or combining variables
into features is called feature extraction [32]. This method considerably minimizes the amount of information that
must be processed while characterizing the actual data set precisely and completely.
3. Model Training: Using the features that were retrieved, a machine learning model is then trained. Different
machine learning algorithms can be applied. The algorithm is trained using data that has been labelled, with real
reviews being classified as positive and false reviews as negative. Data can be categorized in both organized and
unstructured forms. The task of classifying a given data set into groups is known as classification. The initial stage
in the technique is to forecast the class of a given data point. Classes are frequently referred to by the terms target,
label, and class. Classification predictive modelling refers to the issue of approximating the function of mapping
specific input factors to specific output variables [2]. The key task is to find out which class or category the new
data will supposedly belong to.
4. Removal: After classification, the data is clearly classified into two categories, genuine reviews and fake
reviews, and using an appropriate machine learning model, the fake review class is removed. And to also ensure
the reduction of fake reviews in a system figure 3 contains the detailed steps followed for fake reviews removal.
5. Model Evaluation: After the model has been trained, its effectiveness is assessed using a different test dataset.
A variety of measures are used to evaluate the model’s performance, including accuracy, precision, recall, and F1
score.
6. Deployment: The model must then be used to identify and eliminate fraudulent reviews from the online
marketplace or business. The model can be used in real-time to automatically flag and remove fake reviews or
provide a score indicating the likelihood of a review being fake.

V. RESULT AND DISCUSSION

After testing all of the algorithms on three separate data sets, the Amazon product review data set, the Yelp Hotel
Review data set, and the TripAdvisor Review data set, this study was conducted to determine that Support Vector
Machine outperformed all other classifiers. Table II shows the exact accuracy of all machine learning approaches
on various data sets. To facilitate comparing the results, the output of the Deceptive (D) and Truthful (T) review
types produced by the classifier are evaluated with respect to Precision (P), Recall (R), F1-Score (F1), and
Accuracy (Acc). With a mean predictive accuracy of more than 88% , the Support Vector Machines Classifier
predicted the fraudulent reviews the most accurately.
Logistic Regression, which had an average prediction accuracy of just over 86%, was immediately behind it. The
Multinomial Naive Bayes method and the Random Forests Classifier predicted with an average precision level of
about 85%. The Decision Tree Classifier, however, was able to anticipate bogus reviews with an average accuracy
of just over 75%. The prediction accuracy of the k-Nearest Neighbors method, which was the worst performing
algorithm, was only about 61%.
The analysis of false positive and false negative rates to determine how well various machine learning models can
categorize and differentiate between fake and authentic reviews is provided with these formulas

204
After applying the ML model (Support Vector machine algorithm in above case) to detect fake reviews, this model
will remove the review and will also ensure that the same device cannot be used to create more such fake reviews
shown in Fig 3. (With his or her account ID and password, the user will log into the system, see various products,
and provide product reviews. To establish whether a review is legitimate or fake, the algorithm will determine the
user’s IP address. The system will alert the administrator and request that the review must be deleted from the
system if it discovers numerous fraudulent reviews sent from the same IP address).
Even though the overall model may have more time and spatial complexity, none of the current systems combine
these two strategies. The Classification Report and the Confusion Matrix of the Machine Learning algorithm in
the mentioned datasets (Amazon, Yelp and TripAdvisor) are presented in figures 4, 5, 6 correspondingly.

Fig. 4. Amazon Dataset - SVM

Fig. 5. Yelp Dataset – SVM

There are several challenges associated with detecting and removing fake product reviews:
 Volume of reviews: As the number of reviews for a product grows, it becomes increasingly difficult to
manually review every review. This makes it challenging to detect fake reviews, particularly when there are
a large number of them.
 Diversity of language: Fake reviews can be written in a wide range of language styles, making it difficult to
develop Algorithms that can accurately detect them.
 Mimicking genuine reviews: As fake review detection algorithms improve, so too do the techniques used by
reviewers to create fake reviews that closely mimic genuine ones. This makes it challenging to develop
algorithms that can Accurately identify fake reviews.
 Platform policies: Platforms like Amazon have policies that prohibit fake reviews, but they are not always
effective in detecting and removing them. Additionally, the policies themselves can be difficult to enforce in
a consistent manner.

