有哪些信誉好的足球投注网站- 1、本文档共67页，可阅读全部内容。
- 2、有哪些信誉好的足球投注网站（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
- 3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
CS276B
TextInformationRetrieval,Mining,andExploitationLecture523January20031
Recap2
Today’stopicsFeatureselectionfortextclassificationMeasuringclassificationperformanceNearestneighborcategorization3
FeatureSelection:Why?Textcollectionshavealargenumberoffeatures10,000–1,000,000uniquewords–andmoreMakeusingaparticularclassifierfeasibleSomeclassifierscan’tdealwith100,000soffeat’sReducetrainingtimeTrainingtimeforsomemethodsisquadraticorworseinthenumberoffeatures(e.g.,logisticregression)ImprovegeneralizationEliminatenoisefeaturesAvoidoverfitting4
Recap:FeatureReductionStandardwaysofreducingfeaturespacefortextStemmingLaugh,laughs,laughing,laughed-laughStopwordremovalE.g.,eliminateallprepositionsConversiontolowercaseTokenizationBreakonallspecialcharacters:fire-fighter-fire,fighter5
FeatureSelectionYangandPedersen1997ComparisonofdifferentselectioncriteriaDF–documentfrequencyIG–informationgainMI–mutualinformationCHI–chisquareCommonstrategyComputestatisticforeachtermKeepntermswithhighestvalueofthisstatistic6
InformationGain7
(Pointwise)MutualInformation8
Chi-SquareTermpresentTermabsentDocumentbelongstocategoryABDocumentdoesnotbelongtocategoryCDX^2=N(AD-BC)^2/((A+B)(A+C)(B+D)(C+D))UseeithermaximumoraverageX^2Valueforcompleteindependence?9
DocumentFrequencyNumberofdocumentsatermoccursinIssometimesusedforeliminatingbothveryfrequentandveryinfrequenttermsHowisdocumentfrequencymeasuredifferentfromtheother3measures?10
YangPedersen:ExperimentsTwoclassificationmethodskNN(knearestneighbors;morelater)LinearLeastSquaresFitRegressionmethodCollectionsReuters-2217392categories16,000uniquetermsOhsumed:subsetofmedline14,000categories72,000uniquetermsLtctermweighting11
YangPedersen:ExperimentsChoosefeaturesetsizePreprocesscollection,discardingnon-selectedfeatures/wordsApplyt
文档评论（0）