AISQA - 人工免疫问答系统(IJMECS-V4-N3-4)

I.J.ModernEducationandComputerScience,2012,3,28-34PublishedOnlineApril2012inMECS()DOI:10.5815/ijmecs.2012.03.04Copyright©2012MECSI.J.ModernEducationandComputerScience,2012,3,28-34AISQA-AnArtificialImmuneQuestionAnsweringSystemMohsenShakibaFakhrDepartmentofComputerEngineering,dezfulbranch,IslamicAzadUniversity,Dezful,IranEmail:Shakibafakhr@yahoo.comMohammadSanieeAbadehElectricalandComputerEngineeringCollege,TarbiatModaresUniversity,Tehran,IranEmail:Saniee@modares.ac.irAbstract—Questionanswering(QA)isthetaskofautomaticallyansweringaquestionposedinnaturallanguage.Atthistime,thereexistsseveralQAapproaches,and,accordingtorecentevaluationresults,mostofthemarecomplementary.Someofthemusetheevolutionaryalgorithms,suchasthegeneticalgorithm,initself.Inthispaperweproposeaquestionansweringsystemthatusestheartificialimmunealgorithms,forsearchingintheknowledgebasetofindtherightanswer.Thisalgorithmisoneoftheevolutionaryalgorithms.Searchisbasedontwofeatures:(i)thecompatibilitybetweenquestionandanswertypes,(ii)theoverlapandnon-overlapinformationbetweenthequestion-answerpair.Experimentalresultsareencouraging;theyindicatesignificantincreasesintheaccuracyofproposedsystem,incomparisonwiththeprevioussystems.IndexTerms—QA,GA,ArtificialImmuneSystem,Mutation.I.INTRODUCTIONInformationRetrieval(IR)Systems,receiveseveralkeywordsfromtheuser,andthesearchengineretrievesalltherelateddocumentsfromitsdocumentrepositoryinalimitedtime.Mostofretrieveddocumentsarejustsyntactically–andnotsemantically-relatedtotheuserquery.Theseenginesreceivetheusersquerythatconsistofseveralkeywords,andinsteadofgivingexactanswerstotheusersquestionjustretrievethedocumentsthatarerelevanttousersquery[1].Thesesystemshavesomemajorproblems.First,thoseusershavequestion,butinsteadthequestionsomekeywordsshouldbeentered.Ontheotherhand,usuallyusershaveaproblemtoconvertthequestiontotheappropriatekeywords,andthisconversionrequiresskill,thatmustbeachievedovertime.Inaddition,severalkeywordscannotmaketheuserintention,thatthisissue,sometimesimpossibletomakethisconversion.Sowecansaythattheuseofkeywords,notproperandthoroughmethodforcommunicationbetweenthesystemanduser.Ontheotherhand,usually,usersarelookingforexactanswers,whiletheoutputofthesesystemsisagreatdocumentthatmaynothavethecorrectanswerinitself.Thus,userisforcedtoreadalargenumberofdocuments,tofindtheirdesiredanswers.Usersneedexactandaccurateinformationanddon'tliketowastetheirtimebyreadingallretrieveddocumentstofindtheanswer,andIRsystemsarenotsufficientforthisreason[2].So,anewkindofIR,namedQuestionAnswering(QA)systemsappearedfromthelate1970'sandearly1980's.Inthesesystems,theuseraskhis/hernaturallanguagequestionwithnorestrictioninitssyntaxorsemantic.Thesystemisresponsibleforfindingtheexact,short,andcompleteanswerattheshortestpossibletime.Todothis,aQAAISQA-AnArtificialImmuneQuestionAnsweringSystem29Copyright©2012MECSI.J.ModernEducationandComputerScience,2012,3,28-34systemappliesbothIRandNLPtechniques[3].Inasubdivision,questionansweringsystemsaredividedintwocategories[4]:Restricteddomain.Respondstoquestionsonaparticulardomain(egmedicalorcarmaintenance),andcanbeuseofspecificknowledgeofitsdomain,tonaturallanguageprocessing.Opendomain.Thatalmostdealingwithanyquestion,andcanrelyontheglobalontologyandpublicknowledge.AnotherdivisionforQAsystemsisbasedonthenumberoflanguagesacceptedbythesesystems.Monolingualsystems,receivethequestion,andrespondtoit,onlywithonelanguage.Anothergroupthatiscalledmultilingualsystems,haveabilityforunderstandandrespondtoquestionsthatincludeseveraldifferentlanguages[5].Thispaperwillpresentbrieflyreviewonthequestionansweringsystem,andrelatedworkinthisarea.Thethirdsectionisdedicatedtotheproposedwork,inthefourthsection,theproposedsystemperformanceisevaluatedandcomparedwithgeneticalgorithm,andthefifthsectionofpaperisconcluded.II.RELATEDWORKSQAsystemswhicharebasedonsearchingamongasetofdocumentsareusuallycomposedofthreemainmodules[6]:(1)questionanalysisandextension(2)documentretrieval(3)answerextraction.Thefirstmoduleanalysestheuserquestiontoextractthetypeofquestionandtheexpectedtypeofanswer[7]orextendsittobeusedbythenextmodules.Thesecondmodulerelatestoretrievingrelevantdocumentstotheuserquery.Itcanbereplacedbyasearchengine.Thethirdmoduleextractsthefinalanswerfromthedocumentsretrievedbythesecondmodule.Allquestionansweringsystems,havethreestepsabove,butdifferentmethodsareusedtoimplementthisprocess.Thefirstdevicesforaccesstotheinformation,weretextualinformationretrievalsystems,thatdespitethesimple,areuseful,andverywidelyused.Anexampleofthissystems,areGoogle,AltavistaandMSNSearch,whereusedtofindrelevantdocumentsontheInternet.Someofinformationretrievalsystems,aredesignedforuseintextualcollection,outoftheInternet,suchastheSMART[8]andPRISE[9].WebQuestionAnsweringSystem,isanotherexampleofquestionansweringsystems,whichusedthegeneticalgorithmforranking.Inthissystem,atfirst,wordswillbesenttotheWeb,andsentencesthatincludetheanswer,areretrieved.Setofretrievedsentences,arematchedwithknownpreviousparagr