文本分類綜述2002年12月報(bào)告內(nèi)容文本分類的定義和應(yīng)用文本分類的方法文本分類的評(píng)估指標(biāo)參考文獻(xiàn)和資源文本分類的定義和應(yīng)用應(yīng)用垃圾郵件的判定(spamornotspam)類別{spam,not-spam}新聞出版按照欄目分類類別{政治,體育,軍事,…}詞性標(biāo)注類別{名詞,動(dòng)詞,形容詞,…}詞義排歧類別{詞義1,詞義2,…}計(jì)算機(jī)論文的領(lǐng)域類別ACMsystemH:informationsystemsH.3:informationretrievalandstorage文本分類的方法人工方法和自動(dòng)方法人工方法結(jié)果容易理解足球and聯(lián)賽體育類費(fèi)時(shí)費(fèi)力難以保證一致性和準(zhǔn)確性(40%左右的準(zhǔn)確率)專家有時(shí)候憑空想象知識(shí)工程的方法建立專家系統(tǒng)(80年代末期)自動(dòng)的方法(學(xué)習(xí))結(jié)果可能不易理解快速準(zhǔn)確率相對(duì)高(準(zhǔn)確率可達(dá)60%或者更高)來源于真實(shí)文本，可信度高特征抽取(featureextraction)預(yù)處理去掉html一些tag標(biāo)記禁用詞(stopwords)去除、詞根還原(stemming)(中文)分詞、詞性標(biāo)注、短語識(shí)別、…詞頻統(tǒng)計(jì)TFi,j:特征i在文檔j中出現(xiàn)次數(shù)，詞頻(TermFrequency)DFi:所有文檔集合中出現(xiàn)特征i的文檔數(shù)目，文檔頻率(DocumentFrequency)數(shù)據(jù)清洗：去掉不合適的噪聲文檔或文檔內(nèi)垃圾數(shù)據(jù)文本表示向量空間模型降維技術(shù)特征選擇(FeatureSelection)特征重構(gòu)(Re-parameterisation，如LSI)文本表示向量空間模型(VectorSpaceModel)M個(gè)無序標(biāo)引項(xiàng)ti(特征)，詞根/詞/短語/其他每個(gè)文檔dj可以用標(biāo)引項(xiàng)向量來表示(a1j,a2j,…,aMj)權(quán)重計(jì)算，N個(gè)訓(xùn)練文檔AM*N=(aij)相似度比較Cosine計(jì)算內(nèi)積計(jì)算Term的粒度Character，字：中Word，詞：中國Phrase，短語：中國人民銀行Concept，概念同義詞：開心高興興奮相關(guān)詞cluster，wordcluster：葛非/顧俊N-gram，N元組：中國國人人民民銀銀行某種規(guī)律性模式：比如某個(gè)window中出現(xiàn)的固定模式DavidLewis等一致地認(rèn)為：(英文分類中)使用優(yōu)化合并后的Words比較合適特征選擇(1)基于DFTerm的DF小于某個(gè)閾值去掉(太少，沒有代表性)Term的DF大于某個(gè)閾值也去掉(太多，沒有區(qū)分度)信息增益(InformationGain,IG)：該term為整個(gè)分類所能提供的信息量(不考慮任何特征的熵和考慮該特征后的熵的差值)特征選擇(2)term的某種熵：該值越大，說明分布越均勻，越有可能出現(xiàn)在較多的類別中；該值越小，說明分布越傾斜，詞可能出現(xiàn)在較少的類別中相對(duì)熵(not交叉熵)：也稱為KL距離(Kullback-Leiblerdivergence)

