Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

Implementation of fake product review monitoring system and

real review generation by using data mining mechanism
Mupparam Sowjanya,K.Shnati latha,Ch.hyma,K.Naresh
1,2,3,4
Asst.Prof,CSE department,Anurag Group of Institutions
1
[email protected],[email protected],[email protected],4nareshcse@cvsr.
ac.in
Abstract: Most of the people requires genuine information about the online product. Before
spending their economy on particular product can analyze the various reviews in the website.
In this scenario, they did not identify whether it may be fake or genuine. In general, some
reports in the websites are good, company technical people itself add these for making the
product famous. These people belong to media and social organization teams, they give
reviews with a good rating by their own firm. Online purchasers did not identify the fake
product because of this falsification in the reviews of the website. In this research, the SVM
classification mechanism has been used for detect the fake reviews by using IP address. This
implementation helpful for users find out the correct review of online product. In this
accuracy is improved by 98.79%, F1 score increases by 10%.

Keywords: Fake reviews, data mining, online product, real time marketing.

Introduction:

As the general public of the people require audit approximately an object earlier than
spending their coins at the object. So individuals pass over one of a kind audits inside the
website online however those surveys are veritable or counterfeit isn't recognized via the
consumer. In a few survey sites some tremendous audits are blanketed by using the item
enterprise people itself a good way to make so that you can create bogus tremendous item
surveys. They provide super audits for a few, diverse items fabricated via their personal firm.
Client won't have the option to look if the audit is certifiable or counterfeit. To find out
counterfeit survey inside the web page this "Phony Product Review Monitoring and Removal
for Genuine Online Product Reviews Using Opinion Mining" framework is offered. This
framework will discover counterfeit surveys made through posting counterfeit remarks
approximately an item with the aid of distinguishing the IP address along audit posting
designs. Client will login to the framework making use of his customer id and secret phrase
and will see specific items and will supply survey approximately the object. To find out the
audit is phony or actual, framework will find out the IP address of the consumer if the

Volume XII, Issue II, 2020 Page No: 635

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

framework watch counterfeit survey ship by a similar IP Address numerous multiple

instances it's going to train the administrator to expel that survey from the framework. This
framework utilizes statistics mining gadget. This framework encourages the client to find out
proper survey of the object.

In present days the usage of Internet and internet based marketing has gotten mainstream. A
high-quality many objects and administrations are available in web primarily based
showcasing that create huge degree of statistics. Consequently, it is difficult to discover the
quality affordable administrations or objects perfect to the prerequisite. Clients
straightforwardly take preference dependent on audits or conclusions that are composed via
others depending on their encounters. Right now any individual can compose something, this
boost the quantity of phony audits. Different agencies are employing individuals to compose
counterfeit tremendous audits approximately their administrations or items or out of line bad
surveys approximately their opponents' administrations or gadgets. This process gives
incorrect contribution to the brand new customers who wish to buy such things and therefore
we need a framework to differentiate such phony audits and expel them. Right now have a
look at exclusive directed, unaided and semi controlled statistics digging approaches for
counterfeit audit identification depending on diverse highlights.

Literature survey:

As of late, the World Wide Web has greatly changed the technique for imparting the insights.
Online audits are feedback, tweets, posts, conclusions on numerous on line tiers like survey
destinations, news locales, net based totally enterprise destinations or some different lengthy
range interpersonal communication destinations. Sharing audits is one of the strategies to
compose a survey approximately administrations or items [1] [2]. Surveys are considered as a
person's near domestic concept or experience about items or administrations [7] [13]. Client
dissects reachable audits and takes preference whether to shop for the object or no longer [3].
In this way on-line audits are critical wellspring of information approximately patron
conclusions [5]. Phony or spam audit alludes to any spontaneous and superfluous facts about
the item or administration. Spammer composes counterfeit audits approximately the
contenders' item and advances possess items [8] [10]. The surveys composed by spammers
are called phony audits or unsolicited mail audits [2]. In this manner counterfeit surveys
discovery has grown to be fundamental trouble for customers to decide higher preference on
items dependable simply because the sellers to make their purchase [15].

Volume XII, Issue II, 2020 Page No: 636

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

Methodology:

Commentator Centric Approach-This method is predicated upon the behavior of analysts.

