Downloaded 18 times
![[0,1,0,1]
1: Relavant Item
0: Non-Relavant Item
.
- Precision
/ 2/4 = 0.5
- Mean Reciprocal Ranks
mean(1/2 + 1/4) = 0.375
- normalized discounted cumulative gains
(NDCG@K)(1/log(1 + 2) + 1/log(1 + 4)) / (1/log(1 + 1) + 1/log(1 + 2)) = 0.65](https://image.slidesharecdn.com/amazonpersonalize-201017065918/75/AWS-Hero-Amazon-Personalize-AWS-Community-Day-Online-2020-24-2048.jpg)
![[AWS Hero 스페셜] Amazon Personalize를 통한 개인화/추천 서비스 개발 노하우 - 소성운(크로키닷컴) :: AWS Community Day Online 2020](https://image.slidesharecdn.com/amazonpersonalize-201017065918/75/AWS-Hero-Amazon-Personalize-AWS-Community-Day-Online-2020-31-2048.jpg)
Uploaded byAWSKRUG - AWS한국사용자모임
[AWS Hero 스페셜] Amazon Personalize를 통한 개인화/추천 서비스 개발 노하우 - 소성운(크로키닷컴) :: AWS Community Day Online 2020
Amazon Personalize is Amazon's machine learning service for generating personalized recommendations. It has over 3,700 customers and processes over 26TB of data daily using a machine learning stack of 33 DAGs and 200+ tasks in Airflow. Amazon Personalize offers rule-based, collaborative filtering, and deep learning models to generate recommendations and helps with cold start problems through feature engineering and unsupervised learning techniques. It provides an API endpoint and AutoML capabilities to build, train, tune and deploy machine learning models for recommendations.
More Related Content
![[AWS Innovate 온라인 컨퍼런스] Amazon Personalize를 통한 개인화 추천 기능 실전 구현하기 - 최원근, AWS 솔...](https://cdn.slidesharecdn.com/ss_thumbnails/awsinnovateonlineconferenceaimltrack2session4wonkeunchoi-200319071654-thumbnail.jpg?width=640&height=640&fit=bounds)
![[DEVIEW 2021] 1000만 글로벌 유저를 지탱하는 기술과 사람들](https://cdn.slidesharecdn.com/ss_thumbnails/1000-deview-220204010915-thumbnail.jpg?width=640&height=640&fit=bounds)
Similar to [AWS Hero 스페셜] Amazon Personalize를 통한 개인화/추천 서비스 개발 노하우 - 소성운(크로키닷컴) :: AWS Community Day Online 2020
![Machine-Learning-for-Personalization[1].pdf](https://cdn.slidesharecdn.com/ss_thumbnails/machine-learning-for-personalization1-240730065613-9d17b899-thumbnail.jpg?width=640&height=640&fit=bounds)
![IVS CTO Night And Day 2018 Winter - [re:Cap] AI & ML Services](https://cdn.slidesharecdn.com/ss_thumbnails/20181218recap-aimlctonightday2018w-190115184833-thumbnail.jpg?width=640&height=640&fit=bounds)
![[AWS Hero 스페셜] 서버리스 기반 검색 서비스 구축하기 - 이상현(스마일벤처스) :: AWS Community Day Online ...](https://cdn.slidesharecdn.com/ss_thumbnails/aa-201017071333-thumbnail.jpg?width=640&height=640&fit=bounds)
[AWS Hero 스페셜] Amazon Personalize를 통한 개인화/추천 서비스 개발 노하우 - 소성운(크로키닷컴) :: AWS Community Day Online 2020
- 1. Amazon Personalize (Yan So)/ Data Scientist,
- 2.
- 3. 3,700 + 2,000 + 300+ (MAU) 6,000 2019 26TB+ 250M+ 33 DAG, 200+ in Airflow
- 4.
- 5. • Rule-Based • • • • CollaborativeFiltering, Factorization Machine • • (Amazon SageMaker) • Deep Learning • •
- 6. • • Cold Start- , • • • • Machine Learning • ( , , , ) •
- 7.
- 8. Amazon Personalize Improve thepossibilities of personalizing your service! Take the full benefits ofAmazon Personalize
- 9. Benefits • • Amazon.com MachineLearning • API Endpoint • AutoML •
- 10.
- 12. ? 3700 + 10,000 + 300+ (MAU)
- 13. Amazon RDS Amazon DynamoDB Amazon S3 Client Amazon Kinesis Data Firehose Amazon S3 Amazon S3 AWS LambdaAmazon EMR AWS Glue Catalog Amazon Athena Amazon Personalize ML Feature Engineering AB Amazon SageMaker
- 14. Build Train TuneDeploy ● ● hrnn ● ● AutoML ● ● ● ● AWS SDK ● / ● Recommendation Filtering , , HPO ● Datalake ● Feature Engineering ● S3
- 15.
- 17. - User-Item Interaction(M),User(O), Item(O) 3 Dataset - Recipe User, Item dataset Metadata . - Data importing .
- 18.
- 20.
- 24. [0,1,0,1] 1: Relavant Item 0:Non-Relavant Item . - Precision / 2/4 = 0.5 - Mean Reciprocal Ranks mean(1/2 + 1/4) = 0.375 - normalized discounted cumulative gains (NDCG@K)(1/log(1 + 2) + 1/log(1 + 4)) / (1/log(1 + 1) + 1/log(1 + 2)) = 0.65
- 27. ( ) - CTR(Click-through Rate) - 20% - 10% Lesson Learned - - Feature Engineering
- 28. ( ) - CTR - LessonLearned - UI - =>
- 29. Amazon Personalize • (2019.11) • • •https://aws.amazon.com/ko/about-aws/whats-new/2019/11/amazon- personalize-now-supports-batch-recommendations/ • (2020.06) • ? • -> -> • , • https://aws.amazon.com/ko/blogs/machine-learning/introducing- recommendation-filters-in-amazon-personalize/
- 30. Amazon Personalize • UserPersonalization (2020.08) • • Context • https://aws.amazon.com/blogs/machine-learning/amazon-personalize- can-now-create-up-to-50-better-recommendations-for-fast-changing- catalogs-of-new-products-and-fresh-content/