使用 GenAI 将视频广告分类扩展到数百万个类别.pdf

1、2024 Databricks Inc.All rights reserved1Video AD Video AD Classification Classification Across Millions Across Millions of Classesof ClassesPuneet JainPuneet JainThierry SteenberghsThierry SteenberghsJames KimJames Kim2024 Databricks Inc.All rights reservedMotivation&ChallengesSample CodeRay on Data

2、bricks with Spark structured Ray on Databricks with Spark structured streamingstreamingClassification with GenAI across millions Classification with GenAI across millions of classesof classes2Motivation&ChallengesProcess of the ML pipelineSample CodeQ&AAGENDAAGENDAWhat we will cover todayWhat we wil

3、l cover today2024 Databricks Inc.All rights reserved The amount of ads is increasing exponentially year over year.(Online Advertising Revenue went from 8 to 225 Billion from 2000 to 2023)Almost doubled from 2020 to 2023 from 140 to 225 billion.Needed an automated solution to solve the increasing num

4、ber of ads3CLASSIFICATION WITH GENAICLASSIFICATION WITH GENAIMotivation and ChallengesMotivation and Challenges2024 Databricks Inc.All rights reserved MediaRadar|Vivvix is a Advertising Intelligence company We operate on all Medium:TV(Broadcast/Cable&OnDemand)Print(Newspapers&Magazines)RadioDigital(

5、Online&Mobile)PodcastOutdoorsCinema4CLASSIFICATION WITH GENAICLASSIFICATION WITH GENAIMotivation and ChallengesMotivation and Challenges2024 Databricks Inc.All rights reserved Our customers wantAccurate Branding(advertised product/service)Accurate Terms(offers)Near real time reportingRepresentation

6、of multi-lingual creativesNeed to minimize human classification/attribution5CLASSIFICATION WITH GENAICLASSIFICATION WITH GENAIMotivation and ChallengesMotivation and Challenges2024 Databricks Inc.All rights reserved6Driving Factors Behind ArchitectureDriving Factors Behind Architecture2024 Databrick