Identifying Malicious Reviews Using NLP and Bayesian Technique on Ecommerce Historical Data

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 03 | Mar 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1776
Identifying Malicious Reviews Using NLP and Bayesian
Technique on Ecommerce Historical Data
Sagar Ashokrao Mahajan
GovernmentCollegeofEngineering, Aurangabad, Dept. of Computer Science & Engineering, Aurangabad
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract: These days, online item audits assume an essential part in thebuychoiceofshoppers.Ahighextentofpositiveaudits
will bring significant deals development, while negative surveys will cause deals misfortune. Driven by the huge monetary
benefits, numerous spammers attempt to advance their items ordowngradetheirrivals'itemsbypostingphonyandone-sided
online surveys. By enlisting various records or delivering assignments in publicly supporting stages, numerous individual
spammers could be coordinated as spammer gatherings to control the item audits together and can be additionally harming.
Existing deals with spammer bunch discovery separate spammer bunch applicants fromauditinformationanddistinguishthe
genuine spammer bunches utilizing unaided spamicity positioning strategies. As a matter of fact, as per the past examination,
marking few spammer bunches is simpler than one expects, nonetheless, hardly any techniques attempt to utilize this
significant named information. In this paper, we propose a halfway administered learning model (PSGD) to distinguish
spammer gatherings. By naming some spammer bunches as certain examples, PSGD applies positive unlabeled learning (PU-
Learning) to examine a classifier as spammer bunch indicator from positive occasions (marked spammer gatherings) and
unlabeled cases (unlabeled gatherings). In particular, we remove solid negative set regarding the positive occasions and the
unmistakable highlights. By joiningthepositiveexamples, extricatednegativeoccasionsandunlabeledoccurrences, weconvert
the PU-Learning issue into the notable semi supervised learning issue,andafterwardutilizea NaiveBayesianmodel andanEM
calculation to preparea classifierforspammerbunchdiscovery.ExaminationsongenuineAmazon.cninformational indexshow
that the proposed PSGD is viable and outflanks the cutting edge spammer bunch location techniques.
Keywords- Information search and retrieval, NLP, opinion mining, opinion feature.
I Introduction
Opinion mining can likewise be alluded to as estimation investigation whose objective is to break down individuals' opinions,
mentalities, and feelings toward substances, functions, and their characteristics. An opinion assumes a significant part in
dynamic. Slants or opinions explained in audits are analyzed at a scope of goals. This is material to people just as for
associations moreover. At the point when a few associations need to investigate the opinions of clients about their items and
administrations, they can direct overviews.
Clients post their perspectives and opinions on the seller's webpage or on their websites, discussions, and social destinations.
In the present life, profoundly accessible CGM for example shopper produced media like message sheets, wikis, gatherings,
online journals, and news stories board huge accommodation however they are liable for some presentation too. For the
formation of some new creative open doors which are favorable to buyers, endeavors canexaminepurchasercreatedmedia to
understand client's assessment about their items and administrations. At the point when certain issues arenotsettledquickly
and effectively, obliviousness of such shopper created media can influence and produce hazards in brand picture and
undertaking impact on the lookout, in light of the fact that the transmission speed of CGM data could reestablish obstinate
consideration over the web. Ground-breakingdiagnosticmodelsare required whicharefavorableinthe evaluationofcustomer
conclusions.
Document level opinion mining distinguishes the general subjectivity or assessment communicated on an element ina survey
report, yet it doesn't connect opinions with explicit parts of the substance. Purchasers of the item are perpetually discontent
with the opinion rating of that item. People groups are more intrigued to know why it gets the rating, positive just as contrary
ascribes that impacts on conclusive rating of item. In this way, it is basic to mine the exact opinionated features from text
surveys and partner them to opinions. In opinion mining, an opinion featuresshowsan elementora propertyofa substanceon
which clients expresses their opinions.
Opinion mining incorporates opinion include whose errand is to determine an element or a quality of a substance on which
purchasers express their perspectives and opinions. Proposed framework perceives such features fromunstructuredliterary
audits. In opinion mining, a lot more methodologies have been now proposed which unique opinion features . To remove

