編輯反饋的內(nèi)容

1、編輯反饋的內(nèi)容編輯反饋的內(nèi)容AnatomicalentityrecognitionwithahierarchicalframeworkaugmentedbyexternalresourcesPLOSONEThankyouforsubmittingyourmanuscripttoPLOSONE.Aftercarefulconsideration,wefeelthatithasmerit,butisnotsuitableforpublicationasitcurrentlystands.Therefore,mydecisionisMajorRevision.Weinviteyoutosubmita

2、revisedversionofthemanuscriptthataddressesalloftheconcernsraisedbythetworeviewers.Itiscriticalthatyouspecificallyaddressthefollowingissues:1)Providemoredetailsonyourmethodologyanddatasources(possiblywithexamples),sothatthereviewerscanbetterevaluatethesummaryresultsprovidedinthetables;2)Describepreci

3、selywhatwillbepubliclyavailable;3)Thoroughlyedityourrevisedmanuscriptbeforesubmission.PleasenotethatPLoSONEdoesnotprovidecopyediting.Weencourageyoutosubmityourrevisionwithinforty-fivedaysofthedateofthisdecision.Whenyourfilesareready,pleasesubmityourrevisionbyloggingontoandfollowingtheSubmissionsNeed

4、ingRevisionlink.Donotsubmitarevisedmanuscriptasanewsubmission.Beforeuploading,youshouldproofreadyourmanuscriptverycloselyformistakesandgrammaticalerrors.Shouldyourmanuscriptbeacceptedforpublication,youmaynothaveanotherchancetomakecorrectionsaswedonotofferpre-publicationproofs.Ifyouwouldliketomakecha

5、ngestoyourfinancialdisclosure,pleaseincludeyourupdatedstatementinyourcoverletter.Inaddition,whensubmittingyourrevisionpleaseincludethefollowingitems:Arebuttalletterthatrespondstoeachpointbroughtupbytheacademiceditorandreviewer(s).ThislettershouldbeuploadedasaResponsetoReviewersfile.?Acleanrevisedman

6、uscriptasyourManuscriptfile.?Amarked-upcopyofthechangesmadefromthepreviousarticlefileasaRevisedManuscriptwithTrackChangesfile.ThiscanbedoneusingtrackchangesinprogramssuchasMSWordand/orhighlightinganychangesinthenewdocument.?Formoreinformationonhowtouploadyourrevisedsubmission,seeourvideo: HYPERLINK

7、/everyone/2021/05/10/how-to-submit-your-revised-manuscript/ /everyone/2021/05/10/how-to-submit-your-revised-manuscript/Ifyouchoosenottosubmitarevision,pleasenotifyus.Yourssincerely,RaminHomayouni,Ph.D.AcademicEditorPLOSONEJournalrequirements:Whensubmittingyourrevision,weneedyoutoaddresstheseaddition

8、alrequirements:Wenotethatyouhavestatedthatyouwillproviderepositoryinformationforyourdataatacceptance.Shouldyourmanuscriptbeacceptedforpublication,wewillholdyourmanuscriptuntilyougetintouchwithuswiththeaccessionnumbersorDOIsnecessarytoaccessyourdata.Ifyouwishtomakechangestoyourdataavailabilitystateme

9、nt,pleasedescribethesechangesinyourcoverletterandwewillmakethemonyourbehalf.Reviewerscomments:ReviewersResponsestoQuestionsCommentstotheAuthorIsthemanuscripttechnicallysound,anddothedatasupporttheconclusions?Themanuscriptmustdescribeatechnicallysoundpieceofscientificresearchwithdatathatsupportstheco

10、nclusions.Experimentsmusthavebeenconductedrigorously,withappropriatecontrols,replication,andsamplesizes.Theconclusionsmustbedrawnappropriatelybasedonthedatapresented.Reviewer#1:PartlyReviewer#2:YesHasthestatisticalanalysisbeenperformedappropriatelyandrigorously?Reviewer#1:NoReviewer#2:YesDoesthemanu

11、scriptadheretothePLOSDataPolicy?Authorsmustfollowthe,whichrequiresauthorstomakealldataunderlyingthefindingsdescribedintheirmanuscriptfullyavailablewithoutrestriction.PleaserefertotheauthorsDataAvailabilityStatementinthemanuscript.Alldataandrelatedmetadatamustbedepositedinanappropriatepublicrepositor

12、y,unlessalreadyprovidedaspartofthesubmittedarticleorsupportinginformation.Iftherearerestrictionsontheabilityofauthorstopubliclysharedatae.g.privacyoruseofdatafromathirdpartythesereasonsmustbespecified.Reviewer#1:YesReviewer#2:NoIsthemanuscriptpresentedinanintelligiblefashionandwritteninstandardEngli

