【t112017-数据工程和技术分会场】用于图像标记的应用深度学习-旅行推荐引擎应用

《【t112017-数据工程和技术分会场】用于图像标记的应用深度学习-旅行推荐引擎应用》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、PragmaticDeepLearningforimagelabelling-Anapplicationtoatravelrecommendationengine主讲人：Dataiku数据科学家AlexandreHubertOutlineResultsLabelingImagesMoreimages!PragmaticdeeplearningforIntroductionandContextdummiesPostProcessingIterativebuildingofaAKA:ImageforBIonsteroidsrecommendersystem

2、DataikuDataScienceSoftwareEditorofDataikuDSS•Foundedin2013•90+employees,100+clients•Paris,New-York,London,SanFrancisco,SingaporeDESIGNPREPAREANALYSEMODELLoadandprepareVisualizeandshareBuildyouryourdatayourworkmodelsPRODUCTIONAUTOMATEMONITORSCORERe-executeyourFollowyourproduction

3、Getpredictionsworkflowateaseenvironmentinrealtime•KeyFiguresE-businessvacationretailerNegotiatethebestprizefortheirclientsDiscountluxury18Millionsofclients.HundredsofsalesopenedeverydaySaleImageisparamountPurchaseisimpulsiveSpecificitiesHighlytemporarysalesExpensiveProduct->Clas

4、sicalrecommendersystemfail->Fewrecurrentbuyers->Timeeventlinked(Christmas,ski,summer)->AppearancecountsalotIterativeBuildingofaRecommenderSystemBasicRecommendationEnginesOtherFactorsOneMetaModeltoRuleThemAllDescribeRecommendRecommendersasfeaturesCombineMachinelearningtooptimizep

5、urchasingprobabilityRecommendersystemforHomePageOrdering+7%revenue(A/Btesting)OptimizationofhomedisplayCustomervisitsCleaning,combiningRecommendationMetaModelPurchasesandenrichmentofdataEnginesSalesinformationtheapplicationGenerationofMetalmodelcombineEverycustomerisshownthe10sa

6、lesautomaticallyrunsandrecommendationsbasedonrecommendationstoheisthemostlikelytobuycompilesheterogeneoususerbehaviourdirectlyoptimizedatapurchasingprobabilitySalesImagesBatchScoringeverynightWhyuseImage?ApictureisworthathousandwordsWewantdodistinguish«Ski»«SunandBeach»Integrati

7、ngImageInformationCONTENTBASEDSea+Beach+Forest+HotelPool+PalmTreesHotel+MountainsPool+Forest+Hotel+SeaLabelingModelSalesImagesSalesdescriptionsRecommendervectorSystemImageLabellingForRecommendationEnginePragmaticDeeplearningfor“Dummies”UsingDeepLearningmodelsCommonIssues“Idon’th

8、aveGPUs“Idon’thaveadeepleaningexpert”server”“Id