205
 Legal considerations: There are legal considerations associated with the removal of reviews, particularly when
they are posted by individuals or companies that feel they have been wronged. This can make it challenging
to remove fake reviews without causing legal issues.

TABLE II ACCURACY TABLE OF DIFFERENT ALGORITHMS ON VARIOUS DATA SETS

Algorithms Classification Report
Detection Revie Amazon Dataset Yelp Dataset TripAdvisor Dataset
Techniques w
P R F1 Acc(%) P R F1 Acc(%) P R F1 Acc(%)
Type
Multinomial D 0.8 0.8 0.8 0.0 0.0 0.0 0.7 0.9 0.8
Naive-Bayes 2 9 5 0 0 0 9 4 6
84.92 88.07 84.46
T 0.8 0.8 0.8 0.8 1.0 0.9 0.9 0.7 0.8
8 1 4 8 0 4 3 5 3
Logistic D 0.8 0.8 0.8 0.0 0.0 0.0 0.8 0.8 0.8
Regression 7 6 6 0 0 0 6 6 6
86.42 88.07 86.25
T 0.8 0.8 0.8 0.8 1.0 0.9 0.8 0.8 0.8
6 7 7 8 0 4 6 6 6
k-Nearest D 0.5 0.9 0.6 0.1 0.1 0.1 0.6 0.9 0.7
Neighbors 4 7 9 0 8 3 0 5 3
57.37 71.87 65.36
T 0.8 0.1 0.3 0.8 0.7 0.8 0.8 0.3 0.5
7 8 0 8 9 3 8 5 1
Decision Tree D 0.7 0.7 0.7 0.0 0.1 0.0 0.6 0.6 0.6
3 6 4 9 0 9 7 8 7
73.87 76.45 67.14
T 0.7 0.7 0.7 0.8 0.8 0.8 0.6 0.6 0.6
5 2 4 8 5 6 7 7 7
Random Forest D 0.8 0.8 0.8 0.0 0.0 0.0 0.7 0.8 0.8
Classifier 0 9 4 0 0 0 8 6 2
83.48 88.07 81.07
T 0.8 0.7 0.8 0.8 1.0 0.9 0.8 0.7 0.8
8 8 3 8 0 4 5 6 0
Support Vector D 0.8 0.8 0.8 0.0 0.0 0.0 0.8 0.8 0.8
Machine 9 8 8 0 0 0 7 8 7
88.16 88.07 87.32
T 0.8 0.8 0.8 0.8 1.0 0.9 0.8 0.8 0.8
8 8 9 8 0 4 8 7 7

To improve the detection of fake reviews, future research can consider incorporating behavioral factors such as
the frequency of reviews, time taken to complete a review, and the ratio of positive to negative reviews. This
approach can also be used to identify spammer communities by linking reviews to groups and finding the review
with the highest correlation. Additionally, researchers can investigate domain specific features that distinguish real
from fake reviews, such as sentiment analysis, feature-based analysis, and aspect-based analysis.
Semi-supervised and unsupervised learning techniques that use both huge volumes of unlabeled data and small
amounts of labelled data, in addition to conventional supervised machine learning techniques, may also be
successful in detecting false reviews. Data augmentation and generative models can be used to generate synthetic
data for training models.
As NLP techniques advance, they can play a critical role in identifying false reviews. Researchers can explore new
architectures like Transformer-based models to handle the complexity of text data more accurately [23].
Ensembling, which combines the outputs of several machine-learning models, can improve the overall accuracy
of fake review detection. Researchers can develop ensemble models that use different features to identify signs of
fakery in reviews, such as overly positive language, a lack of detail, or suspicious patterns in timing or location.
Removing fake reviews is another challenge that researchers can address. They can develop techniques for
identifying and removing fake reviews in real-time, as well as methods for removing past fake reviews.
Overall, as new technologies and methods emerge, the future of fake review detection looks promising. By
developing more sophisticated and accurate methods, we can ensure that consumers can trust the online reviews
they read and make informed purchasing decisions.