13、sh?PLOSONEdoesnotcopyeditacceptedmanuscripts,sothelanguageinsubmittedarticlesmustbeclear,correct,andunambiguous.Anytypographicalorgrammaticalerrorsshouldbecorrectedatrevision,sopleasenoteanyspecificerrorshere.Reviewer#1:NoReviewer#2:YesReviewCommentstotheAuthorPleaseusethespaceprovidedtoexplainyoura

14、nswerstothequestionsabove.Youmayalsoincludeadditionalcommentsfortheauthor,includingconcernsaboutdualpublication,researchethics,orpublicationethics.(Pleaseuploadyourreviewasanattachmentifitexceeds20,000characters)Reviewer#1:Thispaperpresentsaninterestinghierarchicalframeworktorecognizeanatomicalentit

15、ies,whichisimportantinhealthcaredomain.Authorsalsobringtheimportanceandthechallengesofthistask.Tothebestofmyknowledge,Isummarizemycommentsandsuggestionsasfollows:FeaturesforthesequencelabelingproblemsunderCRFarecomprehensiveandacceptable.Authorsincludebaselinenaturallanguagefeatures,semanticfeatures

16、fromexternalknowledgeaboutWikipediaandWordNet,co-reference,anddictionarymatching.Authorsconductedrelativelycomprehensiveexperimentstoshowthecontributionofeachindividualfeaturesandcombinationoffeaturestotheoverallprecisionandrecall.Problemintroductionandannotationaregoodtoo.However,somemajorpointsnee

17、dtobefixed:1)Thewritingofthispaperisreallypoor.Alltablereferencesarenotcorrect,grammarerrorscanbeseenalmosteveryparagraph.Itisveryverydifficulttoread.Ittookmehundredsofhourstounderstandwhatauthorstrytodeliver.Letmejustshowexamplesbasedontheabstract:a)Thefirstsentenceisnotacompletesentence.Todevelopi

18、nmedicalrecords.Therecordbutalsobyotherdiverse.=Theyinferrelevantanatomicalentitiesbasedonbothexplicitanatomicalexpressionsintherecordandotherdiverse.Thehierarchicalframeworkwasdemonstrated.=Thehierarchicalframeworkwasdemonstrated.inF1comparingto?manyothersinthepaper!Fortheannotation,authorsusedA3to

19、check(A1,A2),thenobtainthecoefficient.WhynotA3-(A1,A2),A1-(A2,A3),andA2-(A1,A3),thenobtaintheaveragecoefficient?Whatifthereisaannotationconflict,meaningthatall3annotatorsdonotagree?Inaddition,authorsclaimthattheirgoldenstandardisnotperfect,thenwhyyoustillusethemtodoevaluations?Fromtheexperimentalres

20、ults,CFseemstobethesmallestcontributiontotheprecisionintable5andtable8,thenwhyaddingCFgetsalotincreaseintable6and9?Idontbelievethisresult.Canyougivesomeexplanations.Inaddition,somesuggestions,Itwouldbegreatifthepapergivessomeformaldefinitionofeachconceptandshowssomerealortoyexamplesinfigure.Theycanh

21、elpreaderstocatchthepoint.Reviewer#2:ThemanuscriptbyYanXuetal.describestheconstructionofananatomicalentityrecognitionframeworkbasedonamachinelearningalgorithm.Thisframeworkcanrecognizenotonlyexplicitexpressionsofanatomicalentities,butalsoimplicitexpressionssuchasdiseases,clinicaltreatments,andclinic

22、altests.Theauthorsinsistedthattherecognitionoftheimplicitexpressionswasimportantbecausetheimplicitexpressionsareabundantinclinicalrecordsanditisfromtheseimplicitexpressionsthatmedicalexpertscaninfertheanatomicalentitiesdescribedinthedocuments.Theframeworkconsistsofthreelayersofentityrecognizers,allo

23、fwhicharebasedonconditionalrandomfield(CRF)models.Thefirstlayeristhemulti-classCRFrecognizerdevelopedforthe2021and2021I2B2challenge;thislayerrecognizesentitiesofthreesemanticclasses:diseases,clinicaltreatments,andclinicaltests.Theothertworecognizerlayersaredevelopedinthisstudy.One(thesecondlayer)isf

24、orexplicitanatomicalexpressionandtheother(thethirdlayer)isforimplicitexpression.ForuseinthetrainingandtestingoftheCRFmodels,theauthorscarefullymadeanannotatedcorpusof300clinicalrecords(i.e.,thedischargesummariesinthisstudy).Theresultingannotationsinclude16690explicitanatomicalentitytokensand5564impl

