資訊擷取與推薦技術(shù)期中報(bào)告

Ch9.ThesaurusConstruction指導(dǎo)教授：黃三益教授組別：第三組組員：周桂穗、孫繡紋、莊士民日期：19.Nov.20021outlineIntroduction索引典(Thesaurus)定義索引典結(jié)構(gòu)INSPECthesaurus索引典參照款目說明FeaturesofthesauriCoordinationlevelTermrelationshipsNumberofentriesforeachtermSpecificityofvocabularyControlontermfrequencyofclassmembersNormalizationofvocabularyThesaurusConstructionManualThesaurusConstructionAutomaticThesaurusConstruction3.2.1ThesaurusConstructionfromTextAutomaticThesaurusConstructionFromaCollectionofDocumentItemsByMergingExistingThesauriUserGeneratedThesaurusConstructionofVocabulary3.2.2MergingexistingthesauriConclusion21.1索引典(Thesaurus)定義就資訊儲(chǔ)存與檢索的範(fàn)疇而言，索引典乃收集足以表示知識(shí)概念的字或詞，並將之以特定的結(jié)構(gòu)加以排列，這些字彙控制了同義字，區(qū)別了同形異義字，並顯現(xiàn)各相關(guān)詞彙間階層及語意互屬上的各種關(guān)係，以作為索引者在分析處理資料及讀者在檢索資料時(shí)能選用一致的、經(jīng)過控制的詞彙。即提供資訊儲(chǔ)存與檢索標(biāo)準(zhǔn)化的用語。1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion31.3INSPECthesaurusThisthesaurusisdesignedfortheINSPECdomain,whichcoversphysics,electricalengineering,electronics,aswellascomputersandcontrol.Thethesaurusislogicallyorganizedasasetofhierarchies.Includesanalphabeticallistingofthesauralterms.Eachhierarchyisbuiltfromaroottermrepresentingahigh-levelconceptinthedomain.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion5Ashortextractfromthe1979INSPECthesaurusCesiumUSEcaesiumComputer-aidedinstructionseealsoeducationUFteachingmachinesBTeducationalcomputingTTcomputerapplicationsRTeducationteachingCCC7810CFCc7810cf1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion61.4索引典參照款目說明USE(被替代)The“seealso(參見)”linkleadstocross-referencedthesauralterms.NT(narrowerterms較狹義字)suggestamorespecificthesauralterm.BT(broaderterms較廣義字)providesamoregeneralthesauralterm.TT(Topterm,最BT的term)RTsignifiesarelatedterm(相關(guān)字).UF(替代)isutilizedtoindicatethechosenformfromasetofalternatives.CCClassificationCodes類別代碼1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion72.1CoordinationlevelTheconstructionofphrasesfromindividualterms.Twocoordinationoptions:pre-coordinationandpost-coordination.Aprecoordinatedthesauruscancontainphrases.Theadvantageisthatthevocabularyisveryprecise.Thedisadvantageisthatthesearcherhastobeawareofthephraseconstructionrulesemployed.Precoordinationismorecommoninmanuallyconstructedthesauri.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion9Apostcoordinatedthesaurusdoesnotallowphrases.Instead,phrasesareconstructedwhilesearching.Theadvantageisthattheuserneednotworryabouttheexactorderingofthewordsinphrase.Thedisadvantageisthatsearchprecisionmayfall.

Automaticthesaurusconstructionusuallyimpliespostcoordination.10Coordinationlevel組合層次前組合將此述語概念視為一個(gè)複合名詞，例如：diesellocomotive(柴油引擎火車頭)後組合以現(xiàn)存的二個(gè)或二個(gè)以上的詞彙代替而組合，例如：dieselenginesANDlocomotive1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion112.2a)EquivalencerelationshipsEquivalencerelationsincludebothsynonymy(同義字)andquasi-synonymy(半同義字).Forexample:genetics(遺傳)andheredity;harshnessandtenderness1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion13同義關(guān)係同義字：同一概念可以用一種以上的詞彙表示時(shí)，索引典多選用較廣為使用或新穎的一種為述語，其他則作為參照款目，例如：storagebatteriesUFsecondarybatteriessecondarybatteriesUSEstoragebatteriesUF(UsedFor)替代USE被替代半同義字：有時(shí)兩個(gè)反義字卻可代表一個(gè)概念一體之兩面，則擇其一為述語，另一為參照款目，例如：stabilityUFinstability，相對(duì)應(yīng)之參照款目為instabilityUSEstability1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion142.2b)Hierarchicalrelationships