opinion include from surveys, regulated learning model work in givendomainjustyetthemodel mustberetrainedinthe event
that it is applied to various domains [1], [4].
Unsupervised learning approaches incorporate regular language handling (NLP) which utilizesdomain-freesyntacticformats
or laws. These layouts and rules are utilized to catch the reliance jobs and nearby setting of the element terms. In any case,
these standards are not material to genuine surveys since they needlegitimateplan.Ruleswhicharenotinlegitimatestructure
can't function admirably on informal genuine surveys. Point displaying approaches can extricate coarse-grained and
nonexclusive subjects, which are really semantic element groups of the exact features remarked on explicitly in reviews [3].
A spammer bunch comprises of a bunch of commentators who co-audits a bunch of normal items.Accordingly,theassessment
mining procedure could be used utilizing NLP to extricate the gatherings [12], [13]. Be that as it may, since numerous clients
might be fortuitously assembled due to the comparative interest, the gatherings removed by FIM are just the spammer bunch
applicants and should be additionally checkedtodistinguishthegenuinespammergatherings.Consequently,therecognitionof
spammer bunches for the most part contains two stages: (I) Discover spammer bunch competitors, (ii) Identify the genuine
spammer bunches from the applicants.
II Literature Review
Opinion mining, which is likewise alluded as notion investigation, incorporates advancement of a framework which ready to
amass and characterize opinions of a buyers about an item. Mechanized opinion mining regularly utilizes AI, a sort of man-
made brainpower (AI), for the reason to dig text for slant. Data accessible in text configuration can be classified into realities
and opinions. A reality speaks to the target explanations about substances and functions where asopinionsrepresentabstract
proclamations. Opinions emulate individuals' feelings about the elementsandfunctions.Opinionsgavebybuyersintextaudits
are inspected from archive, sentences remembered for thatreportand wordandexpressionsrememberedforthatrecord[11].
Objective of such sort of report level (sentence-level) opinion mining is to arrange the general subjectivity or notion
communicated in an individual audit archive (sentence). Assessment of writings at the record or the sentence level does
represents the opinions of clients, for example, differentpreferences.Apositivearchivedoesn'tspeak totheall sureopinions of
customers on features of specific article. Also, a negative reportdoesn'trepresentall negativeopinionsofclientson features of
specific item [7]. Text record which incorporates assessments holds both positive and negative parts of specific article or
element as per client's perspectives.
For the most part, generally assessment on the articlemaycontainsomesureviewpointsandsomenegativeperspectives.Solid
examination of features level is needed to discover total viewpoints about item or element.
For this reason three significant errands are as per the following:
1) Identifying object features
2) Determining opinion directions
3) Grouping equivalent words Identify object features search out for intermittent things and thing phrases as features,which
are generally true features.
Existing data extraction techniques which are appropriate for recognizing object highlights are as restrictive irregular fields
(CRF), shrouded Markov models (HMM). Determining opinion directions close whether the opinions given by buyer on the
highlights of article or element are positive, negative or impartial. Existing vocabulary based methodology utilizes opinion
words and expressions in a sentence to choose the direction of an opinion on a component. One article highlights can be
communicated with various words or expressions, gathering equivalents task gathering's equivalent words together.
To compute sentence subjectivity Hatzivassiloglou and Wiebe [16] presents supervised grouping strategy to figure sentence
subjectivity. Hatzivassiloglou and Wiebeproposedthegeneral impactsofdynamicmodifiers,semanticallysituateddescriptors,
and gradable descriptors on anticipating subjectivity of the content report holding audits.
Ache and Lee [11] proposed a sentence-level subjectivity finder for the reason to discover the sentences in a record as either
emotional or objective. This strategy holds abstract sentences and disposes of the goal sentences. After then they applied
opinion classifier. Errand of assumption classifier is to digest come about subjectivity with improved outcomes.