，反映了文本類別的概率分布和在出現(xiàn)了某個(gè)特定詞匯條件下的文本類別的概率分布之間的距離，該值越大，詞對(duì)文本類別分布的影響也大。特征選擇(3)χ2統(tǒng)計(jì)量(念xi)：度量兩者(term和類別)獨(dú)立性的缺乏程度，χ2越大，獨(dú)立性越小，相關(guān)性越大(若AD<BC,則類和詞獨(dú)立,N=A+B+C+D)互信息(MutualInformation)：MI越大t和c共現(xiàn)程度越大ABCDt~tc~c特征選擇方法的性能比較(1)特征選擇方法的性能比較(2)特征選擇方法的性能比較(3)YangYi-ming自動(dòng)文本分類方法Rocchio方法Na?veBayeskNN方法決策樹方法decisiontreeDecisionRuleClassifierTheWidrow-HoffClassifier神經(jīng)網(wǎng)絡(luò)方法NeuralNetworks支持向量機(jī)SVM基于投票的方法(votingmethod)Rocchio方法可以認(rèn)為類中心向量法是它的特例Rocchio公式分類類C中心向量的權(quán)重訓(xùn)練樣本中正例個(gè)數(shù)文檔向量的權(quán)重Na?veBayes參數(shù)計(jì)算Bayes公式?jīng)Q策樹方法構(gòu)造決策樹CARTC4.5(由ID3發(fā)展而來)CHAID決策樹的剪枝(pruning)DecisionRuleLearningwheat&formWHEATwheat&commodityWHEATbushels&exportWHEATwheat&agricultureWHEATwheat&tonnesWHEATwheat&winter&~softWHEAT(粗糙集)RoughSet邏輯表達(dá)式(AQ11算法)學(xué)習(xí)到如下規(guī)則支持向量機(jī)

SupportVectorMachineSupportVectorOptimalSeparatingHyperplane基于投票的方法Bagging方法訓(xùn)練R個(gè)分類器fi，分類器之間其他相同就是參數(shù)不同。其中fi是通過從訓(xùn)練集合中(N篇文檔)隨機(jī)取(取后放回)N次文檔構(gòu)成的訓(xùn)練集合訓(xùn)練得到的。對(duì)于新文檔d，用這R個(gè)分類器去分類，得到的最多的那個(gè)類別作為d的最終類別Boosting方法類似Bagging方法，但是訓(xùn)練是串行進(jìn)行的，第k個(gè)分類器訓(xùn)練時(shí)關(guān)注對(duì)前k-1分類器中錯(cuò)分的文檔，即不是隨機(jī)取，而是加大取這些文檔的概率AdaBoostAdaBoostMH文本分類的評(píng)估指標(biāo)分類方法的評(píng)估鄰接表每個(gè)類Precision=a/(a+b),Recall=a/(a+c),fallout=b/(b+d)=falsealarmrate,accuracy=(a+d)/(a+b+c+d),error=(b+c)/(a+b+c+d)=1-accuracy,missrate=1-recallF=(β2+1)p.r/(β2p+r)BreakEvenPoint,BEP,p=r的點(diǎn)如果多類排序輸出，采用interpolated11pointaverageprecision所有類：宏平均:對(duì)每個(gè)類求值，然后平均微平均:將所有文檔一塊兒計(jì)算，求值真正對(duì)的錯(cuò)誤標(biāo)YESab標(biāo)NOcd其他分類方法RegressionbasedonLeastSquaresFit(1991)NearestNeighborClassification(1992)*BayesianProbabilisticModels(1992)*SymbolicRuleInduction(1994)DecisionTree(1994)*NeuralNetworks(1995)Rocchioapproach(traditionalIR,1996)*SupportVectorMachines(1997)BoostingorBagging(1997)*HierarchicalLanguageModeling(1998)First-Order-LogicRuleInduction(1999)MaximumEntropy(1999)HiddenMarkovModels(1999)Error-CorrectingOutputCoding(1999)...