欢迎来到天天文库
浏览记录
ID:40353465
大小:653.91 KB
页数:10页
时间:2019-07-31
《Lec14_text_classification》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、OutlineTextClassificationandBasicConceptsTextClassificationSentimentAnalysisTextRepresentationFeatureSelectionClassificationAlgorithmRuiXiaSentimentAnalysisrxia@njust.edu.cnTraditionalMethods2013.11.23NewChallengesNewDirections2Real-worldPR&MLApplicationsStructureofAPR&M
2、LSystemTrainingSamples(LabeledPatterns)FeatureFeatureClassPatternClassifierRepresentationSelectionLabel34ATextClassificationSystemOutlineBasicConceptsTextClassificationTextRepresentationFeatureSelectionClassificationAlgorithmSentimentAnalysisTraditionalMethodsNewChallenge
3、s561TextRepresentationTermWeightVectorSpaceModel(VSM)BOOL(presence)1,iftexistsinidkkialsocalledBag-of-words(BOW)model0,otherwiseTermfrequency(TF)tfkikiInversedocumentNlogfrequency(IDF)idfiNTF-IDFtflogkikidfi78AnExampleofTextRepresentationAnExampleofTR(cont.)Trai
4、ningdata(labeleddocuments)Bagofwords(containing40words)教育体育19582008奥林匹克北京博弈场地创创建大学北京理工大学计算机北京理工大学体育馆的第四第五东亚夺冠高校计算机奖牌届锦专业创建于1958年是2008年中国北京奥标赛军团理工男女年排球设立是双双体育是中国最早设立计算林匹克运动会的排球馆新高学子于预赛运动会在之一中中国专机专业的高校之一预赛场地业总数最早北京理工大学学子在第五届东亚运动会中第四届中国计算机博国军团奖牌总数创新弈锦标赛中夺冠高男女排球双双夺冠910OutlineFeature
5、Selection-FrequencyFilterBasicConceptsDocumentFrequency(DF)TextClassificationFeaturesarerankedaccordingtoitsDocumentTextRepresentationFrequencyinthetrainingcorpusFeatureSelectionTermFrequency(TF)DocumentalFrequency(DF)MutualInformation(MI)FeaturesarerankedaccordingtoitsTe
6、rmInformationGain(IG)FrequencyinthetrainingcorpusClassifierDesignShortcomingsSentimentAnalysisUnsupervised-lackoftheclassinformation11122BasicProbabilityEstimateConceptsfromInformationTheoryEntropyAstatistictableforfeaturetandclasscPc()j(AijCij)NallHX()px()log()pxijxPt
7、()(AB)NiijijallJointEntropyPt()(CD)NclassiijijallfeaturecjcjA1HXY(,)pxy(,)log(,)pxyPc(
8、)tijxyABjitiijijAijBijCConditionalEntropyC1tCDijiijijPc(j
9、)tiHYX(
10、)pxHYX()(
11、x)pxy(,)log(
12、)pyxCDCijijxxyHYX(
13、)HXY(,)HX()1314FeatureSelection-MIFeatureSelection-IGMutu
14、alInformation(MI)InformationGain(IG)MIo
此文档下载收益归作者所有
点击更多查看相关文章~~