To order whole film surveys into positive or negative slants, Pang et al. [14] presented AI model as guileless Bayes, greatest
entropy, and backing vector machines. They finish up results created by standard AI techniques are better than result by
human-produced baselines. In any case, AI strategy performs well on just customary subject based order and need usefulness
on assessment arrangement. An unsupervised learning strategy was proposed to order audit records intopositiveor negative
in which as approval spoke to inspiration of report and disapproval speaks to antagonism of archive [8]. Past work indicated
that customary assessment examination approaches can be very successful.
To mechanize the investigation of estimation materials, various methodologies were utilized for theforecastforthe notions of
words, articulations and furthermore archives [7], which incorporate Natural Language Processing (NLP) and examplebased
[8]–[11], AI calculations, for instance NB, ME, SVM [12], and unsupervised learning [13].
Kim and Hovy [14] first created an equivalent arrangement of competitor words with obscure feelings.
Govindarajan [15] proposed a strategy for slant examination on eatery audits utilizing half and half order innovation. While
most analysts center around AI based opinioninvestigation,otherscenteraroundextremityvocabularies basedstrategies[16].
Kamps et al. [17] decided word assessment direction in the wake of figuring their semantic separationwiththeirbenchmarks
in the WordNet [18] equivalent structure graph.
Wang et al. [19] first examined the characters about the assumption phrases in the NTUSD extremity word bank to get their
polarities and qualities dependent on their characters. Cambria [20] received human-PC communication, data recovery and
multi-modular sign handling advances to extricate individuals' assumptions among the ever-developing on the web social
information base. Since every one of the above investigations had restricted inclusion [21] and deficiencies in forecast, we
should think about semantic fluffiness when buildingslantvocabulary.Thispaperproposedanothermethodology,forexample
Multi-Strategy estimation examination dependent on semantic fluffiness,whichisa blendofAIandopinionvocabulariesbased
methodology.
Normal assessment directions of expressions and words are determined of each audit record to envision estimationofsurvey
report. To register conclusions of expressions in audit record, domain-subordinate relevant data is utilized however this
method has impediment as it relies upon outside web crawler. Zhang et al. [6] presented a standard based semantic
investigation method to arranged notions for text surveys. Word reliancestructuresareutilizedtoarrangethesuppositionofa
sentence.
Zhang et al. anticipated record level assessments by totaling notions of sentence. This procedure has restriction as rule-based
techniques experience helpless presentation as they don't hold completeness in their guidelines.
To stay away from this, Maas et al. [15] introduced technique for both report level and sentence-level slant grouping. This
proposed technique utilizes blend of unsupervised and supervised ways to deal with learn vectors.Forlearningmeasure,they
catch semantic term-report data just as rich conclusion content. It is fundamental to take note of that opinion mining of the
report, sentence, or expression (word) level doesn't figure out what precisely individuals loved and hatedinaudits. Itneglects
to consolidate the recognized suppositions and comparablehighlightsremarkedonintheaudits.Obviously,a removedopinion
without the comparing features (opinionated objective) is of restricted an incentive in all actuality [2].
Opinion Feature Extraction Opinion featuresextractionisa subproblemofopinionmining.Existing methodsofopinioninclude
extraction can be classified into two classifications as, supervised and unsupervised. To check highlights or parts of watched
elements, supervised learning consolidates concealedMarkovmodelsandcontingent arbitraryfields.Thisisotherwisecalled a
joint auxiliary labeling issue. Despite the fact that supervised models perform well on given domain, they required broad
retraining when utilized in a few domains.Toutilizesupervisedmodelsinvariousdomains,movelearningmeasureisrequired.
Unsupervised Natural language Processing NLP techniques use mining of syntactic examples of highlights to extract opinion
highlights. Unsupervised methodologies decide syntactic relations between include terms and opinionwordsinsentences.To
decide relations unsupervised methodologiesutilizecreatedsyntacticguidelinesorsemanticjob marking[10].Thisconnection
helps to find highlights related with opinion words just as mine huge number of invalid highlights of online surveys. With the
end goal of extraction of successive itemsets .
Hu and Liu [12] presented an affiliation rule mining (ARM) method which depends on recurrence of itemsets. Regularitemset
comprises of potential opinion highlights, which are things and thing phrases with high sentence-level recurrence. Yet, this
strategy has limitations as: 1) successive yet invalid highlights are separated erroneously, and 2) uncommon yet legitimate

