資源描述：
《detecting liars and the confused in contradictory online reviews》由會員上傳分享，免費在線閱讀，更多相關內(nèi)容在工程資料-天天文庫。
1、CHAPTER6DetectingLiarsandtheConfusedinContradictoryOnlineReviewsJacobPerkinsDidyouknowthatpeopleliefortheirownselfishreasons?Evenifthisistotallyobvioustoyou,youmaybesurprisedathowblatantthispracticehasbecomeonline,tothepointwheresomepeoplewillexplaintheirreasonsforlyingimmediatelyafterdoingso.Iknewu
2、nethicalpeoplewouldlieinonlinereviewsinordertoinflateratingsorattackcompetitors,butwhatIdidn’tknow,andonlylearnedbyaccident,isthatindividualswillsometimeswritereviewsthatcompletelycontradicttheirassociatedrating,withoutanyregardtohowitaffectsabusiness’sonlinereputation.Andoftenthisisforbusinessestha
3、tanindividuallikes.HowdidIlearnthis?Byusingratingsandreviewstocreateasentimentcorpus,Itrainedasentimentanalysisclassifierthatcouldreliablydeterminethesentimentofareview.Whileevaluatingthisclassifier,Idiscoveredthatitcouldalsodetectdiscrepanciesbe‐tweenthereviewsentimentandthecorrespondingrating,ther
4、ebyfindingliarsandconfusedreviewers.Here’sthewholestoryofhowIusedtextclassificationtoidentifyanunexpectedsourceofbaddata...WeottaAtmycompany,Weotta,1weproduceapplicationsandAPIsfornavigatinglocaldatainwaysthatpeopleactuallycareabout,sowecananswerquestionslike:Isthereakid-friendlyrestaurantnearby?Wha
5、t’sthenearesthipyogastudio?Whatconcertsarehap‐peningthisweekend?1.http://www.weotta.com83Todothis,weanalyze,aggregate,andorganizelocaldatainordertoclassifyitalongdimensionsthatwecanusetoanswerthesequestions.Thisclassificationprocessenablesustoknowwhichrestaurantsareclassy,whichbarsaredivey,andwherey
6、oushouldgoonafirstdate.Onlinebusinessreviewsareoneofthemajorinputsignalsweusetodeterminetheseclassifications.Reviewscantellusthepositiveornegativesentimentofthereviewer,aswellaswhattheyspecificallycareabout,suchasqualityofservice,ambience,andvalue.Whenweaggregatereviews,wecanlearnwhat’spopularaboutt
7、heplaceandwhypeoplelikeordislikeit.Weusemanyothersignalsbesidesreviews,butwiththeproperapplicationofnaturallanguageprocessing,2reviewsarearichsourceofsignificantinformation.GettingReviewsTogetreviews,