參考文獻(xiàn)文獻(xiàn)及其他資源PapersK.AasandL.Eikvil.Textcategorisation:Asurvey.Technicalreport,NorwegianComputingCenter,June1999XiaomengSu，“Textcategorization”，LessonPresentationYimingYangandXinLiu.1999."Are-examinationoftextcategorizationmethods."22ndAnnualInternationalSIGIRASurveyonTextCategorization,NLPLab,KoreanU.龐劍峰，基于向量空間模型的自反饋的文本分類系統(tǒng)的研究與實(shí)現(xiàn)，中科院計(jì)算所碩士論文，2001

黃萱菁等，獨(dú)立于語種的文本分類方法，中文信息學(xué)報(bào)，2000年第6期Software:RainbowBoosTexterTiMBLC4.5Corpus

謝謝！Wangbin@7>3@N)J&G#CWySuOrhnd 9;5<2=M(I$EYBVxRtjpfmc8.4`0-L*H!DXzTvQsioeka:7>3@N)J%F#CWySuOqhnd 9;5~1=M(I$EYAUxRtjpflb8.4`0-K&H!DXzTvPsioeka:6<3@N)J%FZBWySuOqgmd 9;5~1+L(I$EYAUwRtjpflb^7.4`0-K&G#DXzTvPrhoeka:6<2=N)J%FZBVySuOqgmc9;5~1+L*H$EYAUwQsjpflb^7>3`0-K&G#CXzTvPrhndka:6<2=M(J%FZBVxRuOqgmc8.5~1+L*H!EYAUwQsioflb^7>3@N-K&G#CWyTvPrhnd 9:6<2=M(I%FZBVxRtjqgmc8.4`1+L*H!DXAUwQsioekb^7>3@N)K&G#CWySuPrhnd 9;5<2=M(I$EZBVxRtjpfmc8.4`0+L*H!DXzTwQsioeka:7>3@N)J%G#CWySuOqhnd 9;5~2=M(I$EYAVxRtjpflb8.4`0-K*H!DXzTvPsioeka:6>3@N)J%FZCWySuOqgmd 9;5~1+M(I$EYAUwRtjpflb^8.4`0-K&G!DXzTvPrhoeka:6<2@N)J%FZBVySuOqgmc 9;5~1+L*I$EYAUwQsjpflb^7>4`0-K&G#CXzTvPrhneka:6<2=M)J%FZBVxRuOqgmc8;5~1+L*H!EYAUwQsipflb^7>3@0-K&G#CWyTvPrhnd a:6<2=M(I%FZBVxRtOqgmc8.4~1+L*H!DXAUwQsioelb^7>3@N)K&G#CWySvPrhnd 9;6<2=M(I$EZBVxRtjpgmc8.4`0+L*H!DXzUwQsioeka^7>3@N)J%G#CWySuOrhnd 9;5~2=M(I$EYBVxRtjpflc8.4`0-K*H!DXzTvQsioeka:6>3@N)J%F#CWySuOqgnd 9;5~1+M(I$EYAUxRtjpflb^8.4`0-K&H!DXzTvPrioeka:6<2@N)J%FZBWySuOqgmc 9;5~1+L(I$EYAUwQtjpflb^7>4`0-K&G#DXzTvPrhneka:6<2=N)J%FZBVxSuOqgmc8;5~1+L*H$EYAUwQsipflb^7>3`0-K&G#CWzTvPrhnd a:6<2=M(J%FZBVxRtOqgmc8.4~1+L*H!DYAUwQsioelb^7>3@N-K&G#CWySvPrhnd 9;6<2=M(I$FZBVxRtjpgmc8.4`1+L*H!DXzUwQhoeka:6<2@N)J%FZBVySuOqgmc9;5~1+L*I$EYAUwQsjpflb^7>4`0-K&G#CXzTvPrhndka:6<2=M)J%FZBVxRuOqgmc8;5~1+L*H!EYAUwQsioflb^7>3@0-K&G#CWyTvPrhnd