copy谷歌sb

RuleBasedPlagiarismDetectionusingInformationRetrievalAniruddhaGhosh,PinakiBhaskar,SantanuPal,SivajiBandyopadhyayDepartmentofComputerScienceandEngineering,JadavpurUniversity,Kolkata–700032,India{arghyaonline,pinaki.
bhaskar,santanu.
pal.
ju}@gmail.
com,sivaji_cse_ju@yahoo.
comAbstract.
ThispaperreportsaboutthedevelopmentofaPlagiarismdetectionsystemasapartofthePlagiarismdetectiontaskinPAN2011.
TheexternalplagiarismdetectionproblemhasbeensolvedwiththehelpofNutch,anopensourceInformationRetrieval(IR)system.
Thesystemcontainsthreephases–knowledgepreparation,candidateretrievalandplagiarismdetection.
Fromthesourcedocuments,knowledgebasehasbeenpreparedfordevelopingtheNutchindexandthequerieshavebeenformedfromthesuspiciousdocumentsforsubmissiontotheNutchIRsystem.
TheretrievedcandidatesourcesentencesareassignedsimilarityscoresbyNutch.
Dissimilarityscoreisassignedforeachcandidatesentenceandthesuspicioussentence.
Eachcandidatesourcesentenceisrankedbasedonthesetwoscores.
Thetoprankedcandidatesentenceisselectedforeachsuspicioussentence.
Keywords:PlagiarismDetection,InformationRetrievalSystem,SimilarityScore,DissimilarityScore.
1IntroductionPlagiarismmaybedefinedasthewrongfulmisuseandclosereplicationofthoughts,ideas,orexpressionsfromtheoriginalworkofsomeoneinthesamelanguageoffromanotherlanguage.
From18thcentury,plagiarismhasbeenconsideredasacademicdishonesty[1].
Fordecades,researchershaveexploreddifferenttechniquestodetectplagiarism.
Plagiarismcanoccurindifferentforms–fullplagiarism,substantialplagiarism,minimalisticplagiarism,sourcecitationetc.
IthasbecomeachallengingtaskintheareaofNaturalLanguageProcessing.
Inourapproach,wehaveconsideredalltheformsofplagiarismexceptminimalisticplagiarismatthesentencelevel.
Duetoabsenceofcontrolledevaluationenvironmenttocompareresultsofthealgorithms,plagiarismdetectionisstillachallengingtask[2].
Researchershaveorganizedvariousconferences(similartoPAN)toovercometheplagiarismproblem.
Fingerprintretrievalmethod[3],candidateretrieval[4]andpassageretrieval[5]arethemostprominentattemptsonplagiarismdetection.
Thesystemdescribedin[6]workswithanaturallanguageparsertofindswappedwordsandphrasestodetectintentionalplagiarismwhilen-gramco-occurrencestatisticisusedtodetectverbatimcopy.
TheLongestCommonSubsequencetechniquehasbeenusedin[7]tohandletextmodification.
Researchershaveusedcosinesimilarityscoreandn-gramvectorspacemodelatdifferentlevels,i.
e.
,word[8]andcharacter[9]levels.
Inthepresentwork,plagiarismhasbeentreatedasanIRproblem.
Anopensourcesearchengine,Nutch,hasbeenusedtoretrievetheplagiarizedpartsfromthesuspiciousdocuments.
2SystemFrameworkTheInformationRetrieval(Nutch1)basedPlagiarismDetectionsystemframeworkisshowninthefigure1.
Thesystemisdefinedinthreephases:KnowledgePreparation,CandidateRetrieval,i.
e.
,identificationofsuspicioussentenceandtheprobablesetofsourcesentencepairsandfinallyplagiarismdetectionofeachidentifiedsuspicioussentence.
Fig.
1.
SystemArchitecture3KnowledgePreparationEachsourcedocumentisparsedtoidentifyandextractallthesentencesinthedocument.
NowKnowledgefilesaregeneratedforeachsourcesentence.
Thefilenamesofknowledgefilesarecreatedinsuchamannerthatthesourcesentenceintheoriginalsourcedocumentcanbetracked.
Theknowledgeofeachsentenceintheknowledgefileisstoredintheformofstems,synonyms,hyponyms,hypernymsandsynsetsofeachword(afterremovalofthestopwords)thatareextractedfromWordNet3.
02.
Duplicatewordsareremovedtogetthesetofidenticalsenseuniquewords.
Thesewordsareusedtoidentifytheplagiarizedwords,thewordsthataresimilarinsensetotheoriginalwords.
Theoriginalwordsinthesentenceareaddedtothissetofwords.
Thus,eachknowledgefileforasentenceconsistsofasetofwords.
Afteralltheknowledgefilesarebuilt,theseareindexedusingLucene3.
1http://nutch.
apache.
org/2http://wordnet.
princeton.
edu/3http://lucene.
apache.
org/4CandidatesRetrievalEachsuspiciousdocumentisparsedtoidentifyandextractallthesentencesinthesuspiciousdocuments.
EachSuspicioussentenceisconsideredfromtheparsedsuspiciousdocumenttogeneratethequery.
