A Sentiment-aligned Topic Model for Product Aspect Rating Prediction

《A Sentiment-aligned Topic Model for Product Aspect Rating Prediction 》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、ASentiment-alignedTopicModelforProductAspectRatingPredictionHaoWangMartinEsterSchoolofComputingScienceSchoolofComputingScienceSimonFraserUniversitySimonFraserUniversityBurnaby,BC,CanadaBurnaby,BC,Canadahwa63@sfu.caester@sfu.caAbstractnewlineofresearchonaspect-levelopinionmin

2、-ing(HuandLiu,2004).Aspect-basedopinionmininghasattractedAnaspectreferstoarateablefeature,suchaslotsofattentiontoday.Inthispaper,westaffandlocationinhotelreviews,orsizeandaddresstheproblemofproductaspectrat-batteryfordigitalcamerareviews.Inthispaper,ingprediction,wherewewoul

3、dliketoex-wedealwiththeproblemofproductaspectrat-tracttheproductaspects,andpredictas-ingprediction.Theinputisacollectionofprod-pectratingssimultaneously.Topicmod-ucts,andeachproductisassociatedwithasetofelshavebeenwidelyadaptedtojointlyreviews.Thegoalistoextractthecorpus-lev

4、elas-modelaspectsandsentiments,butexist-pects,andpredicttheaspectratingsforeachprod-ingmodelsmaynotdothepredictiontaskuct.Thiskindoffine-grainedsentimentanalysiswellduetotheirweaknessinsentimentwillhelpusersefficientlydigestthereviews,andextraction.Thesentimenttopicsusuallygai

5、nmoreinsightintotheproductquality.donothaveclearcorrespondencetocom-Theproductaspectratingpredictionproblemmonlyusedratings,andthemodelmayusuallyinvolvestwosubtasks:aspectextractionfailtoextractcertainkindsofsentimentsandsentimentidentification(TitovandMcDonald,duetoskeweddat

6、a.Totacklethisprob-2008b).Givensometext,wewouldliketoknowlem,weproposeasentiment-alignedtopicwhataspectsittalksabout,andwhatkindofsen-model(SATM),whereweincorporatetwotimentsareexpressed.Forexample,givenasen-typesofexternalknowledge:product-tence“theroomisfilthy”,wewouldliket

7、oknowleveloverallratingdistributionandword-thatittalksabouttheaspect“room”.Also,“filthy”levelsentimentlexicon.Experimentsonisasentimentword,anditexpressesstronglyneg-realdatasetdemonstratethatSATMisef-ativesentimenttowardstheaspect“room”.fectiveonproductaspectratingprediction

8、,Topicmodels(Bleietal.,2003;Hofmann,1999)anditachievesbetterperformancecom-