This methodology considers data about clients and all surveys that are composed with the aid
of them [1]. Highlights utilized proper now account age, profile image, URL duration, IP
address, variety of composed audits through one commentator, maximum severe rating every
day and so on. Item Centric Approach-This technique for the most component facilities
around the object related information. Right now, rank of object, value of object and so on
are taken into consideration as highlights. At first phony audit identification turned into
supplied via Jinal et al. [12]. There are special methods to differentiate counterfeit surveys.
AI system is one of the procedures to distinguish counterfeit surveys. AI model learns and
make forecast [2]. The essential advances associated with AI are records making ready,
highlight extraction, include determination, characterization model age. This technique is
appeared in Fig. 1:

Figure 1: Fake review detection system.

Different processes were proposed in past to understand counterfeit audits dependent on

styles of information like marked records (as an example, directed learning), unlabeled data
(as an instance, unaided learning), and in element named records (for instance, semi-
regulated discovering) that is portrayed underneath. A. Directed Learning approaches Wael et
al. [6] make use of administered studying calculation for counterfeit audit discovery. Before
applying the arrangement technique, diverse pre-getting ready steps are played out; these

Volume XII, Issue II, 2020 Page No: 637

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

method contain stemming, evacuation of accentuation stamps and forestall phrase expulsion.
They utilize etymological thing to differentiate counterfeit audits. Semantic element
incorporates POS and sack of-words. Sack of-words highlights contain of individual word or
collecting of phrases which can be observed in given content. At that point unique grouping
calculations are applied like preference tree, arbitrary backwoods, bolster vector device,
credulous bayes and inclination supported bushes. Here gullible bayes and bolster vector
device provide better final results. Jitendra et al. [2] applied various highlights dependent on
content material closeness and feeling extremity for spotting phony and proper surveys. Here
creators use assessment rating depending on slant extremity among high quality and terrible
surveys, phonetic and unigram as spotlight. They at that factor implemented three
calculations 1) bolster vector machine, 2) guileless bayes and 3) choice tree. Snehasish et al.
[3] utilizes directed AI calculation. Right now, audits are separated from veritable surveys
making use of 4 phonetic pieces of statistics like degree of element, understandability,
attention markers and composing fashion. Level of element includes exceptional
operationalized highlights like guidance, logical detail, lexical assorted range, paintings
phrases and perceptual element. Instruction became determined through POS (grammatical
characteristic) like component, movement phrase, modifier, action words, intensifiers,
pronouns and so forth. Relevant element contains spatial and brief references whilst
perceptual detail includes feeling and visible words, quantity of aural. Lexical diverse range
words comprise non content material phrases that beat the diploma of element within the
audits. Composing style is predicated upon usage of capitalized, lower case, question marks,
all accentuation, tenses, and feelings. Tenses became predicted dependent on collecting of
destiny, over a vast time span irritating words. Perception marker depends on speculative
phrase, motion phrases, rejection words simply as causal phrases and so on. The creators
make use of various controlled learning calculations like strategic relapse, C4.5, lower back
unfold device, credulous bayes, bolster vector system using polynomial bit, bolster vector
system making use of immediately portion, bolster vector device with radical premise group
bit, casting a poll, k-closest neighbor and arbitrary woodland. The phony and validate audits
are analyzed against two baselines. Standard 1 include one of a kind issue like person
consistent with word, duration of audit in phrases, first individual solitary words, lexical
decent variety, emblem references, first person plural phrases, poor feeling word and fine
feeling words. Second pattern includes action phrases, qualifiers, modifier, phrases per
sentence, individual per word, modular action words, all accentuation, first man or woman
plural words, first man or woman precise phrases, spatial phrases, paintings phrases, fleeting

Volume XII, Issue II, 2020 Page No: 638

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

words, emotiveness, visible words, believing words, aural phrases, terrible feeling phrases
and wonderful feeling phrases. Second sample offers steadily unique final results contrasted
with first gauge.

System works as follows: -

 Admin will add gadgets to the framework.

 Admin will erase the audit that is phony.
 User as soon as get to the framework, patron can see object and can post audit about
the item.
 System will comply with the IP deal with of the consumer.
 If the framework watches counterfeit audit originating from equal IP address several a
couple of times this IP cope with might be followed with the aid of the framework and
will advise the administrator to expel this survey from the framework.