25、icitanatomicalentitytokens.TheauthorsusedthefollowingfeaturesfortheconstructionoftheCRFmodelsandconsideredtherelativeimpactontherecognitionperformanceusingprecision,recall,andF-score:baselinefeatures(astandardsetofusefulfeaturesforgeneralnamedentityrecognitiontasks),ontologicalfeaturesDF1andDF2(base

26、donsomeoftherepresentativeanatomicalontologies:UMLS,MeSH,RadLex,andBodyParts3D),coreferencefeatures,andworldknowledgefeaturesWF1,WF2,WF3,andHF,whichisbasedonthedictionaryconstructedfromthetermsinWikipediaandWordNet,whosedefinitionsentencescontainexplicitanatomicalentities,forthepurposeofextractingim

27、plicitanatomicalentities;HFisreferredtoasahierarchicalfeature.Thisstudyisoriginalandaddressesanimportanttaskinprocessingmedicaldocumentsingeneral.Theiranalyticalapproachseemstobesoundinthesenseofordinalresearchonnaturallanguageprocessing.Therefore,thismanuscriptseemstowarrantpublicationinPLOSONE.The

28、maincriticismIhaveisthelackofconsiderationofconcreteinstancesofanatomicaldictionaries,clinicalrecordcorpuses,annotations,andexperimentresults.Theauthorsonlyprovidedseveralnumericaltablesoftheprecision,recall,andF-score.Allthemainconclusionsweredrawnfromobservationofthesenumericaltables.AlthoughIknow

29、thatthisstyleiscommoninNLPresearchpapers,Ibelievethatwithoutaninvestigationofconcreteinstances,readerscannotevaluatetherelativeimpactofthemanyfactorsthatwillaffectthefinalperformance.Withonlyalittlethought,onecanlistupmanyfactorsthataffectthefinalresults:datasourcesselectionfortheconstructionofthean

30、atomicaldictionaries,relativecontributionofthe(four)datasourcesontheperformance,whetherthereexistssomeparticularanatomicalterminthefourdictionariesthathasasignificanteffectontheperformance,thetotalsizeofanatomicaldictionaries,semantictypeoftermsincludedintheanatomicaldictionaries,typeofclinicalrecor

31、ds,totalnumberofclinicalrecordsandsentenceswhichareannotatedbytheexperts,targetsemantictypes,thechoicesofmachinelearningalgorithms,andtheselectionofthefeaturesfortheCRFmodels,aswellasmanyotherfactors.However,observationoftheseriesofnumericaltablesyieldsonlylimitedinformationabouttheimpactofthefactor

32、sandwhatentitiescan/cannotberecognizedundertheproposedframework.Therefore,atveryleast,theauthorsshouldprovideapartofthelistof16690explicitanatomicalentitytokens|and5564implicitanatomicalentitytokensIIwiththeirnumbersofoccurrencesinthecorpus,becausethesedefinetheproblemthatthismanuscriptisaddressing.

33、Inaddition,theauthorsshoulddiscusswhattermsintheanatomicaldictionarymatchtheannotatedtokensand/ortheresultsoftheBegin/Inside/Outside(BIO)callingbytheCRFmodel.Thensomeexplanationoftherelativeimpactoftheframeworkcomponentsshouldbeprovidedbasedontheconcreteinstancesofmatchingresults.Asecondcriticismcon

34、cernsthereproducibilityofthisstudy.Althoughtheauthorswroteattheendoftheabstractsection,Theresourcesconstructedforthisresearchwillbemadepubliclyavailable.|sincetheresourcesneededforthereproductionofthisstudyarenotprovidedatthistime,Icouldnotevaluatewhethertheresultscanbereproducedusingtheresourcestha

35、ttheauthorssaywillbeeventuallyprovided.IknowthattheauthorshavemadeagreatcontributiontotheNLPresearchfield,notonlybyintroducingnovelconcepts,butalsobyprovidingmanyusefulresources,includingsoftwareandannotatedcorpuses,andsoIbelievethattheresourcesthatwillbeavailabletothepublicwillbequiteusefulforNLPre

36、searchers,butIbelievethatitisquiteimportanttomeetthereproducibilitycriteriastatedinthepublicationcriteriaofPLOSONE(describedinsufficientdetailforanotherresearchertoreproducetheexperimentsdescribedII),andinordertomeetthesecriteria,Iexpectthattheauthorswillneedtowriteadditionalparagraphsdescribinginsu

37、fficientdetailhowtoreproducetheresulttables.Ibelievethattheresultshavebeenlargelyaffectedbythecontentofthedictionariesandannotatedcorpusesconstructedbytheauthors,andtherefore,withouttheseresources,itwillbequitedifficultforotherresearcherstoreproduceexactlytheresultsdescribedinthetables.MinorpointsPage8,lines7-10Idonotunder