highlights might be disregarded. Su et al. [8] proposed a common support grouping (MRC) procedure to handle include based
opinion mining issues. Common support bunching techniques are utilized to mine the relations between features classes and
opinion word gatherings. Extraction measure reliesupon a coevent weightlatticeproducedfromthegivensurvey corpus. MRC
additionally ready to separate inconsistent highlights if the shared connections among features and opinion bunches found
through the grouping stage is precise. MRC's exactness is low as it experiences difficultiesinacquiringgreatgroups ongenuine
surveys.
Latent Dirichlet portion (LDA) [13] approaches have beenusedtosettleperspectivebasedopinionminingundertakings.These
LDA's are developed for extraction of dormant subjects which may not be opinion highlights communicated explicitly in
surveys. Regardless of whetherthesemethodologiesareuseful indeterminingoffundamental structuresofsurveyinformation,
they might be less productive in determining exact element terms remarked in audits. In our proposed framework, we group
some syntactic reliance arrangements to mine up-and-comer highlights, as in NLP approachesandweusetheSPAMANDNON
SPAM FEATURES strategies to classify the favored domain-explicit opinion highlights. The vital differentiation of SPAM AND
NON SPAM FEATURES contrasted with existing techniques lies in its shrewd combination of domain-ward and domain free
data sources.
III Proposed System
NLP consolidates Syntactic assessment and Semantic examination. Feature that discontinuously appears in the given review
domain, and routinely appears outside the domain, for instance, in a domain-self-governingcorpusknownasExplicitfeatures.
Subsequently, domain-explicit opinion features will implyevenmoreonandoninthedomaincorpusofstudies whenstoodout
from a domain-independent corpus.
Figure 1 Proposed Architecture Flow
Fig. 1 portrays the progression of proposed technique. By utilizing physically expressed syntactic principles, from the audit
corpus we first mine top notch of up-and-comer highlights. Later we process SPAM FEATURES and NON SPAM FEATURES for
each preoccupied applicant highlight. SPAM FEATURES portrays the factual relationship of the possibilitytothegivendomain
corpus and EDR, which reproduce the measurable significance of thecontendertothe domainindependentcorpus. Competitor
highlights with SPAM FEATURES increases more noteworthy than a predefined Implicit significance edge and NON SPAM
FEATURES scores not exactly Explicit pertinence limit are affirmed as legitimate opinion highlights.
A. Candidate Feature Extraction
Opinion highlights are comprised of things or thing phrases, by and large which are develop as the subject orobjectofanaudit
sentence. On account of reliance language structure, the subject opinion features has a syntactic relationship of type subject