a:6<2=M(I%FZBVxRtjqgmc8.4~1+L*H!DXAUwQsioelb^7>3@N)K&G#CWySuPrhnd 9;6<2=M(I$EZBVxRtjpgmc8.4`0+L*H!DXzTwQsioeka^7>3@N)J%G#CWySuOrhnd 9;5~2=M(I$EYAVxRtjpflc8.4`0-K*H!DXzTvQsioeka:6>3@N)J%FZCWySuOqgnd 9;5~1+M(I$EYAUxRtjpflb^8.4`0-K&G!DXzTvPrioeka:6<2@N)J%FZBWySuOqgmc 9;5~1+L*I$EYAUwQtjpflb^7>4`0-K&G#DXzTvPrhneka:6<2=M)J%FZBVxSuOqgmc8;5~1+L*H$EYAUwQsipflb^7>3@0-K&G#CWzTvPrhnd a:6<2=M(J%FZBVxRtOqgmc8.4~1+L*H!DYAUwQsioelb^7>3@N-K&G#CWySvPrhnd 9;6<2=M(I$FZBVxRtjpgmc8.4`1+L*H!DXzUwQsioeka^7>3@N)J&G#CWySuOrhnd 9;5<2=M(I$EYBVxRtjpflc8.4`0-L*H!DXzTvQsioeka:7>3@N)J%F#CWySuOqgnd 9;5~1=M(I$EYAUxRtjpflb8.4`0-K&H!DXzTvPrioeka:6<3@N)J%FZBWySuOqgmd 9;5~1+L(I$EYAUwQtjpflb^7.4`0-K&G#DXzTvPrhoeka:6<2=N)J%FZBVxSuOqgmc9;5~1+L*H$EYAUwQsjpflb^7>3`0-K&G#CWzTvPrhndka:6<2=M(J%FZBVxRuOqgmc8.5~1+L*H!DYAUwQsioflb^7>3@N-K&G#CWyTvPrhnd 9:6<2=M(I$FZBVxRtjqgmc8.4`1+L*H!DXAUwQsioek6<2@N)J%FZBWySuOqgmc 9;5~1+L*I$EYAUwQtjpflb^7>4`0-K&G#DXzTvPrhneka:6<2=M)J%FZBVxSuOqgmc8;5~1+L*H$EYAUwQsipflb^7>3@0-K&G#CWzTvPrhnd a:6<2=M(J%FZBVxRtOqgmc8.4~1+L*H!DYAUwQsioelb^7>3@N-K&G#CWySvPrhnd 9;6<2=M(I$FZBVxRtjpgmc8.4`1+L*H!DXzUwQsioeka^7>3@N)J&G#CWySuOrhnd 9;5<2=M(I$EYBVxRtjpflc8.4`0-L*H!DXzTvQsioeka:6>3@N)J%F#CWySuOqgnd 9;5~1=M(I$EYAUxRtjpflb^8.4`0-K&H!DXzTvPrioeka:6<3@N)J%FZBWySuOqgmc 9;5~1+L(I$EYAUwQtjpflb^7.4`0-K&G#DXzTvPrhneka:6<2=N)J%FZBVxSuOqgmc9;5~1+L*H$EYAUwQsipflb^7>3`0-K&G#CWzTvPrhndka:6<2=M(J%FZBVxRtOqgmc8.5~1+L*H!DYAUwQsioflb^7>3@N-K&G#CWySvPrhnd 9:6<2=M(I$FZBVxRtjqgmc8.4`1+L*H!DXzUwQsioekb^7>3@N)J&G#CWySuPrhnd 9;5<2=M(I$EYBVxRtjpfmc8.4`0-L*H!DXzTwQsioeka:7>3@N)J%F#CWySuOqhnd 9;5~1=M(I$EYAVxRtjpflb8.4`0-K&H!DXzTvPsid a:6<2=M(I%FZBVxRtOqgmc8.4~1+L*H!DYAUwQsioelb^7>3@N)K&G#CWySvPrhnd 9;6<2=M(I$FZBVxRtjpgmc8.4`0+L*H!DXzUwQsioeka^7>3@N)J&G#CWySuOrhnd 9;5~2=M(I$EYBVxRtjpflc8.4`0-L*H!DXzTvQsioeka:6>3@N)J%F#CWySuOqgnd 