“Fake

Submitted Capstone
Product
in Department
partial Sudhir
Shreyash
Om
Zeeshan
fulfillment
Diploma
Computer
Mrs.
under painjane
On
of2023
Poonam
the wadekar
A
By Project
Computer
of
of
in shaikh
tukrul
Review
guidance ofDetection”
Engineering
-Engineering
the
2024
Jadhav
requirement
Engineering
for the award of
tukrul
 CERTIFICATE

VIVA COLLEGE OF DIPLOMA ENGINEERING & TECHNOLOGY

VIRAR (W)
2023-24

This is to certify that the Capstone project entitled “Fake product review detection” has been
submitted under the guidance of Poonam Jadhav in partial fulfillment of the requirement for
the award of Diploma of Engineering in Computer Engineering from Maharashtra State
Board of Technical Education.

“Fake Product Review Detection”

GROUP MEMBERS

[Link] shaikh
[Link] painjane
[Link] wadekar
[Link] tukrul

Project Guide H.O. D

Prof. Poonam Jadhav Prof. Nikhil Asolkar
 INDEX

Sr. Name of the topic Page no.

No.

1 Abstract 1

2 Object and Scope of The Project 2-4

3 Methodology 5

4 Requirements For Proposed Project. 6

5 Process Description 7-9

6 References And Bibliography 10

 1.0 Abstract

ABSTRACT

Online spam reviews are deceptive evaluations of products and services. They
are often carried out as a deliberate manipulation strategy to deceive the readers.
Recognizing such reviews is an important but challenging problem. In this
work, I try to solve this problem by using different data mining techniques. I
explore the strength and weakness of those data mining techniques in detecting
fake review. I start with different supervised techniques such as Support Vector
Ma-chine (SVM), Multinomial Naive Bayes (MNB), and Multilayer Perceptron.
The results attest that all the above mentioned supervised techniques
can successfully detect fake review with more than 86% accuracy. Then, I work
on a semi-supervised technique which reduces the dimensionality of the input
features vector but offers similar performance to existing approaches. I use a
combination of topic modeling and SVM for the implementation of the semi-
supervised tech-nique. I also compare the results with other approaches that
consider all the words of a dataset as input features. I found that topic words are
enough as input features to get similar accuracy compared to other approaches
where researchers consider all the words as input features. At the end, I
propose an unsupervised learning approach named as Words Basket Analysis
for fake review detection. I utilize five Amazon products review dataset for an
experiment and report the performance of the proposed on these datasets.
 2.0 Object and Scope of The Project

OBJECTIVE:-

Our study employs statistical methods to evaluate the performance of detection

mechanism for fake reviews and abate the acconcy of this detection. Here, we
present iterature review on studies that applied statistical methods

Sentiment analysis issues

There are several to emsider when conduct in this section, smer issues are
addressed First, the viewpoint for opinion) observed a negative in a simation
might be considered positive in another situation. Second, people do not always
express upon the same way. Most common a processing techniques employ the
fact that minor changes between the two text fagnar unlikely change the accing

Textual reviews

Most of the available reputation models depend on numeric data available in

different fields, an example ratings in e-commerce. Abe most of the reputation
models locas only in the metall ratings of products with condering the news
which are provided by comes [15]. On the other hand, most webs allow conners
to add texmal reviews to provide a detailed opinion about the product [14] [17].
The reviews are available for users to read. Also, customers are lydings reviews
her than on ratings Reputation models can use SA methods to extract users
opinions and use this data in the Reputation syscm. This information may
include consumers opinions about different
Detecting Fake Reviews Using Machine Learning

Filter and fiction fake reviews hasé substanta gnificance (20) Moraes et al [21]
proposed a technique for courting a single pe text new. A sement chuted
document level in applied for stating a sive or positive sentiment Supervised
leaming methods an amused of two phases alycoction and extraction of reviews
utilizing learning models such as SVM