action word (SBV) with the sentence predicate. The article opinion include has a reliance relationship of verbobject (VOB) on
the predicate. Moreover, it additionally has a reliance relationship of relational word object(POB)ontheprepositional wordin
the sentence. From the previously mentioned reliance relations, i.e., SBV, VOB, and POB, we present three syntactic guidelines
as follows:
Table 1.0 Syntactic Rules (Heuristic Rules)
Rules Interpretation
NN+SBV Identify NN as CF, If NN has a SBV dependency relation
NN+VOB Identify NN as CF, If NN has a VOB dependency relation
NN+POB Identify NN as CF, If NN has a POB dependency relation
Component of the technique of competitor features extraction is as per the following: 1) First of all, to perceive the syntactic
association of each sentence in the given audit corpus, relianceparsing(DP)isutilized. 2)Later,thethreeprinciplesin depicted
in Table 1 are applied to the perceived reliance structures, and when a standard is terminated, the relating things or thing
phrases are mined as set of up-and-comerhighlights.Proposedcompetitorinclude extractiontechniqueislanguagedependent.
B. Opinion Feature Identification
Domain Relevance Domain importance depicts the manner by which a term is identified with a specific domain dependent on
two kinds of measurements as scattering and deviation. Scattering checks the number of quantities of times a term is alluded
across records by figuring the distributional importance of that term across various archives in the total domain which is
commonly known as level centrality. Deviation reflects how consistently a term is uncover in a specific record by ascertaining
its distributional essentialness in the archive which is commonly known as vertical centrality. Scattering and deviation are
registered by using the recurrence converse record recurrence (TF-IDF) term loads. Proposed domain reliance measure has
two distinct perspectives as:
• Actually, domain reliance was acquainted with track news functions by choosing theme words from function words
verbalized in the reports. Proposed work doesn't recognize point and functions As another option, proposed strategy use the
proposed domain significance as a measure to characterize opinion highlights from unstructured content audits.
• As proposed framework doesn't separate among subject and function words. Domain significance recipe is modified for
evaluating between corpus measurements uniqueness; especially, it istunedtobind thedistributional incongruitiesofopinion
includes crossways two domain.
C. Implicit and Explicit Domain Relevance
Implicit Domain Relevance represents domain importance of specific opinion includes determined on given domain specific
survey corpus. SPAM FEATURES duplicate the exact substance of the component to the domain audit corpus.
Explicit-domain importance is estimated by domain significance of specific opinion features on given domain independent
survey corpus. NON SPAM FEATURES shows the factual relationship of the element to the domain-free corpus. As applicant
terms are identified with possibly one corpus or other. They never identified with both simultaneously. In such case, NON
SPAM FEATURES likewise shows the immateriality of an element to the given domain audit corpus. There some generally
normal terms that are used all over the place and furthermore in an audit corpus as highlights.
IV Performance Analysis
The experiments were conducted using the customer reviews using Amazon dataset.
A system based on the proposed techniques has been implemented in Java using Stanford NLP classifier.The proposedsystem
evaluated from three perspectives:
 The effectiveness of feature extraction.

 The effectiveness of opinion sentence extraction.
 The accuracy of orientation prediction of opinion sentences.
For each sentence in a review, if it shows user’s opinions, all the features on which the reviewer hasexpressedhis/heropinion
are tagged. Whether the opinion is positive or negative (i.e., the orientation) is also identified. If the user gives no opinion in a
sentence, the sentence is not tagged as we are only interested in sentences with opinions in this work. For each product, we
produced a feature list. All the results generated by our systemarecompared withthe manuallytaggedresults.Taggingisfairly
straightforward for both product features and opinions. A minor complication regardingfeaturetaggingisthatfeaturescanbe
extrinsic or intrinsic in a sentence. Most features appear in opinion sentences, e.g., pictures in “the pictures are absolutely
amazing”. Both extrinsic and intrinsic features are easy to identify by the human tagger.
Another issue is that judging opinions in reviews can be somewhat subjective. It is usually easy to judge whether an
opinion is positive or negative if a sentence clearly expresses an opinion. However, deciding whether a sentence offers an
opinion or not can be debatable.
Table 1.0 Confusion Matrix
Tables given below shows the TP, TN, FP, FN, Precision, Recall, F-score for Amazon Dataset. Based on the results generated of
confusion matrices we computed precision, recall l. Table contains recall and precision of frequentfeaturegenerationfor each
product, which uses association mining.
The precision is improved significantly by this pruning. The recall stays steady. The recall level almost does not change. The
results from clearly demonstrate the effectiveness of these two pruning techniques. The precision drops a few percents on
average.
The reviews represent typical user generated content; they are all written by customers/users instead of being authored by
any professional editor. Texts exhibit a rather informal style they often lack correct English grammar, exhibit the use of slang
words, and contain an above average amount of misspellings. Web crawls are mono lingual,all crawleddocumentsarewritten
in English.
Table 1.0 NLP Classification Results
Table Data Precision Recall F-score
Test Data 97.51 100 98.74
Table 2.0 Naïve Bayes
Table Data Precision Recall F-score
Apparel 82.79 99.57 90.45