9;5~1=M(I$EYAUxRtjpflb^8.4`0-K&H!DXzTvPrioeka:6<3@N)J%FZBWySuOqgmc 9;5~1+L(I$EYAUwQtjpflb^7.4`0-K&G#DXzTvPrhneka:6<2=N)J%FZBVxSuOqgmc9;5~1+L*H$EYAUwQsipflb^7>3`0-K&G#CWzTvPrhndka:6<2=M(J%FZBVxRtOqgmc8.5~1+L*H!DYAUwQsioflb^7>3@N-K&G#CWySvPrhnd 9:6<2=M(I$FZBVxRtjqgmc8.4`1+L*H!DXzUwQsioekb^7>3@N)J&G#CWySuPrhnd 9;5<2=M(I$EYBVxRtjpfmc8.4`0-L*H!DXzTwQsioeka:7>3@N)J%F#CWySuOqhnd 9;5~1=M(I$EYAVxRtjpb^7>3@0-K&G#CWzTvPrhnd a:6<2=M(I%FZBVxRtOqgmc8.4~1+L*H!DYAUwQsioelb^7>3@N)K&G#CWySvPrhnd 9;6<2=M(I$FZBVxRtjpgmc8.4`0+L*H!DXzUwQsioeka^7>3@N)J&G#CWySuOrhnd 9;5~2=M(I$EYBVxRtjpflc8.4`0-L*H!DXzTvQsioeka:6>3@N)J%F#CWySuOqgnd 9;5~1=M(I$EYAUxRtjpflb^8.4`0-K&H!DXzTvPrioeka:6<3@N)J%FZBWySuOqgmc 9;5~1+L(I$EYAUwQtjpflb^7.4`0-K&G#DXzTvPrhneka:6<2=N)J%FZBVxSuOqgmc8;5~1+L*H$EYAUwQsipflb^7>3`0-K&G#CWzTvPrhnd a:6<2=M(J%FZBVxRtOqgmc8.5~1+L*H!DYAUwQsioelb^7>3@N-K&G#CWySvPrhnd 9:6<2=M(I$FZBVxRtjpgmc8.4`1+L*H!DXzUwQsioekb^7>3@N)J&G#CWySuOrhnd 9;5<2=M(I$EYBVxRtjpfmc8.4`0-L*H!DXzTvQsieka:6<2=M)J%FZBVxRuOqgmc8;5~1+L*H!EYAUwQsipflb^7>3@0-K&G#CWyTvPrhnd a:6<2=M(I%FZBVxRtOqgmc8.4~1+L*H!DXAUwQsioelb^7>3@N)K&G#CWySvPrhnd 9;6<2=M(I$EZBVxRtjpgmc8.4`0+L*H!DXzUwQsioeka^7>3@N)J%G#CWySuOrhnd 9;5~2=M(I$EYBVxRtjpflc8.4`0-K*H!DXzTvQsioeka:6>3@N)J%F#CWySuOqgnd 9;5~1+M(I$EYAUxRtjpflb^8.4`0-K&H!DXzTvPrioeka:6<2@N)J%FZBWySuOqgmc 9;5~1+L(I$EYAUwQtjpflb^7>4`0-K&G#DXzTvPrhneka:6<2=N)J%FZBVxSuOqgmc8;5~1+L*H$EYAUwQsipflb^7>3`0-K&G#CWzTvPrhnd a:6<2=M(J%FZBVxRtOqgmc8.5~1+L*H!DYAUwQsioelb^7>3@N-K&G#CWySvPrhnd 9:6<2=M(I$FZBVxRtjpgmc8.4`1+L*H!DXzUwQsioekb^7>3@N)J&G#CWySuOrhnd 9;5<2=M(I$EYBVxRtflb^7>4`0-K&G#CXzTvPrhneka:6<2=M)J%FZBVxRuOqgmc8;5~1+L*H!EYAUwQsipflb^7>3@0-K&G#CWyTvPrhnd a:6<2=M(I%FZBVxRtOqgmc8.4~1+L*H!DXAUwQsioelb^7>3@N)K&G#CWySvPrhnd 9;6<2=M(I$EZBVxRtjpgmc8.4`0+L*H!DXzUwQsioeka^7>3@N)J%G#CWySuOrhnd 