層級(jí)關(guān)係Atypicalexampleofahierarchicalrelationisgenus(屬)-species(種),suchas”dog”and“germanshepherd(牧羊犬近親).”1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion152.2c)Nonhierarchicalrelationships

非層級(jí)關(guān)係Nonhierarchicalrelationshipsalsoidentifyconceptuallyrelatedterms.Therearemanyexamplesincluding:thing—partsuchas“bus”and“seat”;thing—attributesuchas“rose”and“fragrance(香味)”.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion17相關(guān)關(guān)係是指認(rèn)可述語間的關(guān)連，一般採用RT的參照符號(hào)來連結(jié)，例如：表示事或物的全部與部份的關(guān)係，windowsRThouses表示事或物與其處理作業(yè)的關(guān)係，skatesRTskating表示事或物與其應(yīng)用的關(guān)係，railwayconstructionRTrailway表示事或物與其特性的關(guān)係，seawaterRTcorrosion(侵蝕）1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion18Wang,Vandendorpe,andEvens(1985)provideanalternativeclassificationoftermrelationshipsconsistingof:(1)parts—wholes(整部關(guān)係)(2)collocationrelations(排列關(guān)係)(3)paradigmaticrelations(範(fàn)例關(guān)係)(4)taxonomyandsynonymy(分類及同義字)(5)antonymyrelations(反義字)19(2)Collocationrelations

編排關(guān)係Collectionrelateswordsthatfrequentlyco-occurinthesamephraseorsentence.21(3)ParadigmaticrelationsParadigmaticrelationsrelatewordsthathavethesamesemanticcorelike“moon”and“l(fā)unar”andaresomewhatsimilartoAitchisonandGilchrist’squasi-synonymyrelationship.22(4)TaxonomyandsynonymyTaxonomyandsynonymyareself-explanatoryandrefertotheclassicalrelationsbetweenterms.232.3NumberofentriesforeachtermItisingeneralpreferabletohaveasingleentryforeachthesaurusterm.However,thisisseldomachievedduetothepresenceofhomographs—wordswithmultiplemeanings.InamanuallyconstructedthesaurussuchasINSPEC,thisproblemisresolvedbytheuseofparentheticalqualifiers(括弧限定語),asinthepairofhomographs,bonds化學(xué)鍵(chemical)andbonds粘合劑(adhesive膠帶).However,thisishardtoachieveautomatically.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion25同形異義關(guān)係(homographs)當(dāng)許多詞彙的拼法完全相同，但所代表的意義卻不同時(shí)，則以小括號(hào)加修飾語以區(qū)別之，例如：Mercury水銀(metal金屬)、Mercury水星(planet行星)，小括號(hào)內(nèi)的修飾語亦為述語的一部分。1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion262.6Normalizationofvocabulary詞彙的標(biāo)準(zhǔn)化最好以名詞方式表示名詞片語應(yīng)避免採用頭字語，除非大家都知道可採用形容詞Thereareotherrulestodirectissuessuchasthesingularityofterms(單數(shù)),theorderingoftermswithinphrases(在片語中的順序),spelling(拼法),capitalization(大寫),transliteration(字譯),abbreviations(縮寫),initials(字首),acronyms(字首縮寫),andpunctuation(標(biāo)點(diǎn)符號(hào)).Theadvantageisthatvariantformsaremappedintobaseexpressions,therebybringingconsistencytothevocabulary.Thedisadvantageisthat,inordertobeusedeffectively,theuserhastobewellawareofthenormalizationrulesused.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion293.1ManualThesaurusConstructionDefinetheboundariesofthesubjectareaIdentifycentralsubjectareasandperipheralonesPartitionthedomainintodivisionsorsubareasIdentifydesiredcharacteristicsCollecttermsforeachsubareaSourcesfromindex,encyclopedia,handbook,textbook,journal,abstract,catalog,existingthesaurusorvocabularysystemsIncluding:subjectexpertandpotentialuser1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion303.1ManualThesaurusConstruction

(continued)AnalyzeeachtermforitsrelatedvocabularyIncludingsynonyms,broaderandnarrowerterm,definitionandscopenoteOrganizetermandrelationshipintohierarchicalstructureRevieworrefineforconsistencyInvertthestructuredthesaurustoproduceanalphabeticalarrangementofentriesTestthethesaurus1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion313.1ManualThesaurusConstruction

(continued)Conclusion:InvolveagroupofindividualsandavarietyofresourcesNeedtobemaintainedtoensureviabilityandeffectivenessReflectanychangesintheterminologyofthearea

