【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.pdf

《【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.pdf》由会员分享，可在线阅读，更多相关《【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.pdf（27页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、Pragmatic Deep Learning for image labelling-An application to a travel recommendation engine主讲人：Dataiku数据科学家 Alexandre Hubert Introduction and Context Iterative building of a recommender systemLabeling ImagesPragmatic deep learning for dummiesPost Processing AKA:Image for BI on steroidsOutlineResult

2、sMore images!Dataiku Founded in 2013 90+employees,100+clients Paris,New-York,London,San Francisco,SingaporeData Science Software Editor of Dataiku DSSDESIGNLoad and prepareyour dataPREPAREBuild yourmodelsMODELVisualize and shareyour work ANALYSERe-execute your workflow at easeAUTOMATEFollow your pro

3、duction environmentMONITORGet predictionsin real timeSCOREPRODUCTIONE-business vacation retailerNegotiate the best prize for their clientsDiscount luxuryKey FiguresSale Image is paramountPurchase is impulsive18 Millions of clients.Hundreds of sales opened everydaySpecificitiesHighly temporary sales-

4、Classical recommender system fail-Time event linked(Christmas,ski,summer)Expensive Product-Few recurrent buyers-Appearance counts a lotIterative Building of a Recommender SystemBasic Recommendation EnginesOther FactorsOne Meta Model to Rule Them AllRecommendersasfeaturesMachinelearningtooptimizepurc

5、hasingprobabilityCombineRecommendDescribeCleaning,combining and enrichment of dataRecommendation EnginesOptimization of home displaythe application automatically runs and compiles heterogeneous dataGeneration of recommendations based on user behaviourEvery customer is shown the 10 sales he is the mo

6、st likely to buyCustomer visitsPurchases Sales ImagesMetal model combine recommendations to directly optimize purchasing probabilityMeta ModelRecommender system for Home Page Ordering+7%revenueSales information(A/B testing)Batch Scoring every nightWhy use Image?We want do distinguish Sun and Beach S

7、ki A picture is worth a thousand wordsSales ImagesIntegrating Image InformationLabeling ModelPool+Palm Trees Hotel+MountainsPool+Forest+Hotel+SeaSea+Beach+Forest+HotelSales descriptions vectorCONTENTBASEDRecommender SystemImage Labelling For Recommendation EnginePragmaticDeeplearningfor“Dummies”Usin

8、g Deep Learning modelsCommon Issues“I dont have GPUs server”“I dont have a deep leaning expert”“I dont have labelled data”(or too few)“I dont have the time to wait for model training”I dont want to pay to pay for private apis”/“Im afraid their labelling will change over time”“I dont have(or few)labe

9、lled data”-Is there similar data?Solution 1:Pre trained models PLACESDATABASEUSSUNDATABASE205categories2.5Mimages307categories110Kimagestower:0.53skyscraper:0.26swimming_pool/outdoor:0.65inn/outdoor:0.06Solution 1:Pre trained models If there is open data,there is an open pre trained model!Kudos to t

10、he community Check the licensingExamplewithPlaces(CaffeModelZoo):Solution 2:Transfer Learning Credit:Fei-FeiLi&AndrejKarpathy&JustinJohnsonhttp:/cs231n.stanford.edu/slides/winter1516_lecture11.pdfPLACESDATABASEOURDATASUNDATABASETraining(optional)Pre-trainedmodelVGG16tower:0.53skyscraper:0.26Re-Train

11、ingTransferredData:LastconvolutionallayerfeaturesRe-trainedmodelTensorFlow2fullyconnectedlayersCaffeModelZooGPUCPUGPULeverage existing knowledge!Solution 2:Transfer Learning Accuracy:72%,Top-5Acc:90%stateoftheartondatasetalonePost Treatment&Results(Or how we transfer the labelling information)UsingI

12、magesinformationforBIonsteroidsLabels post-processingComplementary informationRedondant informationIssue with our approach:Solution:NMF Matrix FactorizationDimensionReductionExplicabilitySparsityBalancednessImage content detectionTopic scores determine the importance of topics in an imageTOPICTOPICS

13、CORE(%)Golfcourse Fairway Puttinggreen31Hotel Inn Apartmentbuildingoutdoor30Swimmingpool LidoDeck Hottuboutdoor22Beach Coast-Harbor17TOPICTOPICSCORE(%)Tower Skyscraper Officebuilding62Bridge River Viaduct38Results?1)Visits:France and Morocco Pool displayed2)First Recommendation Mostly France&Mediter

14、ranean Fails to display pools3)Only Images recommendation Pool all around the world Does not respect budget 4)Third column=Right Mix 1)2)3)4)ConclusionDo iterative data science!Start simple and grow Evaluate at each stepsImage labelling=BI on steroidsTransfer LearningKick-start your projectGain time

15、 and moneyAny Data Scientist can do itDeep Learning Dont start from scratch!Is there existing data?Is there a pre-trained model?Learned along the way Whats next?Attractiveness=%visitswithtag/%saleswithtagForskisales,indoorpicturesperformsbetterWhats Next?KenyaPragueBerlinCambodiaWhats Next?Customize the Image!KenyaPragueBerlinCambodiaThank you for your attention!