9;5~2=M(I$EYBVxRtjpflc8.4`0-K*H!DXzTvQsioeka:6>3@N)J%F#CWySuOqgnd 9;5~1+M(I$EYAUxRtjpflb^8.4`0-K&H!DXzTvPrioeka:6<2@N)J%FZBWySuOqgmc 9;5~1+L(I$EYAUwQtjpflb^7>4`0-K&G#DXzTvPrhneka:6<2=N)J%FZBVxSuOqgmc8;50-L*H!DXzTwQsioeka:7>3@N)J%G#CWySuOqhnd 9;5~1=M(I$EYAVxRtjpflb8.4`0-K*H!DXzTvPsioeka:6<3@N)J%FZCWySuOqgmd 9;5~1+M(I$EYAUwRtjpflb^7.4`0-K&G!DXzTvPrhoeka:6<2@N)J%FZBVySuOqgmc9;5~1+L*I$EYAUwQsjpflb^7>4`0-K&G#CXzTvPrhndka:6<2=M)J%FZBVxRuOqgmc8;5~1+L*H!EYAUwQsioflb^7>3@0-K&G#CWyTvPrhnd a:6<2=M(I%FZBVxRtjqgmc8.4~1+L*H!DXAUwQsioelb^7>3@N)K&G#CWySuPrhnd 9;6<2=M(I$EZBVxRtjpgmc8.4`0+L*H!DXzTwQsioeka^7>3@N)J%G#CWySuOrhnd 9;5~2=M(I$EYAVxRtjpflc8.4`0-K*H!DXzTvQsioeka:6>3@N)J%FZCWySuOqgnd 9;5~1+M(I$EYAUsioflb^7>3@N-K&G#CWyTvPrhnd 9:6<2=M(I$FZBVxRtjqgmc8.4`1+L*H!DXAUwQsioekb^7>3@N)J&G#CWySuPrhnd 9;5<2=M(I$EZBVxRtjpfmc8.4`0-L*H!DXzTwQsioeka:7>3@N)J%G#CWySuOqhnd 9;5~1=M(I$EYAVxRtjpflb8.4`0-K*H!DXzTvPsioeka:6<3@N)J%FZCWySuOqgmd 9;5~1+M(I$EYAUwRtjpflb^7.4`0-K&G!DXzTvPrhoeka:6<2@N)J%FZBVySuOqgmc9;5~1+L*I$EYAUwQsjpflb^7>4`0-K&G#CXzTvPrhndka:6<2=M)J%FZBVxRuOqgmc8;5~1+L*H!EYAUwQsioflb^7>3@0-K&G#CWyTvPrhnd a:6<2=M(I%FZBVxRtjqgmc8.4~1+&H!DXzTvPrioeka:6<3@N)J%FZBWySuOqgmc 9;5~1+L(I$EYAUwQtjpflb^7.4`0-K&G#DXzTvPrhneka:6<2=N)J%FZBVxSuOqgmc9;5~1+L*H$EYAUwQsipflb^7>3`0-K&G#CWzTvPrhndka:6<2=M(J%FZBVxRtOqgmc8.5~1+L*H!DYAUwQsioflb^7>3@N-K&G#CWySvPrhnd 9:6<2=M(I$FZBVxRtjqgmc8.4`1+L*H!DXzUwQsioekb^7>3@N)J&G#CWySuPrhnd 9;5<2=M(I$EYBVxRtjpfmc8.4`0-L*H!DXzTwQsioeka:7>3@N)J%F#CWySuOqhnd 9;5~1=M(I$EYAVxRtjpflb8.4-K&G#CWzTvPrhnd a:6<2=M(I%FZBVxRtOqgmc8.4~1+L*H!DYAUwQsioelb^7>3@N)K&G#CWySvPrhnd 9;6<2=M(I$FZBVxRtjpgmc8.4`0+L*H!DXzUwQsioeka^7>3@N)J&G#CWySuOrhnd 9;5~2=M(I$EYBVxRtjpflc8.4`0-L*H!DXzTvQsioeka:6>3@N)J%F#CWySuOqgnd 9;5~1=M(I$EYAUxRtjpflb^8.4`0-K&H!DXzTvPrioeka:6<3@N)J%FZBWySuOqgmc 