Modules:

1. The framework incorporates 2 giant modules with their sub-modules as follows:

 Administrator Login: Admin login to the framework utilizing his administrator ID and
mystery key.

Add item:

Admin will upload object to the framework.

Delete Review:

Admin will evacuate the survey which observed by means of the framework as phony.

2. Client Login:

 User will login to the framework making use of his patron ID and mystery phrase.

View object:

User will see object.

Post Review:

User can submit audit approximately the item.

Tracks IP Address:

Volume XII, Issue II, 2020 Page No: 639

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

If the framework reveals an audit is phony it will light up the administrator to expel the
phony survey.

Table: 1.Review Assessment

QF 90 80 70 60 50

electronics 0.983 0.959 0.91 0.83 0.77

Home
appliances 0.987 0.962 0.92 0.85 0.764

Groceries 0.989 0.957 0.915 0.82 0.77

Table 1 explains that the quality factor analysis with respect to different types of product
reviews which are discussed in above table.1. Here the Groceries achieves more review
factor rather than other to digital review techniques.

Figure 2.Review Analysis System

Fig.2 explains that digital review rating system with respect to 2D mechanism. In this
situation 3 level transformation mechanism has been used to extract the reviews with
windowing techniques by coefficients. At this stage efficient review is obtained but
robustness is required to improve.
∑𝑖 ∑𝑗 𝑤(𝑖,𝑗) ∑𝑖 ∑𝑗 𝑤 1 (𝑖,𝑗)
NC =
∑𝑖 ∑𝑗 𝑤(𝑖,𝑗)2

Volume XII, Issue II, 2020 Page No: 640

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

255
Review = 20log ( )
√𝑁𝑅

Table 2: Review with Respect to Falser

parameter review_100 review_100 review_100

0.99 0.994 0.98

electronics
Home 0.995 0.995 0.992
appliances
0.94 9 0.94 0.97
Groceries
0.96 0.967 0.98
electronics
Home 0.88 0.89 0.89
appliances

The above table explains that different types of reviews on online trading. In this all
corrections and elements has attain with efficient manner but extraction is complex procedure
compared to traditional machine learning models.

Figure 3 Graphical Representations of Reviews.

Fig 3 explains that different reviews on online systems, at this stage some of techniques has
been failed because of conventional insecure methods. These are limitations in [8].

Volume XII, Issue II, 2020 Page No: 641

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

Table 3 SVM method analysis

FUZZY logic[10]
SVM method
Attacks Bit error rate (%)
Website- 1 Website- 2 Website- 3 Bit error rate(%)
electronics 10 4.19 6.23 1.12
Home appliances 5.91 3.12 1.13 5.91
Groceries 8.61 7.94 4.75 5.91
electronics 16.14 1.89 9.02 5.92
Home appliances 6.74 3.59 4.13 5.91
electronics 15.72 3.91 5.79 5.92
Home appliances 1.82 1.84 1.47 5.92
Groceries 21.72 1.71 1.78 5.93
electronics 9.23 9.23 9.23 5.94
Home appliances 1.82 1.87 1.89 5.92
electronics 0.32 0.12 0.21 0.3
Home appliances 7.12 5.2 7.92 0
Groceries 0.34 0 0 0
electronics 7.42 0 0 0
electronics 0.052 0 0 9.23
Home appliances 0.83 0.023 0.21 9.23
Groceries 10 2.91 2.91 9.23
electronics 0.052 0 0 9.23

Table 3 explains that different types of review which are discussed above with respect to bit

error rate, existed methods are compared with proposed FUZZY [10] has been compared and

conclude that SVM is a best digital review identification technique. But, needs improvement

at image sharing.

Volume XII, Issue II, 2020 Page No: 642

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

25

20

15
FUZZY logic[10] Bit error rate(%)
Website- 1
FUZZY logic[10] Bit error rate(%)
10
Website- 2
FUZZY logic[10] Bit error rate(%)
Website- 3
5
SVM method Bit error rate(%)

0
electronics

electronics

electronics

electronics

electronics

electronics
Home appliances

Home appliances

Home appliances

Home appliances

Home appliances

Home appliances
Groceries

Groceries

Groceries

Groceries

Groceries

Groceries

Figure 4: Overall Review Analysis.