Exmeting the best and most create app and shiny catering the ones witten
reviews in septive or positive opinies has stracted asennon as a major research
field. Although istianidactors phase, there has been a lot of work related to
several languages (22)-124), Our work used several supervised learning algatus
uch as SVM. NB. KNN-IBK, K and DT-148 fur Somm Classification of test to
detect fake reviews

A Comparative Study of different Classification algorithms

Table 1 shows comparative studies on classification algorithms to verify the best

method for detecting fake reviews using different datasets such as News Group
dataset, text documents, and movie reviews dataset. It alsoproves that NB and
distributed keyword vectors (DKV) are accurate about detecting take reviews [12]
and [13] While [11] finds that NB is accurate and a better choice but it is not
oriented for detecting fake reviews. Using the same datasets, finds that SVM is
accurate with stopwords method, but it does not focus on detecting fake reviews,
while [10] finds that SVM is only accurate without using stopwords method, and
also without detecting fake reviews, Sentiment Analysis is a very significant to
detect fake reviews [1]. However, they used only supervisor leaming techniques
based on accuracy and precision. Fundamentally, classification accuracy and
precision only are typically not enough information to obtain a good result.
However, in our empirical study, resulta in three cases with movie reviews dataset
V1.0 and movie reviews dataset V2.0 and movie reviews dataset V30 prove that
SVM is robust and accurate for detecting fake reviews by evaluation of measuring
the performance with accuracy, precision. E-mesure and recall. However, in our
empirical study, results in three cases with movie reviews dataset V1.0 and movie
reviews dataset V2.0 and movie reviews dataset V3.0 prove that SVM is robust
and accurate for detecting fake reviews.
 3.0 Methodology

To accomplish our goal, we analyze a dataset of Product reviews using

the Weka tool for text classification In the proposed methodology, as shown in
Figure 1, we follow some steps that are involved in SA using the approaches
described below.

Step 1: Product reviews collection

To provide an exhaustive study of machine learning algorithms, the

experiment is based on analyzing the sentiment value of the standard dataset.
We have used the original dataset of the Product reviews to test our methods of
reviews classification. The dataset is available and has been used in [13], which
is frequently conceded as the standard gold dataset for the researchers working
in the field of the Sentiment Analysis. The first dataset is known as Product
reviews dataset V1.0 which consists of 1400 Product reviews out of which 700
reviews are positive, and 700 reviews are negative. The second dataset is
known as Product reviews dataset V2.0, which consists of total 2000 Product
reviews, 1000 of which are positive and 1000 of which are negative. The third
dataset is known as Product reviews dataset V3.0, which consists of total 10662
Product reviews, 5331 of which are positive and $331 of which are negative. A
sammmary of the two datasets collected is described in Table II.

Step 2: Data preprocessing

The preprocessing phase includes two prdiminary operations, shown in

Figure 1, which help in transforming the data before the actual SA task. Data
preprocessing plays a significant role in many supervised learning algorithms.
We divided data preprocessing as follows:

StringToWordVector: To prepare the dataset for learning involves

transforming the data by using the
 StringToWordVector filter, which is the main tool for text analysis in
Weka. The StringToWordVector filter makes the attribute value in the
transformed datasets Positive or Negative for all single-words, depending on
whether the word appears in the document or not. This filtration process is used
for configuring the different steps of the term extraction. The filtration process
comprises the following two sub-processes
Tokenization

This sub-process makes the provided document classifiable by converting

the content into a set of fitues ning machine learning.

Stopwards Removal

The stopwords are the words we want to filter out, eliminate, before
training the classifier. Some of those words are commonly used (eg., "a" "the"
"of." "L" "you," "," "and") but do not give any substantial information to our
labeling scheme, but instead they introduce confusion to our classifier In this
study. we wood a 630 English stopwords list with Product reviews datusets.
Stopwords removal helps to reduce the memory requirements while classifying
the reviews.

Attribute Selection

Removing the poorly describing attributes can significantly increase the

classification accuracy, in order to maintain a better classification accuracy,
because not all attributes are relevant to the classification work, and the
irrelevant attributes can decrease the performance of the used analysis
algoritme, an attribute selection scheme was used for training the classifier.