Graphical Result and Comparison Chart
When precision, recall, and F-measure are applied to aspect term occurrences, TP is the number of aspect term occurrences
tagged (each term occurrence) both by the method being evaluated, FP is the number ofaspecttermoccurrencestaggedbythe
method and FN is the number of aspect term occurrences tagged by themethod. Thethreemeasuresarethendefinedasabove.
They now assign more importance to frequently occurring distinctaspectterms, buttheycanproducemisleadinglyhighscores
when only a few, but very frequent distinct aspect terms are handled correctly. Furthermore,theoccurrence-based definitions
do not take into account that missing several aspect term occurrences or wrongly tagging expressions as aspect term
occurrences may not actually matter, as long as the most frequent distinct aspect terms can be correctly reported.
When precision, recall, and F-measure are computed over aspect term occurrences, all three scores appear to be very high,
mostly because the system performs well with the occurrences of ‘design’, which is a very frequentaspectterm,eventhoughit
performs poorly with all the other aspect terms. This also does not seem right. In the case of distinct aspect terms, precision
and recall do not consider term frequencies at all, and in thecaseofaspecttermoccurrencesthetwomeasuresaretoosensitive
to high-frequency terms.
V Conclusion
Proposed framework set up approach for opinion features extraction which used SPAM AND NON SPAM FEATURES include
sifting standard. SPAM AND NON SPAM FEATURES features sifting measureutilizesthedifferencesindistributional qualitiesof
highlights across two corpora, out of which one is domain-explicit and another is domain-autonomous.SPAMANDNON SPAM
FEATURES perceive up-and-comer highlights as domain-explicit and domain-free. Proposed IDER technique prompts
detectable improvement in both element extractionexecutionand featuresbasedopinionmining resultswhencontrastedwith
existing IDR, NON SPAM FEATURES LDA, ARM, MRC, and DP. We figure the effect of corpus size and subject choice on include
extraction execution. To figure this, great nature of domain autonomous corpus is fundamental. We see that using a domain-
autonomous corpus with comparable size as yet topically unique in relation to the given survey domain will create
predominant opinion features extraction results.
REFERENCES
1) ZHIANG WU et al.: Detecting Spammer Groups From Product Reviews: A Partially Supervised Learning Model
10.1109/ACCESS.2018.2820025
2) B. Liu, “Sentiment Analysis and Opinion Mining,” SynthesisLecturesonHumanLanguageTechnologies,vol.5,no.1, pp.
1-167, May 2012.
3) Y. Jo and A.H. Oh, “Aspect and Sentiment Unification Model for Online Review Analysis,” Proc. Fourth ACM Int’l Conf.
Web Search and Data Mining, pp. 815-824, 2011.
4) N. Jakob and I. Gurevych, “Extracting Opinion Targets in a Singleand Cross-Domain Setting with Conditional Random
Fields,” Proc. Conf. Empirical Methods in Natural Language Processing, pp. 1035- 1045, 2010.
5) F. Li, C. Han, M. Huang, X. Zhu, Y.-J. Xia, S. Zhang, and H. Yu,“Structure-AwareReviewMiningandSummarization,”Proc.
23rd Int’l Conf. Computational Linguistics, pp. 653-661, 2010.
6) C. Zhang, D. Zeng, J. Li, F.-Y. Wang, and W. Zuo, “Sentiment AnalysisofChineseDocuments:FromSentencetoDocument
Level,” J. Am. Soc. Information Science and Technology, vol. 60, no. 12, pp. 2474-2487, Dec. 2009.
7) G. Qiu, C. Wang, J. Bu, K. Liu, and C. Chen, “Incorporate the Syntactic Knowledge in Opinion Mining in User-Generated
Content,” Proc. WWW 2008 Workshop NLP Challenges in the Information Explosion Era, 2008.
8) Q. Su, X. Xu, H. Guo, Z. Guo, X. Wu, X. Zhang, B. Swen, and Z. Su, “Hidden Sentiment Association in Chinese Web Opinion
Mining,” Proc. 17th Int’l Conf. World Wide Web, pp. 959-968, 2008.