Conclusion:
Due to rapid improvement of internet the size of fake and real reviews are increases. Because
of this huge reviews there is no food review has been identified. Some of the false reviews
causes’ bad selection of products happen, genuinity is missing in that product. Therefore in
this research SVM machine based false review detection is designed with the help of python
software. In this work explains about various fake review detection techniques based on
supervised and unsupervised methodologies. This existed methods are gives less accuracy,
have more limitations at identification of fake reviews. This SVM false review detection
system gives the accuracy by 98.79% and F1 score increases by 10%.

References:

[1] A. Rastogi, M. Mehrotra, “Opinion spam Detection in Online Reviews”, Journal of

information and Knowledge Management, vol. 16, no. 04, pp. 1-38, 2017.
[2] J. Rout,S. Singh, S. Jena, and S. Bakshi, “Deceptive review detection using labeled and
unlabelled data”, Multimedia Tools and Applications,vol.76, no. 3, pp. 3187-3211, 2016.
[3] S. Banerjee, A. Chua, J. Kim, “Using Supervised Learning to Classify Authentic and Fake
Online Reviews ”, Proceeding of the 9th International Conference on Ubiquitous Information
Management and Communication”, ACM, 2015.

Volume XII, Issue II, 2020 Page No: 643

Journal of Xi'an University of Architecture & Technology Issn No : 1006-7930

[4] P.Rosso, D.Cabrera, M. Gomez, “Using PU-Learning to Detect Deceptive Opinion Spam”,
pp.38-45, 2013.
[5] R.Narayan,J. Rout and S. Jena, “Review Spam Detection Using Semisupervised Technique”,
Progress in Intelligent Computing Techniques: Theory, Practice, and Applications, pp. 281-286, 2018.
[6] W. Etaiwi,G. Naymat, “The impact of applying preprocessing steps on review spam
detection”, The 8th international conference on emerging ubiquitous system and pervasion networks,
Elsevier, pp. 273-279, 2017.
[7] W. Zhang,R. Y. K. Lau and Li. Chunping, “Adaptive Big Data Analytics for Deceptive
Review Detection in Online Social Media”, Thirty Fifth International Conference on Information
Systems, Auckland 2014,pp.1-19,2014.
[8] C. Lai, K. Xu, R. Y. Lau, Y. Li, and L. Jing, “Toward a Language Modeling Approach for
Consumer Review Spam Detection,” 2010 IEEE 7th International Conference on E-Business
Engineering, pp. 1–8, 2010.
[9] M. I. Ahsan, T. Nahian, A. A. Kafi, M. I. Hossain, and F. M. Shah, “Review spam detection
using active learning,” 2016 IEEE 7th Annual Information Technology, Electronics and Mobile
Communication Conference (IEMCON), 2016.
[10] M. Ott, Y. Choi, C. Cardie and J.T. Hancock, “Finding deceptive opinion spam by any stretch
of the imagination”, ACM, pp.309- 319,2011.
[11] N. Jindal and B. Liu., “Opinion spam and analysis”, Proceedings of the international
conference on Web search and web data mining - WSDM 08 (2008), ACM, pp. 219 –230,2008.
[12] N. Jindal and B. Liu, “Review spam detection”, Proceedings of the 16th international
conference on World Wide Web - WWW 07 (2007), ACM, pp. 1189–1190, 2007
[13] S. Shojaee, A. Azman, M. Murad, N. Sharef and N. Sulaiman, “A Framework for Fake
Review Annotation”, 2015 17th UKSIM-AMSS International Conference on Modelling and
Simulation, IEEE, pp. 153-158,2015.
14] J. Koven, H. Siadati, and C. Y. Lin, “Finding Valuable Yelp Comments by Personality,
Content, Geo, and Anomaly Analysis,” 2014 IEEE International Conference on Data Mining
Workshop, pp. 1215–1218, 2014.
[15] S. Banerjee and A.Y.K. Chua. 2014. “Applauses in hotel reviews: Genuine or deceptive ?”,
2014 Science and Information Conference (2014), pp. 938–942,2014.

Volume XII, Issue II, 2020 Page No: 644