Step 3: Feature Selection

 Feature selection is an approach which is used to identify a subset of
features which are modly related so the target model, and the goal of feature
selection is to increase the level of accuracy. In this study, we implemented one
feature selection method (BestFirst+ CBSubsetEval, Geneticsearch) widely
used for the classification task of SA with Stopwords methods. The results
differ from one method to the other For example, in our analysis of Product
Review datasets, we found that the use of SVM algorithm is proved to be more
accurate in the classification

Step 4: Sentiment Classification algorithms

In this step, we will use sentiment classification algorithms, and they have
been applied in many domains such as commerce, medicine, media, biology,
etc. There are many different techniques in classification method like NB. DT-
148, SVM, K-NN, Neural Networks, and Genetic Algorithm. In this study, we
will use five popular supervised classiters: NB, DT-4S, SVM. K-NN. KStar
algorithms.

• Naive Bayes(NB): The NB classifier is a basic probabilistic classifier

based on applying Hayes theorem. The NB calculates a set of probabilities by
combinations of values in a given dataset. Also, the NB classifier has fast
decision-making process

: Support Vector Machine (SVM): SVM in machine learning is a supervised

learning model with the related learning algorithm, which examines data and
identifies poterms, which is used for regression and classification analysts [25].
Recently, many classification algorithms have been proposed, but SVM is still
one of the most widely and most popular used classifiers.

K-Nearest Neighbor (K-NN) K-NN is a type of lazy learning algorithm

and is a non- parametric approach for categizing objects based on closest
training. The K-NN algorithm is very simple algorithm for all machine
learning. The performance of the K-NN algorithm depends on several different
key factors, such as a suitable distance measure, a similarity measure for
voting, and. k parameter. A set of vectors and class labels which are related to
each vector constitute each of the training data. In the simplest way, it will be
either positive or negative class. In this study, we are using a single number "k"
with values of ke-3. The number decides how many neighbors influence the
classification
KStar (K): K-star (K) is an instance-based classifier. The class of a test
instance is established in the class of those training instances similar to it, as
decided by some similarity function. K algorithm is usually slower to evaluate
the result.
• Decision Tree: The IT-348 approach is useful in the classification
problem. In the testing option, we are using percentage split as the preferred
method

Step 5: Detection Processes

After training, the next step is to predict the output of the model on the testing
dataset, and then a confusion matrix is generated, which classfies the reviews
as positive or negative.
Trie Positive: Real Positive Reviews in the testing data, which are correctly
classified by the model as Pouitive (P
False Positive: Fake Positive Reviews in the testing data, which are incorrectly
classified by the model as Positive (1)
True Negative: Real Negative Reviews in the testing data, which are correctly
classified by the model as Negative (N)
False Negative: Fake Negative Reviews in the testing data, which are
incorrectly classified by the model as Negative (N)
• Trac negative (TN) are events which are real and are effectively labeled as
real, True Positive (TP) are events which are fake and are effectively labeled as
fake. Respectively, False Positives (FP) refer to Read events being classfied as
fakes: False Negativos (FN) are fake events incorrectly classified as Real
events. The confusion matrix, shows numerical parameters that could be
applied following measures to evaluate the Detection Process (DP)
performance. In Table III, the confusion matrix shows the counts of real and
fake predictions obtained with known data, and for each algorithm used in this
study there is a different performance evaluation and confusion matrix.

The confusion matrix is a very important part of our study because we

can classify the reviews from datasets whether they are fake or real reviews.
The confinion matrix is applied to each of the five algorithm discussed in Step
4.

Step 6: Comparison of results

In this step, we compared the different accuracy provided by the dataset

of Product reviews with various classification algorithms and identified the
most significant classification algorithm for detecting Fake positive and
negative Reviews
 4.0 Requirements for Proposed Project.

Software requirements:
1.
 5.0 Process Description

Flow Chart:-

.
6.0 References and Bibliography