=>Anartandascience1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion323.2AutomaticThesaurusConstructionFromdocumentcollectionsUseacollectionofdocumentsasthesourceforthesaurusconstructionApplystatisticalprocedurestoidentifyimportanttermsaswellasrelationshipsUsecomputationallysimplermethodstoidentifythemoreimportantsemanticknowledge1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion333.2AutomaticThesaurusConstruction(continued)MergeexistingthesaurusMergetwoormorethesauriintoasingleunitMergershouldnotviolatetheintegrityofanycomponentthesauruse.g.augmentMeSHfromSNOMED1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion343.2AutomaticThesaurusConstruction(continued)UsergeneratedthesaurusUsesoftermrelationshipinsearchstrategiesCaptureknowledgefromuser’ssearche.g.TEGEN(ThesaurusGeneratingsystem)ThetypesofBooleanoperatorsbetweentermsThetypeofquerymodificationUserfeedbackincluded1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion353.2.1ThesaurusConstructionfromTextsProcess1ConstructionofvocabularyNormalizationSelectionoftermsPhraseconstructionIdentifythestatisticalassociationsbetweenterms2Similaritycomputations3OrganizationofvocabularyOrganizetheselectedvocabularyintohierarchy1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion361ConstructionofVocabularyObjective:Identifythemostinformativeterms(words,phrases)IdentifyanappropriatedocumentcollectionwhichshouldbesizableandrepresentativeofthesubjectareaDeterminetherequiredspecificityVocabularyfornormalizationEliminatetrivialwordsandconstructastoplistStemthevocabulary1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus37SelectionbyFrequencyofOccurrenceSelectionbyfrequencyofoccurrenceEachtermplacedinonefrequencycategory:high,medium,lowMedium:bestforindexingandabstractingLow:minimalimpactonretrievalHigh:toogeneralandnegativelyimpactsearchprecision1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus38SelectionbyDiscriminationValue(DV)Selectionbydiscriminationvalue(DV)DVmeasuresthedegreetowhichatermisabletodiscriminateordistinguishbetweenthedocumentsThemorediscriminatingaterm,thehigheritsvalueasanindextermUsingsomesimilarityfunctionstocomputetheaverageinter-documentsimilarityinthecollectionDV(k)=(Averagesimilaritywithoutk)-(Averagesimilaritywithk)1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus39SelectionbyDiscriminationValue(DV)

(continued)Selectionbydiscriminationvalue(DV)GooddiscriminatorsarethosethatdecreasetheaveragesimilaritybytheirpresenceDVispositivePoordiscriminatorshavenegativeDVNeutraldiscriminatorshavenoeffectonaveragesimilarityTermsthatarepositivediscriminatorscanbeincludedinthevocabularyandtherestrejected1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus40SelectionbyPoissonMethodSelectionbyPoissonMethodPoissondistributionisadiscreterandomdistributionthatcanbeusedtomodelavarietyofrandomphenomenaTrivialwordshaveasinglePoissondistributionDistributionofnontrivialwordsdeviatessignificantlyfromaPoissondistribution1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus41PhraseConstructionPhraseconstruction

decreasethefrequencyofhigh-frequencytermsandincreasetheirvalueforretrievalSaltonandMcGillProcedure:astatisticalalternativetosyntacticand/orsemanticmethodsforidentifyingandconstructingphrasesThecomponentwordsofaphraseshouldoccurfrequentlyinacommoncontextThecomponentwordsshouldrepresentbroadconcepts,andtheirfrequencyofoccurrenceshouldbesufficientlyhigh1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus42PhraseConstruction

(continued)CriteriaComputepairwiseco-occurrenceforhigh-frequencywordsIfthisco-occurrenceislowerthanathreshold,thendonotconsiderthepairanyfurtherForpairsthatqualify,computethecohesionvalue

cohesion(ti,tj)=

co-occurrence-frequency/sqrt(frequency(ti)*(frequency(tj))

cohesion(ti,tj)=

size-factor*(co-occurrence-frequency/(total-frequency(ti)*(total-frequency(tj)))Ifcohesionisaboveasecondthreshold,retainthephraseasavalidvocabularyphrase1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus43ChouekaProcedure(continued)ChouekaProcedure