VI. CONCLUSION
The above study proved the relevance of fake reviews identification and how they affect the online business.
A comparative analysis of existing machine learning approaches is presented. The suggested method is tested
using the Amazon Dataset, Yelp Dataset, and TripAdvisor Dataset. The developed technique uses a variety of
classifiers. According to the findings, SVM classifier performs better than the other classifiers at predicting
fraudulent reviews as shown in
206
 Fig. 7. Accuracy of ML Algorithms on Different Datasets

figure 7 having a mean predictive accuracy over 88%. The behavioral characteristics are not considered in the
study that is being presented. Future research may take into account additional behavioral aspects, such as those
that rely on how frequently reviewers execute reviews, how long it takes them to finish evaluations, and how
frequently reviewers make favorable or negative ratings. It is anticipated that adding more behavioral traits would
improve the performance of the approach for spotting bogus reviews.

REFERENCES
[1] Hadeer Ahmed, Issa Traore, and Sherif Saad. Detection of online fake news using n-gram analysis and machine learning
techniques. In International conference on intelligent, secure, and dependable systems in distributed and cloud
environments, pages 127–138. Springer, 2017.
[2] Hadeer Ahmed, Issa Traore, and Sherif Saad. Detecting opinion spams and fake news using text classification. Security
and Privacy, 1(1):e9, 2018.
[3] Syed Mohammed Anas and Santoshi Kumari. Opinion mining based fake product review monitoring and removal system.
In 2021 6th International Conference on Inventive Computation Technologies (ICICT), pages 985–988. IEEE, 2021.
[4] Rodrigo Barbado, Oscar Araque, and Carlos A Iglesias. A framework for fake review detection in online consumer
electronics retailers. Information Processing & Management, 56(4):1234–1244, 2019.
[5] John Blitzer, Mark Dredze, and Fernando Pereira. Biographies, bollywood, boom-boxes and blenders: Domain adaptation
for sentiment classification. In Proceedings of the 45th annual meeting of the association of computational linguistics,
pages 440–447, 2007.
[6] Nazir M Danish, Sarfraz M Tanzeel, Nasir Usama, Aslam Muhammad, AM Martinez-Enriquez, Adrees Muhammad, et
al. Intelligent interface for fake product review monitoring and removal. In 2019 16th International Conference on
Electrical Engineering, Computing Science and Automatic Control (CCE), pages 1–6. IEEE, 2019.
[7] Elshrif Elmurngi and Abdelouahed Gherbi. Detecting fake reviews through sentiment analysis using machine learning
techniques. IARIA/data analytics, pages 65–72, 2017.
[8] Elshrif Elmurngi and Abdelouahed Gherbi. An empirical study on detecting fake reviews using machine learning
techniques. In 2017 seventh international conference on innovative computing technology (INTECH), pages 107–114.
IEEE, 2017.
[9] Julien Fontanarava, Gabriella Pasi, and Marco Viviani. Feature analysis for fake review detection through supervised
classification. In 2017 IEEE International Conference on Data Science and Advanced Analytics (DSAA), pages 658–666.
IEEE, 2017.
[10] Ruining He and Julian McAuley. Ups and downs: Modeling the visual evolution of fashion trends with one-class
collaborative filtering. In proceedings of the 25th international conference on world wide web, pages 507–517, 2016.
[11] Haruna Isah, Paul Trundle, and Daniel Neagu. Social media analysis for product safety using text mining and sentiment
analysis. In 2014 14th UK workshop on computational intelligence (UKCI), pages 1–7. IEEE, 2014.
[12] Nitin Jindal and Bing Liu. Opinion spam and analysis. In Proceedings of the 2008 international conference on web search
and data mining, pages 219–230, 2008.
[13] Nour Jnoub and Wolfgang Klas. Declarative programming approach for fake review detection. In 2020 15th International
Workshop on Semantic and Social Media Adaptation and Personalization (SMA, pages 1–7. IEEE, 2020.
[14] Parisa Kaghazgaran, James Caverlee, and Majid Alfifi. Behavioral analysis of review fraud: Linking malicious
crowdsourcing to amazon and beyond. In Proceedings of the International AAAI Conference on Web and Social Media,
volume 11, pages 560–563, 2017.
[15] Raymond YK Lau, SY Liao, Ron Chi-Wai Kwok, Kaiquan Xu, Yunqing Xia, and Yuefeng Li. Text mining and
probabilistic language modeling for online review spam detection. ACM Transactions on Management Information
Systems (TMIS), 2(4):1–30, 2012.
[16] Huayi Li, Zhiyuan Chen, Bing Liu, Xiaokai Wei, and Jidong Shao. Spotting fake reviews via collective positive-unlabeled
learning. In 2014 IEEE international conference on data mining, pages 899–904. IEEE, 2014.