9;5~1+L(I$EYAUwQtjpflb^7.4`0-K&G#DXzTvPrhneka:6<2=N)J%FZBVxSuOqgmc9;5~1+L*H$EYAUwQsipflb^7>3`0-K&G#CWzTvPrhndka:6<2=M(J%FZBVxRtOqgmc8.0-K*H!DXzTvPsioeka:6>3@N)J%FZCWySuOqgmd 9;5~1+M(I$EYAUwRtjpflb^8.4`0-K&G!DXzTvPrhoeka:6<2@N)J%FZBVySuOqgmc 9;5~1+L*I$EYAUwQsjpflb^7>4`0-K&G#CXzTvPrhneka:6<2=M)J%FZBVxRuOqgmc8;5~1+L*H!EYAUwQsipflb^7>3@0-K&G#CWyTvPrhnd a:6<2=M(I%FZBVxRtOqgmc8.4~1+L*H!DXAUwQsioelb^7>3@N)K&G#CWySvPrhnd 9;6<2=M(I$EZBVxRtjpgmc8.4`0+L*H!DXzUwQsioeka^7>3@N)J%G#CWySuOrhnd 9;5~2=M(I$EYBVxRtjpflc8.4`0-K*H!DXzTvQsioeka:6>3@N)J%F#VxRuOqgmc8.5~1+L*H!EYAUwQsioflb^7>3@N-K&G#CWyTvPrhnd 9:6<2=M(I%FZBVxRtjqgmc8.4`1+L*H!DXAUwQsioekb^7>3@N)K&G#CWySuPrhnd 9;5<2=M(I$EZBVxRtjpfmc8.4`0+L*H!DXzTwQsioeka:7>3@N)J%G#CWySuOqhnd 9;5~2=M(I$EYAVxRtjpflb8.4`0-K*H!DXzTvPsioeka:6>3@N)J%FZCWySuOqgmd 9;5~1+M(I$EYAUwRtjpflb^8.4`0-K&G!DXzTvPrhoeka:6<2@N)J%FZBVySuOqgmc 9;5~1+L*I$EYAUwQsjpflb^7>4`0-G#CWySuOrhnd 9;5<2=M(I$EYBVxRtjpfmc8.4`0-L*H!DXzTvQsioeka:7>3@N)J%F#CWySuOqhnd 9;5~1=M(I$EYAUxRtjpflb8.4`0-K&H!DXzTvPsioeka:6<3@N)J%FZBWySuOqgmd 9;5~1+L(I$EYAUwRtjpflb^7.4`0-K&G#DXzTvPrhoeka:6<2=N)J%FZBVySuOqgmc9;5~1+L*H$EYAUwQsjpflb^7>3`0-K&G#CXzTvPrhndka:6<2=M(J%FZBVxRuOqgmc8.5~1+L*H!EYAUwQsioflb^7>3@N-K&G#CWSuOqgnd 9;5~1+M(I$EYAUxRtjpflb^8.4`0-K&G!DXzTvPrioeka:6<2@N)J%FZBWySuOqgmc 9;5~1+L*I$EYAUwQtjpflb^7>4`0-K&G#DXzTvPrhneka:6<2=M)J%FZBVxSuOqgmc8;5~1+L*H$EYAUwQsipflb^7>3@0-K&G#CWzTvPrhnd a:6<2=M(J%FZBVxRtOqgmc8.4~1+L*H!DYAUwQsioelb^7>3@N-K&G#CWySvPrhnd 9;6<2=M(I$FZBVxRtjpgmc8.4`1+L*H!DXzUwQsioeka^7>3@N)J&G#CWySuOrhnd 9;5<2=M(IYAUwQsjpflb^7>4`0-K&G#CXzTvPrhndka:6<2=M)J%FZBVxRuOqgmc8;5~1+L*H!EYAUwQsioflb^7>3@0-K&G#CWyTvPrhnd a:6<2=M(I%FZBVxRtjqgmc8.4~1+L*H!DXAUwQsioelb^7>3@N)K&G#CWySuPrhnd 9;6<2=M(I$EZBVxRtjpgmc8.4`0+L*H!DXzTwQsioeka^7>=N)J%FZBVxSuOqgmc9;5~1+L*H$EYAUwQsipflb^7>3`0-K&G#CWzTvPrhndka:6<2=M(J%FZBVxRtOqgmc8.5~1+L*H!DYAUwQsioflb^7