FirstallthestopwordsareremovedfromthesentenceandthentheremainingwordsarebeingstemmedusingWordNet3.
0stemmertogettherootformofeachword.
Aftergeneratingthequeryfromthesuspicioussentences,thequeryisfiredtoNutchtoretrievetheprobablesetofsourcesentencescorrespondingtoeachsuspicioussentence.
Assourcedocumentsaresplitintosentencesintofilesandeachfilecontainsonlyonesentence,Nutchperformsasentence-sentencemappingforaproximalmatchbetweenthequeryandindexedsourcefiles.
AsetofprobablecandidatesourcesentencesisidentifiedbyNutchinrankedorderforeachsuspicioussentence.
Nutchprovidesthesimilarityscorebetweenasuspicioussentenceandthecorrespondingcandidatesourcesentence.
5PlagiarismDetectionAnalgorithmfordissimilaritymeasurement,proposedin[10],hasbeenusedtocalculatethedissimilarityscorebetweenthesuspicioussentenceanditscorrespondingretrievedcandidatesentences.
Foridenticalsentencesthathavemostnumberofidenticaln-grams,thedissimilarityscoreis0.
Usingthismeasurewehavecalculatedthedissimilarityscoresofeachsourcesentencecorrespondingtothesuspicioussentences.
Thedissimilarityscorearesubtractedfromthesimilarityscoreforeachcandidatesourcesentenceandafinalfine-grainedscorehasbeengenerated.
Alltheretrievedcandidatesourcesentencesforeachsuspicioussentencearerankedaccordingtothisfine-grainedscore.
Thetoprankedcandidatesourcesentenceisidentifiedasthesourcesentencefortheplagiarizedsentenceinthesuspiciousdocument.
6EvaluationTheplagiarismdetectionsystemwasevaluatedusingtheevaluationframeworkdescribedin[2].
TheevaluationscoresareshowninTable1.
Table1.
EvaluationMeasurementPrecisionRecallGranularityPladgetScore0.
00118290.
00500522.
00288180.
00120637ConclusionandFutureWorksThepresenttaskisourfirstattemptinplagiarismdetection.
Wehavetestedtheplagiarismatthesentencelevelbutphraselevelexperimentationisstillleftforinvestigate.
Infuture,analgorithmhastobedevelopedtotesttherelevanceofthecandidatesourcesentencesretrievedbyNutchandchoosethemostrelevantplagiarizedpart.
Theknowledgefilesforthesourcedocumentswillalsohavetobeupdated.
AcknowledgmentTheworkhasbeencarriedoutwithsupportfromDepartmentofInformationTechnology(DIT),Govt.
ofIndiafundedProjectDevelopmentof"CrossLingualInformationAccess(CLIA)"SystemPhaseII.
References1.
WikipediaarticleonPlagiarism:http://en.
wikipedia.
org/wiki/Plagiarism2.
PotthastM.
etal.
:AnEvaluationFrameworkforPlagiarismDetection.
InProceedingsoftheCOLING2010,Beijing,China,August2010.
3.
YuriiPalkovskii,AlexeiBelovandIrinaMuzika.
:ExploringFingerprintingasExternalPlagiarismDetectionMethod:LabReportforPANatCLEF2010.
InBraschleretal.
[2].
ISBN978-88-904810-0-0.
4.
VivianeP.
Moreira,RafaelC.
PereiraandGalanteRenata.
:UFRGS@PAN2010:DetectingExternalPlagiarism:LabReportforPanatCLEF2010.
InBraschleretal.
[2].
ISBN978-88-904810-0-0.
5.
ClaraVaniaandMirnaAdriani.
:ExternalPlagiarismDetectionUsingPassageSimilarities:LabReportforPANatCLEF2010.
InBraschleretal.
[2].
ISBN978-88-904810-0-0.
6.
M.
Mozgovoy,T.
KakkonenandE.
Sutinen.
:UsingNaturalLanguageParsersinPlagiarismDetection.
InProceedingofSLaTE'07Workshop,Pennsylvania,USA,October2007.
7.
Chen,Chien-Ying,Jen-YuanYehandHao-RenKe.
:PlagiarismDetectionusingROUGEandWordNet.
JournalofComputing,2(3),pages34-44,March2010.
https://sites.
google.
com/site/journalofcomputing/.
ISSN2151-9617.
8.
CristianGrozeaandMariusPopescu.
:Encoplot-PerformanceintheSecondInternationalPlagiarismDetectionChallenge:LabReportforPANatCLEF2010.
InBraschleretal.
[2].
ISBN978-88-904810-0-0.
9.
Basileetal.
:APlagiarismDetectionProcedureinThreeSteps:Selection,Matchesand"Squares".
InProceedingsoftheSEPLN2009WorkshoponUncoveringPlagiarism,AuthorshipandSocialSoftwareMisuse(PAN2009),Donostia-SanSebastian,Spain.
10.
VladoKeselj,FuchunPeng,NickCerconeandCalvinThomas.
:"N-gram-basedAuthorProfilesforAuthorshipAttribution".
InProceedingsofthePACLING'03,DalhousieUniversity,Halifax,NovaScotia,Canada,pp.
255-264,August2003.