207
[17] Jiwei Li, Myle Ott, Claire Cardie, and Eduard Hovy. Towards a general rule for identifying deceptive opinion spam. In
Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),
pages 1566–1576, 2014.
[18] Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton Van Den Hengel. Image-based recommendations on styles
and substitutes. In Proceedings of the 38th international ACM SIGIR conference on research and development in
information retrieval, pages 43–52, 2015.
[19] Rami Mohawesh, Shuxiang Xu, Son N Tran, Robert Ollington, Matthew Springer, Yaser Jararweh, and Sumbal Maqsood.
Fake reviews detection: A survey. IEEE Access, 9:65771–65802, 2021.
[20] Muhammd Jawad Hamid Mughal. Data mining: Web data mining techniques, tools and algorithms: An overview.
International Journal of Advanced Computer Science and Applications, 9(6), 2018.
[21] Arjun Mukherjee, Vivek Venkataraman, Bing Liu, and Natalie Glance. What yelp fake review filter might be doing? In
Proceedings of the international AAAI conference on web and social media, volume 7, 2013.
[22] Jianmo Ni, Jiacheng Li, and Julian McAuley. Justifying recommendations using distantly-labeled reviews and fine-grained
aspects. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th
international joint conference on natural language processing (EMNLP-IJCNLP), pages 188–197, 2019.
[23] Ray Oshikawa, Jing Qian, and William Yang Wang. A survey on natural language processing for fake news detection.
arXiv preprint arXiv:1811.00770, 2018.
[24] Myle Ott, Claire Cardie, and Jeffrey T Hancock. Negative deceptive opinion spam. In Proceedings of the 2013 conference
of the north american chapter of the association for computational linguistics: human language technologies, pages 497–
501, 2013.
[25] Myle Ott, Yejin Choi, Claire Cardie, and Jeffrey T Hancock. Finding deceptive opinion spam by any stretch of the
imagination. arXiv preprint arXiv:1107.4557, 2011.
[26] Shebuti Rayana and Leman Akoglu. Collective opinion spam detection: Bridging review networks and metadata. In
Proceedings of the 21th acm sigkdd international conference on knowledge discovery and data mining, pages 985–994,
2015.
[27] Jitendra Kumar Rout, Amiya Kumar Dash, and Niranjan Kumar Ray. A framework for fake review detection: issues and
challenges. In 2018 International Conference on Information Technology (ICIT), pages 7–10. IEEE, 2018.
[28] Joni Salminen, Chandrashekhar Kandpal, Ahmed Mohamed Kamel, Soon-gyo Jung, and Bernard J Jansen. Creating and
detecting fake reviews of online products. Journal of Retailing and Consumer Services, 64:102771, 2022.
[29] Sunil Saumya and Jyoti Prakash Singh. Detection of spam reviews: a sentiment analysis approach. Csi Transactions on
ICT, 6(2):137–148, 2018.
[30] Kolli Shivagangadhar, H Sagar, Sohan Sathyan, and CH Vanipriya. Fraud detection in online reviews using machine
learning techniques. International Journal of Computational Engineering Research (IJCER), 5(5):52–56, 2015.
[31] Chengai Sun, Qiaolin Du, and Gang Tian. Exploiting product related review features for fake review detection.
Mathematical Problems in Engineering, 2016, 2016.
[32] Eka Dyar Wahyuni and Arif Djunaidy. Fake review detection from a product review using modified method of iterative
computation framework. In MATEC web of conferences, volume 58, page 03003. EDP
Sciences, 2016.

208