9) R. Mcdonald, K. Hannan, T. Neylon, M. Wells, and J. Reynar, “StructuredModelsforFine-to-CoarseSentimentAnalysis,”
Proc. 45th Ann. Meeting of the Assoc. of Computational Linguistics, pp. 432- 439, 2007.
10) S.-M. Kim and E. Hovy, “Extracting Opinions, Opinion Holders, andTopicsExpressedin OnlineNewsMedia Text,”Proc.
ACL/COLING Workshop Sentiment and Subjectivity in Text, 2006.
11) B. Pang and L. Lee, “A Sentimental Education: Sentiment Analysis Using Subjectivity Summarization Based on
Minimum Cuts,” Proc. 42nd Ann. Meeting on Assoc. for Computational Linguistics, 2004.
12) M. Hu and B. Liu, “Mining and Summarizing Customer Re- views,” Proc. 10th ACM SIGKDD Int’l Conf. Knowledge
Discovery and Data Mining, pp. 168-177, 2004.
13) D.M. Blei, A.Y. Ng, and M.I. Jordan, “Latent Dirichlet Allocation,” J. Machine Learning Research, vol. 3, pp. 993-1022,
Mar. 2003.
14) B. Pang, L. Lee, and S. Vaithyanathan, “Thumbs up?: Sentiment Classification Using Machine Learning Techniques,”
Proc. Conf. Empirical Methods in Natural Language Processing, pp. 79-86, 2002.
15) P.D. Turney, “Thumbs Up or Thumbs Down?:SemanticOrientationAppliedtoUnsupervisedClassification ofReviews,”
Proc. 40th Ann. Meeting on Assoc. for Computational Linguistics, pp. 417- 424, 2002.
16) V. Hatzivassiloglou and J.M. Wiebe, “Effects of Adjective Orientation and Gradability on Sentence Subjectivity,” Proc.
18th Conf. Computational Linguistics, pp. 299-305, 2000.
17) J. Kamps, `Ùsing wordnet to measure semantic orientation of adjectives,'' in Proc. Int. Conf. Lang. Resour. Eval.,2004,
pp. 1115_1118. [Online]. Available: http://www.baidu.com
18) D. Lin, ``WordNet: An electronic lexical database,'' Comput. Linguistics, vol. 25, no. 2, pp. 292_296, 1999.
19) M. Wang and H. Shi, ``Research on sentiment analysis technology and polarity computation of sentiment words,'' in
Proc. IEEE Int. Conf. Prog. Informat. Comput. (PIC), vol. 1. Dec. 2010, pp. 331_334.
20) E. Cambria, `Àffective computing and sentiment analysis,'' IEEE Intell. Syst., vol. 31, no. 2, pp. 102_107, Mar./Apr.
2016.
21) F. Zhou, R. J. Jiao, and J. S. Linsey, ``Latent customer needs elicitation by use case analogical reasoning from sentiment
analysis of online product reviews,'' J. Mech. Des., vol. 137, no. 7, p. 071401,2015.
22) R. Li, S. Shi, H. Huang, C. Su, and T. Wang, `À method of polarity computation of chinese sentiment words based on
Gaussian distribution,'' in Proc. 15th Comput. Linguistics Intell. Text Process. (CICLing), vol. 8404.

Identifying Malicious Reviews Using NLP and Bayesian Technique on Ecommerce Historical Data

More Related Content

Similar to Identifying Malicious Reviews Using NLP and Bayesian Technique on Ecommerce Historical Data (20)

More from IRJET Journal (20)

Recently uploaded (20)

Identifying Malicious Reviews Using NLP and Bayesian Technique on Ecommerce Historical Data