Identifyingcollocationalexpressionsbythephraseswhosemeaningcannotbederivedinasimplewayfromthatofthecomponentwords(e.g.artificialintelligence)SelecttherangeoflengthallowedforeachcollocationalexpressionBuildalistofallpotentialexpressionsfromthecollectionwithprescribedlengththathaveaminimumfrequencyDeletesequencesthebeginorendwithatrivialword1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus44ChouekaProcedure(continued)4.Deleteexpressionsthatcontainhigh-frequencynontrivialwords5.Givenanexpressionsuchabcdevaluateanypotentialsubexpressionsforrelevance.Discardanythatarenotsufficientlyrelevant6.Trytomergesmallexpressionsintolargeandmoremeaningfulones1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus454.2SimilaritycomputationsbetweentermsTodeterminethestatisticalsimilaritybetweenpairsofterms.Dice:,ifl1=0orl2=0return0Cosine:,ifl1=0orl2=0return0 l1:#oftermsassociatedwithdocument1 l2:#oftermsassociatedwithdocument2 common:#oftermsincommonbetweenthem1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus464.3OrganizationofvocabularyTwoassumptionHigh-frequencywordshavebroadmeaningIfthedensityfunctionsofthetwoterms,pandq(ofvaryingfrequencies)havethesameshape,thenthetwowordshavesimilarmeaning.Astwoassumptions,ifpisthetermwiththehigherfrequency,thenqbecomesachildofp.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus47Identifyasetoffrequencyranges.Groupthevocabularytermsintodifferentclassesbasedontheirfrequencies.Thehighestfrequencyclassisassignedlevel0,thenext,level1andsoon.Parent-childlinksaredetermined.Create“dummy”term.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionProcess1Constructionofvocabulary2Similaritycomputations3OrganizationofvocabularyFromdocumentcollectionsMergeexistingthesaurus48MergingexistingthesauriSimple-mergeTwotermsindifferenthierarchiesaremergediftheyareidentical.Complex-mergeAnytwotermsindifferenthierarchiesaremergediftheyhave‘similar’parentsandchildren.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.ConclusionFromdocumentcollectionsMergeexistingthesaurus49conclusionThischapterbeganwithanintroductiontothesauri.Twomajorautomaticthesaurusconstructionmethodshavebeendetailed.Afewrelatedissuestothesaurihavenotbeenconsideredhere:Evaluationofthesauri.Maintenanceofthesauri.Howtoautomatetheusageofthesauri.1.Introduction2.Featuresofthesauri3.ThesaurusConstruction4.Conclusion50+B3E6H9LcOfRjUmXp!s&w)z0C4F7IaMdPhSkVnZq$t*x-A2D5G8KbNeQiTlXo#r%v(y+B3E6I9LcOgRjUmYp!t&w)z1C4F7JaMePhSkWnZq$u*x+A2D5H8KbNfQiTlXo#s%v(y0B3E6I9LdOgRjVmYp!t&w-z1C4G7JaMePhTkWnZr$u*x+A2E5H8KcNfQiUlXp#s%v)y0B3F6IaLdOgSjVmYq!t&w-z1D4G7JbMePhTkWoZr$u(x+A2E5H9KcNfRiUlXp#s&v)y0C3F6IaLdPgSjVnYq!t*w-A1D4G8JbMeQhTlWoZr%u(x+B2E5H9KcOfRiUmXp#s&v)z0C3F7IaLdPgSkVnYq$t*w-A1D5G8JbNeQhTlWo#r%u(y+B2E6H9LcOfRjUmXp!s&w)z0C4F7IaMdPhSkVnZq$t*x-A1D5G8KbNeQiTlWo#r%v(y+B3E6H9LcOgRjUmYp!s&w)z1C4F7JaMdPhSkWnZq$u*x-A2D5H8KbNfQiTlXo#s%v(y0B3E6I9LcOgRjVmYp!t&w)z1C4G7JaMePhSkWnZr$u*x+A2D5H8KcNfQiUlXo#s%v)y0B3F6I9LdOgSjVmYq!t&w-z1D4G7JbMePhTkWnZr$u(x+A2E5H8KcNfRiUlXp#s%v)y0C3F6IaLdOgSjVnYq!t*w-z1D4G8JbMeQhTkWoZr%u(x+B2E5H9KcOfRiUmXp#s&v)z0C3F7IaLdPgSjVnYq$t*w-A1D4G8JbNeQhTlWoZr%u(y+B2E6H9KcOfRjUmXp!s&v)z0C4F7IaMdPgSkVnZq$t*x-A1D5G8KbNeQiTlWo#r%u(y+B3E6H9LcOfRjUmYp!s&w)z0C4F7JaMdPhSkVnZq$u*x-A2D5G8KbNfQiTlXo#r%v(y0B3E6I9LcOgRjVmYp!t&w)z1C4G7JaMePhWo#r%u(y+B3E6H9LcOfRjUmYp!s&w)z0C4F7JaMdPhSkVnZq$u*x-A2D5G8KbNfQiTlXo#r%v(y0B3E6I9LcOgRjUmYp!t&w)z1C4F7JaMePhSkWnZq$u*x+A2D5H8KbNfQiUlXo#s%v(y0B3F6I9LdOgRjVmYq!t&w-z1C4G7JbMePhTkWnZr$u(x+A2E5H8KcNfQiUlXp#s%v)y0B3F6IaLdOgSjVmYq!t*w-z1D4G7JbMeQhTkWoZr$u(x+B2E5H9KcNfRiUmXp#s&v)y0C3F7IaLdPgSjVnYq!t*w-A1D4G8JbMeQhTlWoZr%u(x+B2E6H9KcOfRiUmXp!s&v)z0C3F7IaMdPgSkVnYq$t*x-A1D5G8JbNeQiTlWo#r%u(y+B2E6H9LcOfRjUmXp!s&w)z0C4F7IaMdPhSkVnZq$t*x-A2D5G8KbNeQiTlXo#r%v(y+B3E6I9LcOgRjUmYp!t&w)z1C4F7JaMePhSkWnZq$u*x-A2D5H8KbNfQiTlXo#s%v(y0B3E6I9LdOgRjVmYp!t&w-z1C4G7JaMePhTkWnZr$u*x+A2E5H8KcNfQiUlXp#s%v)y0B3F6I9LdOgSjVmYq!t&w-z1D4G7JbMePhTkWoZr$u(x+A2E5H9KcNfRiUlXp#s&v)y0C3F6IaLdPgSjVnYq!t*w-A1D4G8JbMeQhTkWoZr%u(x+B2E5H9KcOfRiUmXp#s&v)z0C3F7IaLdPgSkVnYq$t*w-A1D5G8JbNeQhTlWo#r%u(y+B2E6H9LcOfRjUmXp!s&w)z0C4F7IaMdPgSkVnZq$t*x-A1D5G8KbNeQiTlWo#r%v(y+B3E6H9LcOgRjUmYp!s&w)z1C4F7JaMdPhSkWnZq$u*x-A2D5H8KbNfQiTlXo#r%v(y0B3E6IPgSkVnZq$t*x-A1D5G8KbNeQiTlWo#r%v(y+B3E6H9LcOgRjUmYp!s&w)z1C4F7JaMdPhSkWnZq$u*x-A2D5G8KbNfQiTlXo#r%v(y0B3E6I9LcOgRjVmYp!t&w)z1C4G7JaMePhSkWnZr$u*x+A2D5H8KcNfQiUlXo#s%v)y0B3F6I9LdOgRjVmYq!t&w-z1C4G7JbMePhTkWnZr$u(x+A2E5H8KcNfRiUlXp#s%v)y0C3F6IaLdOgSjVnYq!t*w-z1D4G8JbMeQhTkWoZr$u(x+B2E5H9KcNfRiUmXp#s&v)y0C3F7IaLdPgSjVnYq$t*w-A1D4G8JbNeQhTlWoZr%u(y+B2E6H9KcOfRjUmXp!s&v)z0C4F7IaMdPgSkVnYq$t*x-A1D5G8JbNeQiTlWo#r%u(y+B3E6H9LcOfRjUmYp!s&w)z0C4F7JaMdPhSkVnZq$u*x-A2D5G8KbNfQiTlXo#r%v(y+B3E6I9LcOgRjUmYp!t&w)z1C4F7JaMePhSkWnZq$u*x+A2D5H8KbNfQiUlXo#s%v(y0B3FaMdPhSkVnZq$u*x-A2D5G8KbNeQiTlXo#r%v(y+B3E6I9LcOgRjUmYp!t&w)z1C4F7JaMePhSkWnZq$u*x+A2D5H8KbNfQiUlXo#s%v(y0B3F6I9LdOgRjVmYp!t&w-z1C4G7JaMePhTkWnZr$u*x+A2E5H8KcNfQiUlXp#s%v)y0B3F6IaLdOgSjVmYq!t*w-z1D4G7JbMeQhTkWoZr$u(x+B2E5H9KcNfRiUlXp#s&v)y0C3F6IaLdPgSjVnYq!t*w-A1D4G8JbMeQhTlWoZr%u(x+B2E6H9KcOfRiUmXp!s&v)z0C3F7IaMdPgSkVnYq$t*w-A1D5G8JbNeQhTlWo#r%u(y+B2E6H9LcOfRjUmXp!s&w)z0C4F7IaM